{
  "text": "一个孩子出生6周后开始癫痫发作\n常规基因检测没有找到答案\n全基因组测序也没有锁定病因\n连血液里的RNA都给不出明确线索\n每个人体内有数百万处遗传差异\n真正致病的变化可能\n只藏在其中一个位置\n研究人员面对的难题很残酷\n候选项多到几乎无法逐一验证\nGoogle deep mind刚刚发布的alphagenome Atlas\n把这个搜索范围提前计算了一遍\n它覆盖人类基因组中大约\n九十亿种可能的单碱基变化\n也就是把约30亿个DNA位置\n分别换成另外3种字母\n再预测每一种变化可能怎样影响\n基因表达RNA剪接和染色质开放程度\n整套数据约1PB\n规模超过alphafold数据库的30倍\n过去需要调用模型逐个分析的任务\n如今可以先在一张\n预计算地图里查找\n这项发布很容易被概括\n成人类基因突变数据库\n但数据库里装的主要是AI预测\n并非九十亿次实验结果\n它也没有把每一种突变\n与一种疾病直接对应\nAlice提供的是分子影响线索\n帮助研究者从海量候选中\n判断哪些变化更值得进入实验\n它改变的是搜索顺序和实验效率\n还没有取代临床诊断\n人类基因组大约有30亿个DNA字母\n一个位置上的碱基如果发生替换\n就形成单核苷酸变异\n每个位置理论上可以\n换成另外3种碱基\n因此可能的单字母变化接近90亿种\n实验室不可能逐个制造\n培养和测量这些变异\n即使一次实验只花一美元\n成本也会达到90亿美元\n现实中的功能实验远比一美元昂贵\n还需要细胞类型\n培养条件和重复验证\n更棘手的地方在于\n人类过去比较容易理解\n的只是蛋白质编码区\n它只占基因组约2%\n这些区域像制造蛋白质的句子\n碱基改变后可能\n直接换掉一个氨基酸\n甚至让蛋白提前停止\n剩下约98%的非编码区域\n长期被误解成无用空间\n实际上其中大量序列\n控制基因在什么时间\n什么组织里打开\n又以多大强度表达\n许多与疾病和复杂性状相关的变异\n恰好落在这片更难解释的区域\n一处非编码变异不会简单的写\n着导致癫痫或增加糖尿病风险\n它可能破坏一个转录因子结合位点\n让某个基因在神经元里少表达一点\n也可能制造错误的RNA剪接位置\n使细胞拼出异常转录本\n影响还会随着细胞类型改变\n同一个位置在血液里看起来平静\n在脑组织里可能非常关键\n传统规则很难把这种长距离组织\n特异和多层调控同时放进判断\nalphagenome的作用就是从DNA序列\n直接预测这些分子层变化模型\n一次最多读取100万个碱基\n在单碱基分辨率上\n输出数千项调控特征\n输入参考序列以后\n它会预测基因起始位置与RNA产量\n判断剪接位置和染色质可及性\n也会分析DNA是否被特定蛋白结合\n把某个位置换成另一种碱基\n再对比突变前后的预测差异\n就能得到这次变化可能造成的影响\n长序列与高分辨率必须同时存在\n许多调控元件离目标基因很远\n只看附近几百个\n字母会漏掉作用关系\n如果为了扩大窗口而降低分辨率\n又会看不清单个碱基带来的差异\nalphagenome用卷积层寻找局部序列模式\n再让Transformer在整段序列中交换信息\n最后输出不同实验模态的预测\ndeep mind称\n单个变异的多项影响\n可以在约1秒内完成评分\n模型训练依赖Encode和\nGtex等公共科研项目\n也使用了4D nucleus与Phantom的数据\n这些项目已经在数百种人类与小鼠\n细胞和组织中测量了基因调控现象\nAlpha genome\n从这些数据里学习DNA序列与\n分子活动之间的统计关系\n再把规律推广到从未\n做过实验的变异上\nAtlas则进一步把模型在整个\n人类参考基因组上大规模运行\n提前计算全部可能的单碱基变化\n还加入超过1亿种在人群数据库\n中观察到的插入和缺失变异\n这里出现了一个重要转变\nalphagenome本来是一台分析机器\n研究者拿着候选变异去调用它\nAtlas把机器的结果\n预先铺成基础设施\n医生或研究者不需要掌握复杂代码\n也能在网页中定位一个基因区域\n查看不同碱基替换的预测\n需要批量分析的团队还可以使用API\n它与alphafold数据库的关系很相似\n模型负责计算\n数据库负责把计算结果变成整个\n科研社区都能检索的公共资源\n云计算还有一个容易被忽略的优势\n不同实验室过去各自选择\n输入窗口组织轨迹和评分方法\n同一个变异可能\n得到难以比较的结果\nAtlas用统一流程处理\n整个参考基因组\n研究者能够在相同尺度上横向排序\n计算资源较少的罕见病\n团队不必先租用大量TPU\n也能获得基础预测\n大型机构仍可调用\n模型进行定制分析\n基础门槛却明显降低\nAtlas覆盖的内容也超过九十亿种\n单碱基替换\n团队还计算了genomead\n英国生物样本库和美国\nAllofus等人群项目中观察到\n的超过1亿种插入与缺失\n此类变异会改变序列长度\n分析难度高于单字母替换\n论文通过位置偏移\n增强方法减少边界误差\n再把这些结果纳入统一评分\n数据库因此更接近真实\n患者会携带的变异类型\n九十亿种变化对应的输出仍然太多\n每一种变异横跨数百种组织\n还可能有数千项分子特征\n研究者如果看到一整面信号轨迹\n依旧很难迅速排序\nDeepmind因此发布了\nalphagenome variant impact评分简称AVI\n它把alphagenome与alphafold\n的优势合在一起\n为每个变异生成一个统一分数\n让研究者先找到\n可能影响最大的候选\n再继续查看具体机制\nalphafoldsense擅长判断蛋白质\n编码区里的错义变异\nalphagenome更擅长覆盖调控过程\nAVI还加入蛋白质提前\n终止起始位点丢失等注释\n以及跨物种保守性信息\n模型总共使用18类特征训练\n其中包括10类Alpha genome分子预测\n分数被转换成类似Pre的排序尺度\n10分代表进入全部预测\n的前12%10分代表前1%\n它提供的是优先级\n不是患病概率\nAVI的可解释性来自特征归因\n每个分数都能拆开显示\n主要信号究竟来自RNA\n剪接与基因表达\n还是染色质开放\n蛋白质改变或进化保守性\nAdler还整理出超过2,500类\n反复出现的DNA序列基序\n并标出它们的位置\n研究者不仅能看到\n某个变异分数很高\n还能继续追问它破坏\n了哪一种调控单词\n可能在哪类细胞里产生后果\n统一分数很方便\n也会压缩复杂性\n两个AVI分数相近的变异\n背后机制可能完全不同\n一个直接破坏蛋白质\n另一个只在某类\n神经元里轻微改变表达\n临床后果不能从排名本身推出\nAtlas保留特征归因和原始轨迹\n就是为了让研究者从哪个更值得看\n继续走到为什么值得看\n如果只导出一列分数\n系统最重要的解释能力就会被丢掉\nAVI还利用跨物种保守性\n某段DNA在漫长进化中很少改变\n通常说明它承担重要功能\n保守性可以提高致病变异排序\n却可能掩盖只在人类或特定\n灵长类中出现的新调控机制\n论文承认\n当保守性成为主要驱动特征时\n分子解释会变得模糊\n未来版本需要在预测准确率\n与机制透明度之间继续调整\n开头那个癫痫病例展示了\n这套系统真正有用的地方\n患者携带一个位于DNM\ne基因深层内涵子中的变异\nDNM1与癫痫性脑病关系密切\n可这个变化不在常见\n的蛋白质编码位置\n还涉及只在脑组织中\n使用的特殊转录本\n常规检测没有把它挑出来\n血液RNA检测也没有看到明显异常\n因为相关外显子在\n血液中几乎不表达\navi把这个变异排到了候选首位\n分数显示主要影响来自剪接\nalphagenome进一步预测\n它会在谷氨酸能神经元中\n制造一个错误的剪接受体位点\n让生成的蛋白多出一段异常序列\n研究团队随后对目标\n区域进行高通量实验\n把外显子上游265个碱基逐一改变\n并在5种细胞系中测试\n实验找到了12个会\n造成框内延长的变异\n其中包括已经报道的\n3个可能致病变异\nAlpha genome剪接分数与\n实验分类的AUPRC达到0.943\n这个结果具有说服力\n因为预测没有停在相关性\n系统先给出候选位置\n再指出脑细胞特异的剪接机制\n最后由实验验证\n他还解释了血液检测为何失败\n罕见病诊断常常卡在这一步\n患者身上有大量意义不明的变异\n临床团队缺少足够时间\n和组织样本逐个研究\n论文指出\n每名患者平均可能需要\n评估400万到500万个候选变异\n绝大多数没有危害或意义未知\n在已经解决的罕见病病例中\nAVI也显示出更好的排序能力\n研究团队把已知致病变异重新\n放回每名患者的背景变异里\n在前50个候选范围内\nAVI找回约29.5%的已知致病变异\n对比方法CADD1.7约为12.5%\n如果先按人群频率过滤\n前50名的召回率分别提高到74.3%和61%\n这些数字没有让诊断自动完成\n却能显著缩短专家\n需要人工阅读的清单\nAtlas的第二个价值出现\n在大规模人群研究\n常见性状往往受到\n大量微小效应影响\n罕见非编码变异的信号\n很容易淹没在统计噪声里\n埃克塞特大学研究人员把Adles\n用于英国生物样本库超过54,000\n名参与者的全基因组数据\n分析2,028种循环蛋白水平\n按照预测的分子作用\n筛选并组合罕见变异后\n研究发现的\n非编码遗传关联增加了22%\n其中一个例子涉及Plato G7\n这个基因与炎症和衰老研究有关\n普通区域筛选会把大量方向\n不同或没有功能的变异混在一起\n真正信号因此被稀释\n研究人员只保留AVI评分最高的1%\n候选被压缩到15个碱基\n范围内的4个变异统计\n关联反而更清晰\nalphagenome还预测这些变化会\n影响巨噬细胞中的基因转录\n使统计关联能够落到\n可检验的分子机制上\n研究团队又分析身高体重指数\n和糖化血红蛋白等复杂性状\n针对体重指数\n他们聚焦Atlas预测影响\n最大的1%的非编码变异\n找到了19个值得继续\n研究的遗传区域\n这个结果仍然属于研究线索\n不能拿来给个人计算精确体重命运\n它展示的是一种新的统计工具\n先用分子预测减少无关变异\n再把人群数据的检验能力\n集中到更可能有功能的位置\n传统全基因组关联研究擅长\n发现某片区域与疾病相关\n却常常很难解释区域里\n究竟哪个变异起作用\n相邻DNA会一起遗传统计信号\n可能覆盖几十甚至上百个候选\nAtlas为每个候选增加分子层说明\n帮助研究者设计下一轮实验\n药物研发也可能从中受益\n因为更清楚的致病机制\n可以帮助团队选择靶点\n识别适合入组的患者\n并减少在错误假设上投入的时间\n这种价值尤其适合罕见病\n单个疾病患者少\n传统统计方法缺乏样本量\n功能实验又昂贵\nAVI可以先从数百万变异\n中提出一小组优先级\n再由遗传学家结合家系\n信息症状和实验结果判断\n它不会替专家做出诊断\n却能把专家的注意力\n放到更值得看的地方\n医疗AI最现实的进步往往就在这里\n减少搜索成本\n而非直接宣布机器已经理解生命\n性能数据同样需要细看\n论文显示\nAVI在多个Cleanva分类\n任务上超过现有方法\n针对内涵子单碱基变异AUPRC达到0.76\n对比一个主要竞争方法的0.44\n同义变异为0.57 对比0.35\n三撇非翻译区为0.50 对比0.18\n在蛋白质改变变异上AVI为0.90\n对比0.86 他也没有赢下所有项目\n在5撇非翻译区和某项孟德尔\n疾病基准上略低于最佳方法\n这种不完全胜利反而更可信\n基因组预测不存在一把通吃的尺\n不同任务的数据量标签\n质量和生物机制差异很大\n一个模型在简介上表现突出\n未必能同样准确的预测远距离调控\n研究者需要根据疾病\n与变异类型选择工具\n还要保留多个方法交叉比较\n把多数基准领先改写\n成所有突变都能判断\n会直接越过证据\nAUPRC也不能被直接读成准确率\n它衡量模型在类别极度不平衡时\n精确率与召回率之间的综合表现\n致病变异远少于良性变异\n单看普通准确率\n一个把几乎所有候选都判成\n无害的系统也可能看起来很高\nAUPRC更适合这种任务\n但分数仍会随着测试集组成改变\n0.76代表在特定内\n涵子评测中排序更强\n不代表临床判断有76%的把握\n论文对10个独立的饱和基因组\n编辑实验进行了比较AVI在其中\n8项获得最高的斯皮尔曼相关性\n把全部实验合并后也取得最佳AUPRC\n这类实验会在一小段关键\n区域里系统制造大量变异\n在测量细胞功能\n能够为AI提供接近\n真实机制的压力测试\n范围仍然有限\n10个基因无法代表整个基因组\n尤其无法覆盖所有\n远距离调控和发育阶段\nAVI训练方式也带来限制\n团队把在人群中频率高于0.1%\n的变异当做相对中性代理\n把低于0.1%的罕见变异\n当做可能有影响的代理\n这个规则利用了\n自然选择留下的信号\n但罕见不等于有害\n常见也不等于完全无影响\n模型学到的是一种概率排序\n某个高分变异需要结合\n患者表型家族遗传和实验\n功能才能判断临床意义\noffgenome本身也受训练数据覆盖影响\n标准RNA测序容易漏掉没有\n多聚腺苷尾的非编码RNA\n不同细胞类型的数据分布不均\n一些疾病相关组织\n甚至缺少足够样本\n模型主要学习顺式调控\n也就是附近DNA序列怎样影响\n同一条染色体上的基因\n转录因子表达变化等\n反式机制没有被直接建模\n而这些机制解释着复杂\n性状遗传度的重要部分\n组织特异性是另一个难点\n人体有数百种细胞状态\n疾病还会让同一种细胞发生变化\nAtlas可以输出数百种\n组织与细胞类型的预测\n可它不可能覆盖所有发育\n阶段疾病环境和治疗状态\n开头病例能成功\n是因为系统恰好拥有与\n谷氨酸能神经元相关的预测轨迹\n面对缺失组织数据的疾病\n分数可能低估真正关键的变异\n论文还提醒\n序列基序地图依赖模型预测质量\n不能自动证明体内存在直接因果\n一个高分基序可能与\n真实调控位点重合\n也可能只是模型在训练\n数据中学到的近似模式\n研究团队对部分转录\n因子做了实验验证\n但2,500多类基序不可能全部完成\n验证Atlas适合生成更好的实验假设\n最终的生物学结论仍要\n回到细胞动物与患者数据\nAlpha genome Atlas不能用于个人医疗决策\ndeep mind在发布页面上明确写明\n平台没有通过临床用途验证\n也没有获得任何临床批准\n不能替代专业诊断或治疗建议\n学术研究可以免费使用网页\n基础模型和API也提供研究访问\n商业用途将通过Google cloud提供\n这个安排同时说明\nAtlas既是科学资源\n也是Google生命科学AI\n生态的一块基础设施\n它的商业价值可能\n晚于科研价值出现\n制药公司不会因为一个AVI\n分数立即启动临床试验\n但可以用它缩小实验范围\n诊断公司也不能把\n预测直接写进报告\n却可以把它与Cleanvar\n家系信息和功能实验组合\n谁能把基因组预测稳定接近药物\n发现患者分层和临床解释流程\n谁才会把一拍字节\n数据变成真正的生产力\ndeep mind还把Atlas接进\nGoogle Anti Gravity研究袋里\n未来的科研工作流可能\n由代理读取病历和文献\n再调用Atlas排列候选\n设计验证实验并追踪结果\n这里产生的变化很深\n模型不再只给研究者一个分数\n而是逐渐进入从假设到\n实验设计的连续流程\nalphagenome提供分子预测\nAVI压缩优先级\n代理系统负责组织步骤\n三者合在一起才接近Google\n真正想建的科学平台\n这条路线延续了alphafold的经验\nalphafold解决了蛋白质结构预测\n中的一个核心瓶颈数据库\n把数以亿计的结构开放给研究社区\nalphagenome面对的问题更复杂\n因为DNA调控受细胞类型和环境影响\n预测结果也更难用一个结构图\n表达Atlas仍然选择先大规模计算\n再降低查询门槛\nGoogle希望把过去只有少数计算\n生物学团队能运行的模型变成\n普通实验室也能使用的公共工具\n两者之间还有一个关键区别\n蛋白质结构通常对应\n相对稳定的物理对象\n基因变异影响却\n经常依赖组织和时间\nAtlas里的地图会随着训练\n数据与模型更新而改变\n它更接近一版持续\n修订的天气预测图\n而非不可更改的地理测绘\n科研人员必须记录使用的\n模型版本和数据库版本\n否则同一个变异以后\n可能得到不同解释\n数据规模本身也会带来治理问题\nAtlas基于人类参考基因组\nHG38进行全基因组计算\n而参考基因组无法完整\n代表所有人群的遗传多样性\n论文提出\n未来还要在不同遗传\n组员中校准和评估\n许多基因数据库长期\n偏向欧洲血统样本\n如果新模型继续沿用这种分布\n预测性能可能在代表\n不足的人群中下降\n更大的数据库不会自动\n消除旧数据里的偏差\n参考基因组还会漏掉结构差异\n单一线性序列很难完整\n表示不同人群携带的大\n片段插入倒位和复杂重复区域\n91种单碱基变化覆盖了\n一个清晰的问题空间\n却没有覆盖所有\n可能的人类遗传变化\n未来更完整的图基因组和\n泛基因组参考可能要求Atlas\n重新计算一部分地图\n发布规模很大\n边界仍然明确\n数据更新也会改变\n研究可重复性的要求\n团队在论文中提到\n英国生物样本库研究曾因分析\n平台临时停机使用略早版本的Atlast\n并检查了发布版本与\n旧版本之间的相关性\n模型和数据库持续更新以后\n论文需要保存版本号\n查询日期和具体评分轨迹\n否则几年后重新查同一变异\n研究者可能无法解释\n为何分数发生变化\n研究资源的开放范围也值得注意\n网页目前面向非商业研究免费开放\n商业服务将进入Google cloud\n公开访问降低了学术门槛\n但EPB预计算结果和持续\n更新能力仍掌握在Google手中\nAPI可以让实验室快速工作\n也可能把越来越多科研\n流程绑定到单一云平台\nalphagenome的科学贡献十分明确\n基础设施集中带来的\n依赖同样需要长期观察\n对普通人而言\n这项发布最重要的意义不在于\n马上得到一份基因命运报告\n真正变化发生在看不见的地方过去\n医生找到一个意义不明的变异后\n可能要等待多年文献积累\n现在研究者能先查看它对特定\n组织简接和表达的预测\n再决定是否投入实验\n一次诊断仍然需要严格证据\n但寻找证据的路线变得更短\n对科研机构而言\nAtlas把实验资源从广\n撒网推向精准验证\n九十亿种候选永远\n无法全部进入实验\nAI的价值是把最可能产生信息\n增量的那一小部分提到前面\n失败实验不会消失\n候选排序只会更合理\n随着真实实验不断回流\n模型还能发现自己在哪些细胞\n哪些调控模式上最容易出错\n下一代地图会继续改进\n它还可能改变科研竞争的节奏\n过去拥有大型计算团队的\n机构可以更快跑完变异评分\n小型实验室则要\n等待合作或购买服务\n统一Atlas把第一轮\n计算变成公共底座\n差距会转移到谁能提出更好的问题\n获得稀缺患者样本\n并设计高质量验证\n算力的重要性没有消失\n只是基础预测不再需要\n每家实验室重复支付一次\n药物公司的决策也可能提前\n一个疾病相关区域\n如果缺少明确作用机制\n团队通常难以判断该瞄准哪个基因\nlookalike可以显示某组变异主要\n影响哪个组织的表达或剪接\n帮助研究人员决定\n先做哪类细胞实验\n它不会直接给出药物分子\n更不会保证靶点有效\n却能减少候选机制之间的盲目试错\n真正节省的是研发最\n早期反复走错方向的时间\nAlpha GENOME Atlas也揭示了\n科学AI的一条成熟路径\n第一步训练模型\n预测可测量的生物现象\n重点不放在生成一段像论文的文字\n第二步把预测扩展到整个问题空间\n形成所有研究者都能查询的资源\n第三步让独立实验检验关键结果\n只有完成这三步\nAI才真正改变知识生产\n而非增加更多未经验证的答案\n开头那个病例给这条\n路径提供了一个清晰样本\n算法从数百万差异中挑出DNME\n生成内涵子变异\n预测脑细胞里的异常剪接\n又用实验确认该区域的功能\n它没有独立完成临床诊断\n也没有治愈患者\n却让一个长期藏在\n非编码区里的机制浮出水面\n对罕见病家庭而言\n知道病因可能帮助评估复发风险\n寻找同类病例\n并为未来治疗研究建立起点所以\nalphagenome Atlas的突破不在于AI\n已经读懂全部人类基因\n它第一次把每一种可能的\n单字母变化都放进同一套\n可搜索的分子预测地图\n并把蛋白质编码区与更\n庞大的非编码区连接起来\n九十亿是计算覆盖的规模\n不是已经证实的答案数量\n一拍字节是资源体量\n也不是临床可靠性的保证\n这张地图会把基因\n研究的瓶颈向后移动\n研究者以后更少困在\n应该先看哪一个变异\n更多困在怎样用实验确认这项预测\n这是更高级的难题\n也是科学真正前进的方式\nAI把搜索空间压缩\n人类负责判断证据\n两者各自守住边界\nalphagenome Atlas才可能成为\n基因组学里的下一座alphafold数据库",
  "language": "zh",
  "duration": 1582.0,
  "segments": [
    {
      "id": 0,
      "start": 0.133,
      "end": 2.9,
      "text": "一个孩子出生6周后开始癫痫发作",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 1,
      "start": 3.2,
      "end": 5.333,
      "text": "常规基因检测没有找到答案",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 2,
      "start": 5.533,
      "end": 8.133,
      "text": "全基因组测序也没有锁定病因",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 3,
      "start": 8.366,
      "end": 11.466,
      "text": "连血液里的RNA都给不出明确线索",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 4,
      "start": 11.733,
      "end": 14.266,
      "text": "每个人体内有数百万处遗传差异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 5,
      "start": 14.333,
      "end": 15.733,
      "text": "真正致病的变化可能",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 6,
      "start": 15.733,
      "end": 17.333,
      "text": "只藏在其中一个位置",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 7,
      "start": 17.566,
      "end": 19.866,
      "text": "研究人员面对的难题很残酷",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 8,
      "start": 20.366,
      "end": 23.0,
      "text": "候选项多到几乎无法逐一验证",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 9,
      "start": 23.266,
      "end": 26.933,
      "text": "Google deep mind刚刚发布的alphagenome Atlas",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 10,
      "start": 27.666,
      "end": 30.166,
      "text": "把这个搜索范围提前计算了一遍",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 11,
      "start": 30.3,
      "end": 32.6,
      "text": "它覆盖人类基因组中大约",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 12,
      "start": 32.6,
      "end": 35.333,
      "text": "九十亿种可能的单碱基变化",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 13,
      "start": 35.8,
      "end": 38.5,
      "text": "也就是把约30亿个DNA位置",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 14,
      "start": 38.5,
      "end": 40.7,
      "text": "分别换成另外3种字母",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 15,
      "start": 40.933,
      "end": 44.066,
      "text": "再预测每一种变化可能怎样影响",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 16,
      "start": 44.066,
      "end": 48.566,
      "text": "基因表达RNA剪接和染色质开放程度",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 17,
      "start": 48.733,
      "end": 50.366,
      "text": "整套数据约1PB",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 18,
      "start": 50.5,
      "end": 53.4,
      "text": "规模超过alphafold数据库的30倍",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 19,
      "start": 53.733,
      "end": 56.4,
      "text": "过去需要调用模型逐个分析的任务",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 20,
      "start": 56.6,
      "end": 58.066,
      "text": "如今可以先在一张",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 21,
      "start": 58.066,
      "end": 59.733,
      "text": "预计算地图里查找",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 22,
      "start": 59.766,
      "end": 61.366,
      "text": "这项发布很容易被概括",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 23,
      "start": 61.366,
      "end": 63.466,
      "text": "成人类基因突变数据库",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 24,
      "start": 63.866,
      "end": 66.733,
      "text": "但数据库里装的主要是AI预测",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 25,
      "start": 66.733,
      "end": 68.9,
      "text": "并非九十亿次实验结果",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 26,
      "start": 69.333,
      "end": 70.9,
      "text": "它也没有把每一种突变",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 27,
      "start": 70.9,
      "end": 72.666,
      "text": "与一种疾病直接对应",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 28,
      "start": 72.866,
      "end": 75.366,
      "text": "Alice提供的是分子影响线索",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 29,
      "start": 75.533,
      "end": 77.733,
      "text": "帮助研究者从海量候选中",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 30,
      "start": 77.733,
      "end": 80.4,
      "text": "判断哪些变化更值得进入实验",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 31,
      "start": 80.566,
      "end": 83.366,
      "text": "它改变的是搜索顺序和实验效率",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 32,
      "start": 83.666,
      "end": 85.5,
      "text": "还没有取代临床诊断",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 33,
      "start": 85.666,
      "end": 88.8,
      "text": "人类基因组大约有30亿个DNA字母",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 34,
      "start": 89.3,
      "end": 91.9,
      "text": "一个位置上的碱基如果发生替换",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 35,
      "start": 91.9,
      "end": 93.9,
      "text": "就形成单核苷酸变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 36,
      "start": 94.066,
      "end": 95.666,
      "text": "每个位置理论上可以",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 37,
      "start": 95.666,
      "end": 97.3,
      "text": "换成另外3种碱基",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 38,
      "start": 97.8,
      "end": 100.933,
      "text": "因此可能的单字母变化接近90亿种",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 39,
      "start": 101.1,
      "end": 102.966,
      "text": "实验室不可能逐个制造",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 40,
      "start": 102.966,
      "end": 104.866,
      "text": "培养和测量这些变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 41,
      "start": 105.0,
      "end": 107.166,
      "text": "即使一次实验只花一美元",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 42,
      "start": 107.3,
      "end": 109.3,
      "text": "成本也会达到90亿美元",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 43,
      "start": 109.566,
      "end": 112.533,
      "text": "现实中的功能实验远比一美元昂贵",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 44,
      "start": 112.566,
      "end": 113.933,
      "text": "还需要细胞类型",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 45,
      "start": 113.933,
      "end": 115.933,
      "text": "培养条件和重复验证",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 46,
      "start": 115.966,
      "end": 117.566,
      "text": "更棘手的地方在于",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 47,
      "start": 117.8,
      "end": 119.466,
      "text": "人类过去比较容易理解",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 48,
      "start": 119.466,
      "end": 121.466,
      "text": "的只是蛋白质编码区",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 49,
      "start": 121.6,
      "end": 124.0,
      "text": "它只占基因组约2%",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 50,
      "start": 124.333,
      "end": 127.2,
      "text": "这些区域像制造蛋白质的句子",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 51,
      "start": 127.6,
      "end": 129.066,
      "text": "碱基改变后可能",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 52,
      "start": 129.066,
      "end": 130.733,
      "text": "直接换掉一个氨基酸",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 53,
      "start": 131.0,
      "end": 133.166,
      "text": "甚至让蛋白提前停止",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 54,
      "start": 133.333,
      "end": 136.166,
      "text": "剩下约98%的非编码区域",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 55,
      "start": 136.166,
      "end": 138.3,
      "text": "长期被误解成无用空间",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 56,
      "start": 138.766,
      "end": 140.333,
      "text": "实际上其中大量序列",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 57,
      "start": 140.333,
      "end": 142.0,
      "text": "控制基因在什么时间",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 58,
      "start": 142.0,
      "end": 143.366,
      "text": "什么组织里打开",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 59,
      "start": 143.366,
      "end": 145.066,
      "text": "又以多大强度表达",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 60,
      "start": 145.333,
      "end": 148.166,
      "text": "许多与疾病和复杂性状相关的变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 61,
      "start": 148.666,
      "end": 150.933,
      "text": "恰好落在这片更难解释的区域",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 62,
      "start": 151.0,
      "end": 153.6,
      "text": "一处非编码变异不会简单的写",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 63,
      "start": 153.6,
      "end": 157.133,
      "text": "着导致癫痫或增加糖尿病风险",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 64,
      "start": 157.266,
      "end": 159.866,
      "text": "它可能破坏一个转录因子结合位点",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 65,
      "start": 160.066,
      "end": 162.9,
      "text": "让某个基因在神经元里少表达一点",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 66,
      "start": 163.266,
      "end": 165.966,
      "text": "也可能制造错误的RNA剪接位置",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 67,
      "start": 166.0,
      "end": 168.166,
      "text": "使细胞拼出异常转录本",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 68,
      "start": 168.266,
      "end": 170.566,
      "text": "影响还会随着细胞类型改变",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 69,
      "start": 170.766,
      "end": 173.266,
      "text": "同一个位置在血液里看起来平静",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 70,
      "start": 173.466,
      "end": 175.666,
      "text": "在脑组织里可能非常关键",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 71,
      "start": 176.333,
      "end": 179.166,
      "text": "传统规则很难把这种长距离组织",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 72,
      "start": 179.166,
      "end": 182.266,
      "text": "特异和多层调控同时放进判断",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 73,
      "start": 182.5,
      "end": 185.8,
      "text": "alphagenome的作用就是从DNA序列",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 74,
      "start": 185.8,
      "end": 188.4,
      "text": "直接预测这些分子层变化模型",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 75,
      "start": 188.8,
      "end": 191.0,
      "text": "一次最多读取100万个碱基",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 76,
      "start": 191.266,
      "end": 192.966,
      "text": "在单碱基分辨率上",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 77,
      "start": 192.966,
      "end": 195.0,
      "text": "输出数千项调控特征",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 78,
      "start": 195.333,
      "end": 197.166,
      "text": "输入参考序列以后",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 79,
      "start": 197.3,
      "end": 200.766,
      "text": "它会预测基因起始位置与RNA产量",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 80,
      "start": 200.766,
      "end": 203.8,
      "text": "判断剪接位置和染色质可及性",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 81,
      "start": 203.933,
      "end": 207.333,
      "text": "也会分析DNA是否被特定蛋白结合",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 82,
      "start": 207.7,
      "end": 209.933,
      "text": "把某个位置换成另一种碱基",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 83,
      "start": 210.166,
      "end": 212.533,
      "text": "再对比突变前后的预测差异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 84,
      "start": 212.7,
      "end": 215.533,
      "text": "就能得到这次变化可能造成的影响",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 85,
      "start": 215.766,
      "end": 218.6,
      "text": "长序列与高分辨率必须同时存在",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 86,
      "start": 218.8,
      "end": 221.533,
      "text": "许多调控元件离目标基因很远",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 87,
      "start": 221.666,
      "end": 223.0,
      "text": "只看附近几百个",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 88,
      "start": 223.0,
      "end": 224.8,
      "text": "字母会漏掉作用关系",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 89,
      "start": 225.466,
      "end": 228.166,
      "text": "如果为了扩大窗口而降低分辨率",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 90,
      "start": 228.333,
      "end": 230.966,
      "text": "又会看不清单个碱基带来的差异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 91,
      "start": 231.2,
      "end": 234.866,
      "text": "alphagenome用卷积层寻找局部序列模式",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 92,
      "start": 235.266,
      "end": 238.666,
      "text": "再让Transformer在整段序列中交换信息",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 93,
      "start": 238.9,
      "end": 241.266,
      "text": "最后输出不同实验模态的预测",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 94,
      "start": 241.5,
      "end": 242.4,
      "text": "deep mind称",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 95,
      "start": 242.7,
      "end": 244.366,
      "text": "单个变异的多项影响",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 96,
      "start": 244.366,
      "end": 246.4,
      "text": "可以在约1秒内完成评分",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 97,
      "start": 246.866,
      "end": 248.866,
      "text": "模型训练依赖Encode和",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 98,
      "start": 248.866,
      "end": 251.0,
      "text": "Gtex等公共科研项目",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 99,
      "start": 251.366,
      "end": 254.533,
      "text": "也使用了4D nucleus与Phantom的数据",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 100,
      "start": 254.733,
      "end": 257.733,
      "text": "这些项目已经在数百种人类与小鼠",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 101,
      "start": 257.733,
      "end": 261.166,
      "text": "细胞和组织中测量了基因调控现象",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 102,
      "start": 261.366,
      "end": 262.333,
      "text": "Alpha genome",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 103,
      "start": 262.733,
      "end": 265.4,
      "text": "从这些数据里学习DNA序列与",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 104,
      "start": 265.4,
      "end": 267.566,
      "text": "分子活动之间的统计关系",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 105,
      "start": 267.733,
      "end": 269.733,
      "text": "再把规律推广到从未",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 106,
      "start": 269.733,
      "end": 271.4,
      "text": "做过实验的变异上",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 107,
      "start": 271.766,
      "end": 274.133,
      "text": "Atlas则进一步把模型在整个",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 108,
      "start": 274.133,
      "end": 276.766,
      "text": "人类参考基因组上大规模运行",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 109,
      "start": 277.066,
      "end": 279.9,
      "text": "提前计算全部可能的单碱基变化",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 110,
      "start": 280.0,
      "end": 282.766,
      "text": "还加入超过1亿种在人群数据库",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 111,
      "start": 282.766,
      "end": 285.333,
      "text": "中观察到的插入和缺失变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 112,
      "start": 285.666,
      "end": 287.933,
      "text": "这里出现了一个重要转变",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 113,
      "start": 288.166,
      "end": 291.133,
      "text": "alphagenome本来是一台分析机器",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 114,
      "start": 291.333,
      "end": 294.1,
      "text": "研究者拿着候选变异去调用它",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 115,
      "start": 294.3,
      "end": 296.133,
      "text": "Atlas把机器的结果",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 116,
      "start": 296.133,
      "end": 297.9,
      "text": "预先铺成基础设施",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 117,
      "start": 298.3,
      "end": 300.9,
      "text": "医生或研究者不需要掌握复杂代码",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 118,
      "start": 301.066,
      "end": 303.766,
      "text": "也能在网页中定位一个基因区域",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 119,
      "start": 304.0,
      "end": 306.266,
      "text": "查看不同碱基替换的预测",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 120,
      "start": 306.766,
      "end": 309.733,
      "text": "需要批量分析的团队还可以使用API",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 121,
      "start": 309.9,
      "end": 312.733,
      "text": "它与alphafold数据库的关系很相似",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 122,
      "start": 313.066,
      "end": 314.4,
      "text": "模型负责计算",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 123,
      "start": 314.6,
      "end": 317.533,
      "text": "数据库负责把计算结果变成整个",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 124,
      "start": 317.533,
      "end": 320.3,
      "text": "科研社区都能检索的公共资源",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 125,
      "start": 320.533,
      "end": 323.2,
      "text": "云计算还有一个容易被忽略的优势",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 126,
      "start": 323.566,
      "end": 325.9,
      "text": "不同实验室过去各自选择",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 127,
      "start": 325.966,
      "end": 329.166,
      "text": "输入窗口组织轨迹和评分方法",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 128,
      "start": 329.2,
      "end": 330.333,
      "text": "同一个变异可能",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 129,
      "start": 330.333,
      "end": 332.1,
      "text": "得到难以比较的结果",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 130,
      "start": 332.366,
      "end": 334.533,
      "text": "Atlas用统一流程处理",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 131,
      "start": 334.533,
      "end": 335.966,
      "text": "整个参考基因组",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 132,
      "start": 336.4,
      "end": 339.666,
      "text": "研究者能够在相同尺度上横向排序",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 133,
      "start": 339.9,
      "end": 341.966,
      "text": "计算资源较少的罕见病",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 134,
      "start": 341.966,
      "end": 344.7,
      "text": "团队不必先租用大量TPU",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 135,
      "start": 344.766,
      "end": 346.466,
      "text": "也能获得基础预测",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 136,
      "start": 346.566,
      "end": 348.3,
      "text": "大型机构仍可调用",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 137,
      "start": 348.3,
      "end": 349.966,
      "text": "模型进行定制分析",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 138,
      "start": 350.266,
      "end": 352.366,
      "text": "基础门槛却明显降低",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 139,
      "start": 352.533,
      "end": 355.333,
      "text": "Atlas覆盖的内容也超过九十亿种",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 140,
      "start": 355.566,
      "end": 356.866,
      "text": "单碱基替换",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 141,
      "start": 357.333,
      "end": 359.566,
      "text": "团队还计算了genomead",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 142,
      "start": 359.6,
      "end": 361.4,
      "text": "英国生物样本库和美国",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 143,
      "start": 361.4,
      "end": 363.866,
      "text": "Allofus等人群项目中观察到",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 144,
      "start": 363.866,
      "end": 366.3,
      "text": "的超过1亿种插入与缺失",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 145,
      "start": 366.7,
      "end": 368.9,
      "text": "此类变异会改变序列长度",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 146,
      "start": 369.0,
      "end": 371.166,
      "text": "分析难度高于单字母替换",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 147,
      "start": 371.4,
      "end": 373.0,
      "text": "论文通过位置偏移",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 148,
      "start": 373.1,
      "end": 375.333,
      "text": "增强方法减少边界误差",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 149,
      "start": 375.366,
      "end": 377.866,
      "text": "再把这些结果纳入统一评分",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 150,
      "start": 378.533,
      "end": 380.466,
      "text": "数据库因此更接近真实",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 151,
      "start": 380.466,
      "end": 382.4,
      "text": "患者会携带的变异类型",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 152,
      "start": 382.766,
      "end": 385.933,
      "text": "九十亿种变化对应的输出仍然太多",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 153,
      "start": 386.266,
      "end": 388.7,
      "text": "每一种变异横跨数百种组织",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 154,
      "start": 388.866,
      "end": 391.1,
      "text": "还可能有数千项分子特征",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 155,
      "start": 391.4,
      "end": 394.1,
      "text": "研究者如果看到一整面信号轨迹",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 156,
      "start": 394.366,
      "end": 396.0,
      "text": "依旧很难迅速排序",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 157,
      "start": 396.5,
      "end": 397.8,
      "text": "Deepmind因此发布了",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 158,
      "start": 397.8,
      "end": 401.3,
      "text": "alphagenome variant impact评分简称AVI",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 159,
      "start": 401.866,
      "end": 403.766,
      "text": "它把alphagenome与alphafold",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 160,
      "start": 403.966,
      "end": 405.4,
      "text": "的优势合在一起",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 161,
      "start": 405.466,
      "end": 407.966,
      "text": "为每个变异生成一个统一分数",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 162,
      "start": 408.066,
      "end": 409.566,
      "text": "让研究者先找到",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 163,
      "start": 409.566,
      "end": 411.4,
      "text": "可能影响最大的候选",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 164,
      "start": 411.466,
      "end": 413.333,
      "text": "再继续查看具体机制",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 165,
      "start": 413.566,
      "end": 416.366,
      "text": "alphafoldsense擅长判断蛋白质",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 166,
      "start": 416.366,
      "end": 418.133,
      "text": "编码区里的错义变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 167,
      "start": 418.5,
      "end": 421.466,
      "text": "alphagenome更擅长覆盖调控过程",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 168,
      "start": 421.8,
      "end": 424.1,
      "text": "AVI还加入蛋白质提前",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 169,
      "start": 424.1,
      "end": 426.966,
      "text": "终止起始位点丢失等注释",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 170,
      "start": 427.133,
      "end": 429.466,
      "text": "以及跨物种保守性信息",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 171,
      "start": 429.7,
      "end": 432.366,
      "text": "模型总共使用18类特征训练",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 172,
      "start": 432.7,
      "end": 435.866,
      "text": "其中包括10类Alpha genome分子预测",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 173,
      "start": 436.533,
      "end": 439.666,
      "text": "分数被转换成类似Pre的排序尺度",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 174,
      "start": 439.8,
      "end": 441.733,
      "text": "10分代表进入全部预测",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 175,
      "start": 441.733,
      "end": 445.4,
      "text": "的前12%10分代表前1%",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 176,
      "start": 445.766,
      "end": 447.9,
      "text": "它提供的是优先级",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 177,
      "start": 447.9,
      "end": 449.166,
      "text": "不是患病概率",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 178,
      "start": 449.3,
      "end": 452.133,
      "text": "AVI的可解释性来自特征归因",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 179,
      "start": 452.366,
      "end": 454.333,
      "text": "每个分数都能拆开显示",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 180,
      "start": 454.766,
      "end": 456.7,
      "text": "主要信号究竟来自RNA",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 181,
      "start": 456.766,
      "end": 458.166,
      "text": "剪接与基因表达",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 182,
      "start": 458.166,
      "end": 459.733,
      "text": "还是染色质开放",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 183,
      "start": 459.733,
      "end": 462.1,
      "text": "蛋白质改变或进化保守性",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 184,
      "start": 462.366,
      "end": 464.966,
      "text": "Adler还整理出超过2,500类",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 185,
      "start": 464.966,
      "end": 467.3,
      "text": "反复出现的DNA序列基序",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 186,
      "start": 467.466,
      "end": 469.1,
      "text": "并标出它们的位置",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 187,
      "start": 469.3,
      "end": 470.533,
      "text": "研究者不仅能看到",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 188,
      "start": 470.533,
      "end": 472.166,
      "text": "某个变异分数很高",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 189,
      "start": 472.266,
      "end": 473.933,
      "text": "还能继续追问它破坏",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 190,
      "start": 473.933,
      "end": 475.566,
      "text": "了哪一种调控单词",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 191,
      "start": 475.7,
      "end": 478.1,
      "text": "可能在哪类细胞里产生后果",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 192,
      "start": 478.2,
      "end": 479.5,
      "text": "统一分数很方便",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 193,
      "start": 479.5,
      "end": 481.1,
      "text": "也会压缩复杂性",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 194,
      "start": 481.6,
      "end": 483.533,
      "text": "两个AVI分数相近的变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 195,
      "start": 483.533,
      "end": 485.566,
      "text": "背后机制可能完全不同",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 196,
      "start": 485.766,
      "end": 487.533,
      "text": "一个直接破坏蛋白质",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 197,
      "start": 487.866,
      "end": 489.1,
      "text": "另一个只在某类",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 198,
      "start": 489.1,
      "end": 491.166,
      "text": "神经元里轻微改变表达",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 199,
      "start": 491.566,
      "end": 494.166,
      "text": "临床后果不能从排名本身推出",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 200,
      "start": 494.333,
      "end": 497.1,
      "text": "Atlas保留特征归因和原始轨迹",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 201,
      "start": 497.4,
      "end": 500.266,
      "text": "就是为了让研究者从哪个更值得看",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 202,
      "start": 500.266,
      "end": 502.266,
      "text": "继续走到为什么值得看",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 203,
      "start": 502.866,
      "end": 504.566,
      "text": "如果只导出一列分数",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 204,
      "start": 504.666,
      "end": 507.4,
      "text": "系统最重要的解释能力就会被丢掉",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 205,
      "start": 507.566,
      "end": 509.933,
      "text": "AVI还利用跨物种保守性",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 206,
      "start": 510.0,
      "end": 513.1,
      "text": "某段DNA在漫长进化中很少改变",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 207,
      "start": 513.2,
      "end": 515.533,
      "text": "通常说明它承担重要功能",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 208,
      "start": 515.7,
      "end": 518.266,
      "text": "保守性可以提高致病变异排序",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 209,
      "start": 518.733,
      "end": 520.933,
      "text": "却可能掩盖只在人类或特定",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 210,
      "start": 520.933,
      "end": 523.466,
      "text": "灵长类中出现的新调控机制",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 211,
      "start": 523.8,
      "end": 524.6,
      "text": "论文承认",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 212,
      "start": 524.6,
      "end": 527.4,
      "text": "当保守性成为主要驱动特征时",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 213,
      "start": 527.4,
      "end": 529.266,
      "text": "分子解释会变得模糊",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 214,
      "start": 529.766,
      "end": 531.866,
      "text": "未来版本需要在预测准确率",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 215,
      "start": 531.866,
      "end": 534.533,
      "text": "与机制透明度之间继续调整",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 216,
      "start": 534.6,
      "end": 536.8,
      "text": "开头那个癫痫病例展示了",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 217,
      "start": 536.8,
      "end": 538.966,
      "text": "这套系统真正有用的地方",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 218,
      "start": 539.133,
      "end": 541.3,
      "text": "患者携带一个位于DNM",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 219,
      "start": 541.3,
      "end": 543.9,
      "text": "e基因深层内涵子中的变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 220,
      "start": 544.3,
      "end": 547.4,
      "text": "DNM1与癫痫性脑病关系密切",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 221,
      "start": 547.666,
      "end": 549.266,
      "text": "可这个变化不在常见",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 222,
      "start": 549.266,
      "end": 550.933,
      "text": "的蛋白质编码位置",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 223,
      "start": 551.133,
      "end": 552.933,
      "text": "还涉及只在脑组织中",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 224,
      "start": 552.933,
      "end": 554.533,
      "text": "使用的特殊转录本",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 225,
      "start": 554.766,
      "end": 557.066,
      "text": "常规检测没有把它挑出来",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 226,
      "start": 557.266,
      "end": 560.466,
      "text": "血液RNA检测也没有看到明显异常",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 227,
      "start": 560.9,
      "end": 562.2,
      "text": "因为相关外显子在",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 228,
      "start": 562.2,
      "end": 564.066,
      "text": "血液中几乎不表达",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 229,
      "start": 564.266,
      "end": 567.3,
      "text": "avi把这个变异排到了候选首位",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 230,
      "start": 568.1,
      "end": 570.366,
      "text": "分数显示主要影响来自剪接",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 231,
      "start": 570.6,
      "end": 572.466,
      "text": "alphagenome进一步预测",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 232,
      "start": 572.6,
      "end": 574.866,
      "text": "它会在谷氨酸能神经元中",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 233,
      "start": 574.866,
      "end": 577.666,
      "text": "制造一个错误的剪接受体位点",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 234,
      "start": 577.7,
      "end": 580.8,
      "text": "让生成的蛋白多出一段异常序列",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 235,
      "start": 581.0,
      "end": 582.7,
      "text": "研究团队随后对目标",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 236,
      "start": 582.7,
      "end": 584.533,
      "text": "区域进行高通量实验",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 237,
      "start": 584.8,
      "end": 588.466,
      "text": "把外显子上游265个碱基逐一改变",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 238,
      "start": 588.566,
      "end": 590.6,
      "text": "并在5种细胞系中测试",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 239,
      "start": 590.933,
      "end": 592.666,
      "text": "实验找到了12个会",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 240,
      "start": 592.666,
      "end": 594.7,
      "text": "造成框内延长的变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 241,
      "start": 594.966,
      "end": 596.6,
      "text": "其中包括已经报道的",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 242,
      "start": 596.6,
      "end": 598.333,
      "text": "3个可能致病变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 243,
      "start": 598.6,
      "end": 600.5,
      "text": "Alpha genome剪接分数与",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 244,
      "start": 600.5,
      "end": 603.9,
      "text": "实验分类的AUPRC达到0.943",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 245,
      "start": 604.166,
      "end": 605.9,
      "text": "这个结果具有说服力",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 246,
      "start": 605.933,
      "end": 608.2,
      "text": "因为预测没有停在相关性",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 247,
      "start": 608.866,
      "end": 610.8,
      "text": "系统先给出候选位置",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 248,
      "start": 611.266,
      "end": 613.7,
      "text": "再指出脑细胞特异的剪接机制",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 249,
      "start": 613.866,
      "end": 615.4,
      "text": "最后由实验验证",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 250,
      "start": 615.8,
      "end": 618.366,
      "text": "他还解释了血液检测为何失败",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 251,
      "start": 618.6,
      "end": 620.866,
      "text": "罕见病诊断常常卡在这一步",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 252,
      "start": 621.066,
      "end": 623.566,
      "text": "患者身上有大量意义不明的变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 253,
      "start": 623.8,
      "end": 625.6,
      "text": "临床团队缺少足够时间",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 254,
      "start": 625.6,
      "end": 627.566,
      "text": "和组织样本逐个研究",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 255,
      "start": 628.0,
      "end": 628.866,
      "text": "论文指出",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 256,
      "start": 628.866,
      "end": 630.7,
      "text": "每名患者平均可能需要",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 257,
      "start": 630.7,
      "end": 633.566,
      "text": "评估400万到500万个候选变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 258,
      "start": 633.766,
      "end": 636.4,
      "text": "绝大多数没有危害或意义未知",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 259,
      "start": 636.466,
      "end": 638.9,
      "text": "在已经解决的罕见病病例中",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 260,
      "start": 639.1,
      "end": 641.966,
      "text": "AVI也显示出更好的排序能力",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 261,
      "start": 642.166,
      "end": 644.666,
      "text": "研究团队把已知致病变异重新",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 262,
      "start": 644.666,
      "end": 647.133,
      "text": "放回每名患者的背景变异里",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 263,
      "start": 647.466,
      "end": 649.5,
      "text": "在前50个候选范围内",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 264,
      "start": 649.566,
      "end": 654.4,
      "text": "AVI找回约29.5%的已知致病变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 265,
      "start": 654.533,
      "end": 659.3,
      "text": "对比方法CADD1.7约为12.5%",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 266,
      "start": 659.533,
      "end": 661.533,
      "text": "如果先按人群频率过滤",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 267,
      "start": 661.866,
      "end": 667.733,
      "text": "前50名的召回率分别提高到74.3%和61%",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 268,
      "start": 668.0,
      "end": 670.866,
      "text": "这些数字没有让诊断自动完成",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 269,
      "start": 671.166,
      "end": 673.333,
      "text": "却能显著缩短专家",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 270,
      "start": 673.333,
      "end": 675.3,
      "text": "需要人工阅读的清单",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 271,
      "start": 675.533,
      "end": 677.166,
      "text": "Atlas的第二个价值出现",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 272,
      "start": 677.166,
      "end": 678.766,
      "text": "在大规模人群研究",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 273,
      "start": 679.1,
      "end": 680.7,
      "text": "常见性状往往受到",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 274,
      "start": 680.7,
      "end": 682.333,
      "text": "大量微小效应影响",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 275,
      "start": 682.666,
      "end": 684.566,
      "text": "罕见非编码变异的信号",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 276,
      "start": 684.566,
      "end": 686.8,
      "text": "很容易淹没在统计噪声里",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 277,
      "start": 687.166,
      "end": 689.666,
      "text": "埃克塞特大学研究人员把Adles",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 278,
      "start": 689.733,
      "end": 692.8,
      "text": "用于英国生物样本库超过54,000",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 279,
      "start": 692.8,
      "end": 694.966,
      "text": "名参与者的全基因组数据",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 280,
      "start": 695.533,
      "end": 698.3,
      "text": "分析2,028种循环蛋白水平",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 281,
      "start": 698.466,
      "end": 700.466,
      "text": "按照预测的分子作用",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 282,
      "start": 700.666,
      "end": 703.266,
      "text": "筛选并组合罕见变异后",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 283,
      "start": 703.333,
      "end": 704.266,
      "text": "研究发现的",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 284,
      "start": 704.266,
      "end": 707.5,
      "text": "非编码遗传关联增加了22%",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 285,
      "start": 707.6,
      "end": 710.366,
      "text": "其中一个例子涉及Plato G7",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 286,
      "start": 710.4,
      "end": 713.333,
      "text": "这个基因与炎症和衰老研究有关",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 287,
      "start": 713.8,
      "end": 715.933,
      "text": "普通区域筛选会把大量方向",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 288,
      "start": 715.933,
      "end": 718.733,
      "text": "不同或没有功能的变异混在一起",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 289,
      "start": 718.9,
      "end": 720.866,
      "text": "真正信号因此被稀释",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 290,
      "start": 721.133,
      "end": 724.566,
      "text": "研究人员只保留AVI评分最高的1%",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 291,
      "start": 724.666,
      "end": 726.7,
      "text": "候选被压缩到15个碱基",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 292,
      "start": 726.7,
      "end": 728.5,
      "text": "范围内的4个变异统计",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 293,
      "start": 728.9,
      "end": 730.3,
      "text": "关联反而更清晰",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 294,
      "start": 730.6,
      "end": 733.0,
      "text": "alphagenome还预测这些变化会",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 295,
      "start": 733.0,
      "end": 735.366,
      "text": "影响巨噬细胞中的基因转录",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 296,
      "start": 735.566,
      "end": 737.4,
      "text": "使统计关联能够落到",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 297,
      "start": 737.4,
      "end": 739.533,
      "text": "可检验的分子机制上",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 298,
      "start": 740.1,
      "end": 743.066,
      "text": "研究团队又分析身高体重指数",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 299,
      "start": 743.066,
      "end": 745.9,
      "text": "和糖化血红蛋白等复杂性状",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 300,
      "start": 746.1,
      "end": 747.4,
      "text": "针对体重指数",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 301,
      "start": 747.466,
      "end": 749.8,
      "text": "他们聚焦Atlas预测影响",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 302,
      "start": 749.8,
      "end": 752.333,
      "text": "最大的1%的非编码变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 303,
      "start": 752.5,
      "end": 754.3,
      "text": "找到了19个值得继续",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 304,
      "start": 754.3,
      "end": 755.733,
      "text": "研究的遗传区域",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 305,
      "start": 755.966,
      "end": 758.1,
      "text": "这个结果仍然属于研究线索",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 306,
      "start": 758.266,
      "end": 761.3,
      "text": "不能拿来给个人计算精确体重命运",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 307,
      "start": 761.5,
      "end": 763.866,
      "text": "它展示的是一种新的统计工具",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 308,
      "start": 764.333,
      "end": 766.9,
      "text": "先用分子预测减少无关变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 309,
      "start": 767.066,
      "end": 769.2,
      "text": "再把人群数据的检验能力",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 310,
      "start": 769.2,
      "end": 771.7,
      "text": "集中到更可能有功能的位置",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 311,
      "start": 771.9,
      "end": 774.466,
      "text": "传统全基因组关联研究擅长",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 312,
      "start": 774.466,
      "end": 777.1,
      "text": "发现某片区域与疾病相关",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 313,
      "start": 777.3,
      "end": 779.7,
      "text": "却常常很难解释区域里",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 314,
      "start": 779.7,
      "end": 781.466,
      "text": "究竟哪个变异起作用",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 315,
      "start": 781.766,
      "end": 785.166,
      "text": "相邻DNA会一起遗传统计信号",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 316,
      "start": 785.166,
      "end": 788.133,
      "text": "可能覆盖几十甚至上百个候选",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 317,
      "start": 788.333,
      "end": 791.566,
      "text": "Atlas为每个候选增加分子层说明",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 318,
      "start": 791.566,
      "end": 793.9,
      "text": "帮助研究者设计下一轮实验",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 319,
      "start": 794.366,
      "end": 796.333,
      "text": "药物研发也可能从中受益",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 320,
      "start": 796.466,
      "end": 798.3,
      "text": "因为更清楚的致病机制",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 321,
      "start": 798.3,
      "end": 800.666,
      "text": "可以帮助团队选择靶点",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 322,
      "start": 800.733,
      "end": 802.733,
      "text": "识别适合入组的患者",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 323,
      "start": 802.766,
      "end": 805.733,
      "text": "并减少在错误假设上投入的时间",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 324,
      "start": 805.866,
      "end": 808.6,
      "text": "这种价值尤其适合罕见病",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 325,
      "start": 809.4,
      "end": 810.966,
      "text": "单个疾病患者少",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 326,
      "start": 811.266,
      "end": 813.766,
      "text": "传统统计方法缺乏样本量",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 327,
      "start": 813.9,
      "end": 815.533,
      "text": "功能实验又昂贵",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 328,
      "start": 815.8,
      "end": 818.066,
      "text": "AVI可以先从数百万变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 329,
      "start": 818.066,
      "end": 819.966,
      "text": "中提出一小组优先级",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 330,
      "start": 820.166,
      "end": 822.2,
      "text": "再由遗传学家结合家系",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 331,
      "start": 822.2,
      "end": 824.566,
      "text": "信息症状和实验结果判断",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 332,
      "start": 824.966,
      "end": 826.9,
      "text": "它不会替专家做出诊断",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 333,
      "start": 827.066,
      "end": 828.666,
      "text": "却能把专家的注意力",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 334,
      "start": 828.666,
      "end": 830.366,
      "text": "放到更值得看的地方",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 335,
      "start": 830.6,
      "end": 833.733,
      "text": "医疗AI最现实的进步往往就在这里",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 336,
      "start": 833.933,
      "end": 835.333,
      "text": "减少搜索成本",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 337,
      "start": 835.333,
      "end": 838.166,
      "text": "而非直接宣布机器已经理解生命",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 338,
      "start": 838.5,
      "end": 840.566,
      "text": "性能数据同样需要细看",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 339,
      "start": 840.8,
      "end": 841.8,
      "text": "论文显示",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 340,
      "start": 842.0,
      "end": 844.166,
      "text": "AVI在多个Cleanva分类",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 341,
      "start": 844.166,
      "end": 846.1,
      "text": "任务上超过现有方法",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 342,
      "start": 846.366,
      "end": 851.066,
      "text": "针对内涵子单碱基变异AUPRC达到0.76",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 343,
      "start": 851.133,
      "end": 853.866,
      "text": "对比一个主要竞争方法的0.44",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 344,
      "start": 854.1,
      "end": 857.533,
      "text": "同义变异为0.57 对比0.35",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 345,
      "start": 857.533,
      "end": 861.566,
      "text": "三撇非翻译区为0.50 对比0.18",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 346,
      "start": 861.566,
      "end": 865.5,
      "text": "在蛋白质改变变异上AVI为0.90",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 347,
      "start": 865.766,
      "end": 869.0,
      "text": "对比0.86 他也没有赢下所有项目",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 348,
      "start": 869.133,
      "end": 871.7,
      "text": "在5撇非翻译区和某项孟德尔",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 349,
      "start": 871.7,
      "end": 874.133,
      "text": "疾病基准上略低于最佳方法",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 350,
      "start": 874.2,
      "end": 876.666,
      "text": "这种不完全胜利反而更可信",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 351,
      "start": 877.166,
      "end": 879.866,
      "text": "基因组预测不存在一把通吃的尺",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 352,
      "start": 880.6,
      "end": 882.666,
      "text": "不同任务的数据量标签",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 353,
      "start": 882.666,
      "end": 885.166,
      "text": "质量和生物机制差异很大",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 354,
      "start": 885.2,
      "end": 887.766,
      "text": "一个模型在简介上表现突出",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 355,
      "start": 887.8,
      "end": 890.9,
      "text": "未必能同样准确的预测远距离调控",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 356,
      "start": 891.0,
      "end": 892.5,
      "text": "研究者需要根据疾病",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 357,
      "start": 892.5,
      "end": 894.4,
      "text": "与变异类型选择工具",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 358,
      "start": 894.533,
      "end": 896.9,
      "text": "还要保留多个方法交叉比较",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 359,
      "start": 897.2,
      "end": 899.0,
      "text": "把多数基准领先改写",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 360,
      "start": 899.0,
      "end": 900.966,
      "text": "成所有突变都能判断",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 361,
      "start": 901.1,
      "end": 902.566,
      "text": "会直接越过证据",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 362,
      "start": 902.733,
      "end": 906.1,
      "text": "AUPRC也不能被直接读成准确率",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 363,
      "start": 906.5,
      "end": 909.6,
      "text": "它衡量模型在类别极度不平衡时",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 364,
      "start": 909.766,
      "end": 912.7,
      "text": "精确率与召回率之间的综合表现",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 365,
      "start": 913.0,
      "end": 915.566,
      "text": "致病变异远少于良性变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 366,
      "start": 916.2,
      "end": 917.766,
      "text": "单看普通准确率",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 367,
      "start": 917.866,
      "end": 919.9,
      "text": "一个把几乎所有候选都判成",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 368,
      "start": 919.9,
      "end": 922.2,
      "text": "无害的系统也可能看起来很高",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 369,
      "start": 922.3,
      "end": 924.566,
      "text": "AUPRC更适合这种任务",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 370,
      "start": 924.733,
      "end": 927.666,
      "text": "但分数仍会随着测试集组成改变",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 371,
      "start": 927.8,
      "end": 929.6,
      "text": "0.76代表在特定内",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 372,
      "start": 929.6,
      "end": 931.6,
      "text": "涵子评测中排序更强",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 373,
      "start": 931.766,
      "end": 934.866,
      "text": "不代表临床判断有76%的把握",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 374,
      "start": 935.3,
      "end": 937.4,
      "text": "论文对10个独立的饱和基因组",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 375,
      "start": 937.4,
      "end": 940.3,
      "text": "编辑实验进行了比较AVI在其中",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 376,
      "start": 940.3,
      "end": 943.066,
      "text": "8项获得最高的斯皮尔曼相关性",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 377,
      "start": 943.066,
      "end": 946.9,
      "text": "把全部实验合并后也取得最佳AUPRC",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 378,
      "start": 946.966,
      "end": 948.666,
      "text": "这类实验会在一小段关键",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 379,
      "start": 948.666,
      "end": 950.733,
      "text": "区域里系统制造大量变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 380,
      "start": 951.2,
      "end": 952.8,
      "text": "在测量细胞功能",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 381,
      "start": 952.9,
      "end": 954.533,
      "text": "能够为AI提供接近",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 382,
      "start": 954.533,
      "end": 956.266,
      "text": "真实机制的压力测试",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 383,
      "start": 956.7,
      "end": 957.9,
      "text": "范围仍然有限",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 384,
      "start": 958.2,
      "end": 960.6,
      "text": "10个基因无法代表整个基因组",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 385,
      "start": 960.766,
      "end": 962.133,
      "text": "尤其无法覆盖所有",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 386,
      "start": 962.133,
      "end": 964.133,
      "text": "远距离调控和发育阶段",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 387,
      "start": 964.333,
      "end": 966.6,
      "text": "AVI训练方式也带来限制",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 388,
      "start": 966.733,
      "end": 969.666,
      "text": "团队把在人群中频率高于0.1%",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 389,
      "start": 969.666,
      "end": 971.9,
      "text": "的变异当做相对中性代理",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 390,
      "start": 972.066,
      "end": 974.8,
      "text": "把低于0.1%的罕见变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 391,
      "start": 974.966,
      "end": 976.866,
      "text": "当做可能有影响的代理",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 392,
      "start": 977.166,
      "end": 978.3,
      "text": "这个规则利用了",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 393,
      "start": 978.3,
      "end": 980.1,
      "text": "自然选择留下的信号",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 394,
      "start": 980.133,
      "end": 981.933,
      "text": "但罕见不等于有害",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 395,
      "start": 982.2,
      "end": 984.466,
      "text": "常见也不等于完全无影响",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 396,
      "start": 985.1,
      "end": 987.1,
      "text": "模型学到的是一种概率排序",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 397,
      "start": 987.3,
      "end": 989.066,
      "text": "某个高分变异需要结合",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 398,
      "start": 989.066,
      "end": 991.5,
      "text": "患者表型家族遗传和实验",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 399,
      "start": 991.5,
      "end": 993.366,
      "text": "功能才能判断临床意义",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 400,
      "start": 993.7,
      "end": 996.733,
      "text": "offgenome本身也受训练数据覆盖影响",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 401,
      "start": 996.9,
      "end": 998.966,
      "text": "标准RNA测序容易漏掉没有",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 402,
      "start": 998.966,
      "end": 1001.533,
      "text": "多聚腺苷尾的非编码RNA",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 403,
      "start": 1001.566,
      "end": 1003.866,
      "text": "不同细胞类型的数据分布不均",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 404,
      "start": 1004.066,
      "end": 1005.566,
      "text": "一些疾病相关组织",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 405,
      "start": 1005.566,
      "end": 1007.266,
      "text": "甚至缺少足够样本",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 406,
      "start": 1007.566,
      "end": 1009.8,
      "text": "模型主要学习顺式调控",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 407,
      "start": 1009.866,
      "end": 1012.9,
      "text": "也就是附近DNA序列怎样影响",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 408,
      "start": 1012.9,
      "end": 1014.966,
      "text": "同一条染色体上的基因",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 409,
      "start": 1015.166,
      "end": 1016.933,
      "text": "转录因子表达变化等",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 410,
      "start": 1016.933,
      "end": 1019.1,
      "text": "反式机制没有被直接建模",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 411,
      "start": 1019.4,
      "end": 1021.2,
      "text": "而这些机制解释着复杂",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 412,
      "start": 1021.2,
      "end": 1023.266,
      "text": "性状遗传度的重要部分",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 413,
      "start": 1023.666,
      "end": 1025.666,
      "text": "组织特异性是另一个难点",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 414,
      "start": 1025.8,
      "end": 1027.866,
      "text": "人体有数百种细胞状态",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 415,
      "start": 1028.166,
      "end": 1030.866,
      "text": "疾病还会让同一种细胞发生变化",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 416,
      "start": 1031.133,
      "end": 1032.8,
      "text": "Atlas可以输出数百种",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 417,
      "start": 1032.8,
      "end": 1034.933,
      "text": "组织与细胞类型的预测",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 418,
      "start": 1035.1,
      "end": 1037.066,
      "text": "可它不可能覆盖所有发育",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 419,
      "start": 1037.066,
      "end": 1039.6,
      "text": "阶段疾病环境和治疗状态",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 420,
      "start": 1039.733,
      "end": 1041.2,
      "text": "开头病例能成功",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 421,
      "start": 1041.566,
      "end": 1043.533,
      "text": "是因为系统恰好拥有与",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 422,
      "start": 1043.533,
      "end": 1046.366,
      "text": "谷氨酸能神经元相关的预测轨迹",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 423,
      "start": 1046.6,
      "end": 1048.933,
      "text": "面对缺失组织数据的疾病",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 424,
      "start": 1049.466,
      "end": 1052.0,
      "text": "分数可能低估真正关键的变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 425,
      "start": 1052.733,
      "end": 1053.8,
      "text": "论文还提醒",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 426,
      "start": 1054.133,
      "end": 1056.933,
      "text": "序列基序地图依赖模型预测质量",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 427,
      "start": 1056.933,
      "end": 1059.766,
      "text": "不能自动证明体内存在直接因果",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 428,
      "start": 1059.8,
      "end": 1061.733,
      "text": "一个高分基序可能与",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 429,
      "start": 1061.733,
      "end": 1063.6,
      "text": "真实调控位点重合",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 430,
      "start": 1063.933,
      "end": 1065.866,
      "text": "也可能只是模型在训练",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 431,
      "start": 1065.866,
      "end": 1067.966,
      "text": "数据中学到的近似模式",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 432,
      "start": 1068.166,
      "end": 1070.066,
      "text": "研究团队对部分转录",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 433,
      "start": 1070.066,
      "end": 1071.9,
      "text": "因子做了实验验证",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 434,
      "start": 1072.4,
      "end": 1075.133,
      "text": "但2,500多类基序不可能全部完成",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 435,
      "start": 1075.133,
      "end": 1078.866,
      "text": "验证Atlas适合生成更好的实验假设",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 436,
      "start": 1079.133,
      "end": 1081.066,
      "text": "最终的生物学结论仍要",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 437,
      "start": 1081.066,
      "end": 1083.766,
      "text": "回到细胞动物与患者数据",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 438,
      "start": 1083.966,
      "end": 1087.0,
      "text": "Alpha genome Atlas不能用于个人医疗决策",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 439,
      "start": 1087.7,
      "end": 1090.366,
      "text": "deep mind在发布页面上明确写明",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 440,
      "start": 1090.5,
      "end": 1092.766,
      "text": "平台没有通过临床用途验证",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 441,
      "start": 1092.866,
      "end": 1094.966,
      "text": "也没有获得任何临床批准",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 442,
      "start": 1095.0,
      "end": 1097.7,
      "text": "不能替代专业诊断或治疗建议",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 443,
      "start": 1097.8,
      "end": 1100.166,
      "text": "学术研究可以免费使用网页",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 444,
      "start": 1100.566,
      "end": 1103.4,
      "text": "基础模型和API也提供研究访问",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 445,
      "start": 1103.966,
      "end": 1106.5,
      "text": "商业用途将通过Google cloud提供",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 446,
      "start": 1106.5,
      "end": 1108.1,
      "text": "这个安排同时说明",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 447,
      "start": 1108.466,
      "end": 1110.466,
      "text": "Atlas既是科学资源",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 448,
      "start": 1110.5,
      "end": 1112.5,
      "text": "也是Google生命科学AI",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 449,
      "start": 1112.5,
      "end": 1114.2,
      "text": "生态的一块基础设施",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 450,
      "start": 1114.5,
      "end": 1115.8,
      "text": "它的商业价值可能",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 451,
      "start": 1115.8,
      "end": 1117.666,
      "text": "晚于科研价值出现",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 452,
      "start": 1118.066,
      "end": 1120.166,
      "text": "制药公司不会因为一个AVI",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 453,
      "start": 1120.3,
      "end": 1122.366,
      "text": "分数立即启动临床试验",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 454,
      "start": 1122.666,
      "end": 1124.9,
      "text": "但可以用它缩小实验范围",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 455,
      "start": 1125.333,
      "end": 1126.7,
      "text": "诊断公司也不能把",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 456,
      "start": 1126.7,
      "end": 1128.4,
      "text": "预测直接写进报告",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 457,
      "start": 1128.666,
      "end": 1130.7,
      "text": "却可以把它与Cleanvar",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 458,
      "start": 1130.8,
      "end": 1133.333,
      "text": "家系信息和功能实验组合",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 459,
      "start": 1133.466,
      "end": 1136.3,
      "text": "谁能把基因组预测稳定接近药物",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 460,
      "start": 1136.3,
      "end": 1139.533,
      "text": "发现患者分层和临床解释流程",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 461,
      "start": 1139.866,
      "end": 1141.333,
      "text": "谁才会把一拍字节",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 462,
      "start": 1141.333,
      "end": 1143.3,
      "text": "数据变成真正的生产力",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 463,
      "start": 1143.466,
      "end": 1145.166,
      "text": "deep mind还把Atlas接进",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 464,
      "start": 1145.166,
      "end": 1147.133,
      "text": "Google Anti Gravity研究袋里",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 465,
      "start": 1147.5,
      "end": 1149.266,
      "text": "未来的科研工作流可能",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 466,
      "start": 1149.266,
      "end": 1151.3,
      "text": "由代理读取病历和文献",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 467,
      "start": 1151.666,
      "end": 1153.8,
      "text": "再调用Atlas排列候选",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 468,
      "start": 1153.9,
      "end": 1156.7,
      "text": "设计验证实验并追踪结果",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 469,
      "start": 1156.8,
      "end": 1158.566,
      "text": "这里产生的变化很深",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 470,
      "start": 1158.766,
      "end": 1161.2,
      "text": "模型不再只给研究者一个分数",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 471,
      "start": 1161.266,
      "end": 1163.366,
      "text": "而是逐渐进入从假设到",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 472,
      "start": 1163.366,
      "end": 1165.133,
      "text": "实验设计的连续流程",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 473,
      "start": 1165.533,
      "end": 1167.466,
      "text": "alphagenome提供分子预测",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 474,
      "start": 1167.6,
      "end": 1169.166,
      "text": "AVI压缩优先级",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 475,
      "start": 1169.333,
      "end": 1171.3,
      "text": "代理系统负责组织步骤",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 476,
      "start": 1171.466,
      "end": 1173.9,
      "text": "三者合在一起才接近Google",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 477,
      "start": 1173.9,
      "end": 1175.566,
      "text": "真正想建的科学平台",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 478,
      "start": 1175.6,
      "end": 1178.266,
      "text": "这条路线延续了alphafold的经验",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 479,
      "start": 1178.566,
      "end": 1181.266,
      "text": "alphafold解决了蛋白质结构预测",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 480,
      "start": 1181.266,
      "end": 1183.366,
      "text": "中的一个核心瓶颈数据库",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 481,
      "start": 1183.766,
      "end": 1186.766,
      "text": "把数以亿计的结构开放给研究社区",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 482,
      "start": 1187.066,
      "end": 1189.3,
      "text": "alphagenome面对的问题更复杂",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 483,
      "start": 1189.533,
      "end": 1192.733,
      "text": "因为DNA调控受细胞类型和环境影响",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 484,
      "start": 1193.2,
      "end": 1195.333,
      "text": "预测结果也更难用一个结构图",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 485,
      "start": 1195.333,
      "end": 1198.533,
      "text": "表达Atlas仍然选择先大规模计算",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 486,
      "start": 1198.533,
      "end": 1200.0,
      "text": "再降低查询门槛",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 487,
      "start": 1200.266,
      "end": 1202.9,
      "text": "Google希望把过去只有少数计算",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 488,
      "start": 1202.9,
      "end": 1205.866,
      "text": "生物学团队能运行的模型变成",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 489,
      "start": 1205.866,
      "end": 1208.733,
      "text": "普通实验室也能使用的公共工具",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 490,
      "start": 1208.9,
      "end": 1210.9,
      "text": "两者之间还有一个关键区别",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 491,
      "start": 1211.333,
      "end": 1212.866,
      "text": "蛋白质结构通常对应",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 492,
      "start": 1212.866,
      "end": 1214.533,
      "text": "相对稳定的物理对象",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 493,
      "start": 1215.066,
      "end": 1216.4,
      "text": "基因变异影响却",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 494,
      "start": 1216.4,
      "end": 1218.166,
      "text": "经常依赖组织和时间",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 495,
      "start": 1218.5,
      "end": 1220.5,
      "text": "Atlas里的地图会随着训练",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 496,
      "start": 1220.5,
      "end": 1222.566,
      "text": "数据与模型更新而改变",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 497,
      "start": 1222.733,
      "end": 1224.366,
      "text": "它更接近一版持续",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 498,
      "start": 1224.366,
      "end": 1226.133,
      "text": "修订的天气预测图",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 499,
      "start": 1226.333,
      "end": 1228.566,
      "text": "而非不可更改的地理测绘",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 500,
      "start": 1229.2,
      "end": 1231.066,
      "text": "科研人员必须记录使用的",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 501,
      "start": 1231.066,
      "end": 1233.133,
      "text": "模型版本和数据库版本",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 502,
      "start": 1233.466,
      "end": 1235.066,
      "text": "否则同一个变异以后",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 503,
      "start": 1235.066,
      "end": 1236.666,
      "text": "可能得到不同解释",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 504,
      "start": 1236.866,
      "end": 1239.4,
      "text": "数据规模本身也会带来治理问题",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 505,
      "start": 1240.166,
      "end": 1242.266,
      "text": "Atlas基于人类参考基因组",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 506,
      "start": 1242.3,
      "end": 1244.966,
      "text": "HG38进行全基因组计算",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 507,
      "start": 1245.1,
      "end": 1246.966,
      "text": "而参考基因组无法完整",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 508,
      "start": 1246.966,
      "end": 1249.266,
      "text": "代表所有人群的遗传多样性",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 509,
      "start": 1249.766,
      "end": 1250.733,
      "text": "论文提出",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 510,
      "start": 1250.733,
      "end": 1252.5,
      "text": "未来还要在不同遗传",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 511,
      "start": 1252.5,
      "end": 1254.466,
      "text": "组员中校准和评估",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 512,
      "start": 1254.666,
      "end": 1256.533,
      "text": "许多基因数据库长期",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 513,
      "start": 1256.533,
      "end": 1258.366,
      "text": "偏向欧洲血统样本",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 514,
      "start": 1258.666,
      "end": 1261.4,
      "text": "如果新模型继续沿用这种分布",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 515,
      "start": 1261.933,
      "end": 1263.533,
      "text": "预测性能可能在代表",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 516,
      "start": 1263.533,
      "end": 1265.266,
      "text": "不足的人群中下降",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 517,
      "start": 1265.466,
      "end": 1267.333,
      "text": "更大的数据库不会自动",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 518,
      "start": 1267.333,
      "end": 1269.4,
      "text": "消除旧数据里的偏差",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 519,
      "start": 1269.566,
      "end": 1271.8,
      "text": "参考基因组还会漏掉结构差异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 520,
      "start": 1272.066,
      "end": 1273.9,
      "text": "单一线性序列很难完整",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 521,
      "start": 1273.9,
      "end": 1275.966,
      "text": "表示不同人群携带的大",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 522,
      "start": 1275.966,
      "end": 1278.966,
      "text": "片段插入倒位和复杂重复区域",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 523,
      "start": 1279.333,
      "end": 1281.566,
      "text": "91种单碱基变化覆盖了",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 524,
      "start": 1281.566,
      "end": 1283.466,
      "text": "一个清晰的问题空间",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 525,
      "start": 1283.7,
      "end": 1285.066,
      "text": "却没有覆盖所有",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 526,
      "start": 1285.066,
      "end": 1286.866,
      "text": "可能的人类遗传变化",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 527,
      "start": 1287.266,
      "end": 1289.266,
      "text": "未来更完整的图基因组和",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 528,
      "start": 1289.266,
      "end": 1292.166,
      "text": "泛基因组参考可能要求Atlas",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 529,
      "start": 1292.166,
      "end": 1294.066,
      "text": "重新计算一部分地图",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 530,
      "start": 1294.266,
      "end": 1295.7,
      "text": "发布规模很大",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 531,
      "start": 1295.7,
      "end": 1297.1,
      "text": "边界仍然明确",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 532,
      "start": 1297.266,
      "end": 1298.533,
      "text": "数据更新也会改变",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 533,
      "start": 1298.533,
      "end": 1300.333,
      "text": "研究可重复性的要求",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 534,
      "start": 1300.866,
      "end": 1302.133,
      "text": "团队在论文中提到",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 535,
      "start": 1302.2,
      "end": 1304.4,
      "text": "英国生物样本库研究曾因分析",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 536,
      "start": 1304.4,
      "end": 1307.366,
      "text": "平台临时停机使用略早版本的Atlast",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 537,
      "start": 1307.566,
      "end": 1309.4,
      "text": "并检查了发布版本与",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 538,
      "start": 1309.4,
      "end": 1311.133,
      "text": "旧版本之间的相关性",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 539,
      "start": 1311.466,
      "end": 1313.666,
      "text": "模型和数据库持续更新以后",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 540,
      "start": 1313.8,
      "end": 1315.666,
      "text": "论文需要保存版本号",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 541,
      "start": 1315.733,
      "end": 1318.066,
      "text": "查询日期和具体评分轨迹",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 542,
      "start": 1318.333,
      "end": 1320.766,
      "text": "否则几年后重新查同一变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 543,
      "start": 1321.1,
      "end": 1322.7,
      "text": "研究者可能无法解释",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 544,
      "start": 1322.7,
      "end": 1324.4,
      "text": "为何分数发生变化",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 545,
      "start": 1324.666,
      "end": 1327.166,
      "text": "研究资源的开放范围也值得注意",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 546,
      "start": 1327.6,
      "end": 1330.8,
      "text": "网页目前面向非商业研究免费开放",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 547,
      "start": 1331.1,
      "end": 1333.166,
      "text": "商业服务将进入Google cloud",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 548,
      "start": 1333.5,
      "end": 1335.766,
      "text": "公开访问降低了学术门槛",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 549,
      "start": 1336.333,
      "end": 1338.8,
      "text": "但EPB预计算结果和持续",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 550,
      "start": 1338.8,
      "end": 1341.266,
      "text": "更新能力仍掌握在Google手中",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 551,
      "start": 1341.533,
      "end": 1344.133,
      "text": "API可以让实验室快速工作",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 552,
      "start": 1344.3,
      "end": 1346.066,
      "text": "也可能把越来越多科研",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 553,
      "start": 1346.066,
      "end": 1348.266,
      "text": "流程绑定到单一云平台",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 554,
      "start": 1349.066,
      "end": 1351.7,
      "text": "alphagenome的科学贡献十分明确",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 555,
      "start": 1351.966,
      "end": 1353.666,
      "text": "基础设施集中带来的",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 556,
      "start": 1353.666,
      "end": 1355.9,
      "text": "依赖同样需要长期观察",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 557,
      "start": 1355.966,
      "end": 1357.2,
      "text": "对普通人而言",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 558,
      "start": 1357.366,
      "end": 1359.666,
      "text": "这项发布最重要的意义不在于",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 559,
      "start": 1359.666,
      "end": 1362.0,
      "text": "马上得到一份基因命运报告",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 560,
      "start": 1362.566,
      "end": 1365.666,
      "text": "真正变化发生在看不见的地方过去",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 561,
      "start": 1365.7,
      "end": 1368.1,
      "text": "医生找到一个意义不明的变异后",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 562,
      "start": 1368.1,
      "end": 1370.3,
      "text": "可能要等待多年文献积累",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 563,
      "start": 1370.5,
      "end": 1373.366,
      "text": "现在研究者能先查看它对特定",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 564,
      "start": 1373.366,
      "end": 1375.666,
      "text": "组织简接和表达的预测",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 565,
      "start": 1375.766,
      "end": 1377.733,
      "text": "再决定是否投入实验",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 566,
      "start": 1378.066,
      "end": 1380.733,
      "text": "一次诊断仍然需要严格证据",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 567,
      "start": 1380.866,
      "end": 1383.6,
      "text": "但寻找证据的路线变得更短",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 568,
      "start": 1383.666,
      "end": 1385.2,
      "text": "对科研机构而言",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 569,
      "start": 1385.333,
      "end": 1387.466,
      "text": "Atlas把实验资源从广",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 570,
      "start": 1387.466,
      "end": 1389.166,
      "text": "撒网推向精准验证",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 571,
      "start": 1389.533,
      "end": 1391.133,
      "text": "九十亿种候选永远",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 572,
      "start": 1391.133,
      "end": 1392.866,
      "text": "无法全部进入实验",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 573,
      "start": 1393.1,
      "end": 1396.066,
      "text": "AI的价值是把最可能产生信息",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 574,
      "start": 1396.066,
      "end": 1398.5,
      "text": "增量的那一小部分提到前面",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 575,
      "start": 1398.8,
      "end": 1400.4,
      "text": "失败实验不会消失",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 576,
      "start": 1400.533,
      "end": 1402.4,
      "text": "候选排序只会更合理",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 577,
      "start": 1402.533,
      "end": 1404.4,
      "text": "随着真实实验不断回流",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 578,
      "start": 1404.466,
      "end": 1406.866,
      "text": "模型还能发现自己在哪些细胞",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 579,
      "start": 1407.066,
      "end": 1409.466,
      "text": "哪些调控模式上最容易出错",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 580,
      "start": 1409.5,
      "end": 1411.366,
      "text": "下一代地图会继续改进",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 581,
      "start": 1411.533,
      "end": 1413.9,
      "text": "它还可能改变科研竞争的节奏",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 582,
      "start": 1414.3,
      "end": 1416.266,
      "text": "过去拥有大型计算团队的",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 583,
      "start": 1416.266,
      "end": 1418.566,
      "text": "机构可以更快跑完变异评分",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 584,
      "start": 1418.7,
      "end": 1420.1,
      "text": "小型实验室则要",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 585,
      "start": 1420.1,
      "end": 1421.9,
      "text": "等待合作或购买服务",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 586,
      "start": 1422.0,
      "end": 1423.8,
      "text": "统一Atlas把第一轮",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 587,
      "start": 1423.8,
      "end": 1425.6,
      "text": "计算变成公共底座",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 588,
      "start": 1426.333,
      "end": 1429.5,
      "text": "差距会转移到谁能提出更好的问题",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 589,
      "start": 1429.733,
      "end": 1431.366,
      "text": "获得稀缺患者样本",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 590,
      "start": 1431.366,
      "end": 1433.133,
      "text": "并设计高质量验证",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 591,
      "start": 1433.366,
      "end": 1435.333,
      "text": "算力的重要性没有消失",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 592,
      "start": 1435.5,
      "end": 1437.566,
      "text": "只是基础预测不再需要",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 593,
      "start": 1437.566,
      "end": 1440.1,
      "text": "每家实验室重复支付一次",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 594,
      "start": 1440.166,
      "end": 1442.4,
      "text": "药物公司的决策也可能提前",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 595,
      "start": 1442.933,
      "end": 1444.4,
      "text": "一个疾病相关区域",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 596,
      "start": 1444.4,
      "end": 1446.366,
      "text": "如果缺少明确作用机制",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 597,
      "start": 1446.7,
      "end": 1449.666,
      "text": "团队通常难以判断该瞄准哪个基因",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 598,
      "start": 1450.1,
      "end": 1452.166,
      "text": "lookalike可以显示某组变异主要",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 599,
      "start": 1452.166,
      "end": 1454.333,
      "text": "影响哪个组织的表达或剪接",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 600,
      "start": 1454.666,
      "end": 1455.966,
      "text": "帮助研究人员决定",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 601,
      "start": 1455.966,
      "end": 1457.666,
      "text": "先做哪类细胞实验",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 602,
      "start": 1457.8,
      "end": 1459.8,
      "text": "它不会直接给出药物分子",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 603,
      "start": 1459.933,
      "end": 1461.733,
      "text": "更不会保证靶点有效",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 604,
      "start": 1462.0,
      "end": 1464.8,
      "text": "却能减少候选机制之间的盲目试错",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 605,
      "start": 1464.966,
      "end": 1467.133,
      "text": "真正节省的是研发最",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 606,
      "start": 1467.133,
      "end": 1469.266,
      "text": "早期反复走错方向的时间",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 607,
      "start": 1469.5,
      "end": 1471.733,
      "text": "Alpha GENOME Atlas也揭示了",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 608,
      "start": 1471.733,
      "end": 1473.733,
      "text": "科学AI的一条成熟路径",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 609,
      "start": 1474.166,
      "end": 1475.466,
      "text": "第一步训练模型",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 610,
      "start": 1475.666,
      "end": 1477.6,
      "text": "预测可测量的生物现象",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 611,
      "start": 1477.6,
      "end": 1480.866,
      "text": "重点不放在生成一段像论文的文字",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 612,
      "start": 1481.066,
      "end": 1484.0,
      "text": "第二步把预测扩展到整个问题空间",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 613,
      "start": 1484.3,
      "end": 1487.1,
      "text": "形成所有研究者都能查询的资源",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 614,
      "start": 1487.466,
      "end": 1490.2,
      "text": "第三步让独立实验检验关键结果",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 615,
      "start": 1490.366,
      "end": 1491.733,
      "text": "只有完成这三步",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 616,
      "start": 1491.733,
      "end": 1494.0,
      "text": "AI才真正改变知识生产",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 617,
      "start": 1494.0,
      "end": 1496.6,
      "text": "而非增加更多未经验证的答案",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 618,
      "start": 1496.766,
      "end": 1498.333,
      "text": "开头那个病例给这条",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 619,
      "start": 1498.333,
      "end": 1500.4,
      "text": "路径提供了一个清晰样本",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 620,
      "start": 1501.0,
      "end": 1504.0,
      "text": "算法从数百万差异中挑出DNME",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 621,
      "start": 1504.533,
      "end": 1506.0,
      "text": "生成内涵子变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 622,
      "start": 1506.166,
      "end": 1508.4,
      "text": "预测脑细胞里的异常剪接",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 623,
      "start": 1508.533,
      "end": 1511.1,
      "text": "又用实验确认该区域的功能",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 624,
      "start": 1511.2,
      "end": 1513.333,
      "text": "它没有独立完成临床诊断",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 625,
      "start": 1513.466,
      "end": 1514.933,
      "text": "也没有治愈患者",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 626,
      "start": 1515.066,
      "end": 1516.533,
      "text": "却让一个长期藏在",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 627,
      "start": 1516.533,
      "end": 1519.1,
      "text": "非编码区里的机制浮出水面",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 628,
      "start": 1519.366,
      "end": 1520.966,
      "text": "对罕见病家庭而言",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 629,
      "start": 1521.0,
      "end": 1523.933,
      "text": "知道病因可能帮助评估复发风险",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 630,
      "start": 1524.1,
      "end": 1525.5,
      "text": "寻找同类病例",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 631,
      "start": 1525.666,
      "end": 1529.066,
      "text": "并为未来治疗研究建立起点所以",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 632,
      "start": 1529.4,
      "end": 1532.4,
      "text": "alphagenome Atlas的突破不在于AI",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 633,
      "start": 1532.4,
      "end": 1534.266,
      "text": "已经读懂全部人类基因",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 634,
      "start": 1534.366,
      "end": 1536.133,
      "text": "它第一次把每一种可能的",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 635,
      "start": 1536.133,
      "end": 1538.6,
      "text": "单字母变化都放进同一套",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 636,
      "start": 1538.6,
      "end": 1540.866,
      "text": "可搜索的分子预测地图",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 637,
      "start": 1540.933,
      "end": 1543.0,
      "text": "并把蛋白质编码区与更",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 638,
      "start": 1543.0,
      "end": 1545.2,
      "text": "庞大的非编码区连接起来",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 639,
      "start": 1545.533,
      "end": 1547.7,
      "text": "九十亿是计算覆盖的规模",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 640,
      "start": 1547.733,
      "end": 1549.933,
      "text": "不是已经证实的答案数量",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 641,
      "start": 1550.0,
      "end": 1552.533,
      "text": "一拍字节是资源体量",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 642,
      "start": 1552.6,
      "end": 1554.9,
      "text": "也不是临床可靠性的保证",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 643,
      "start": 1555.0,
      "end": 1556.666,
      "text": "这张地图会把基因",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 644,
      "start": 1556.666,
      "end": 1558.4,
      "text": "研究的瓶颈向后移动",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 645,
      "start": 1558.6,
      "end": 1560.2,
      "text": "研究者以后更少困在",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 646,
      "start": 1560.2,
      "end": 1561.7,
      "text": "应该先看哪一个变异",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 647,
      "start": 1562.133,
      "end": 1565.6,
      "text": "更多困在怎样用实验确认这项预测",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 648,
      "start": 1565.733,
      "end": 1567.366,
      "text": "这是更高级的难题",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 649,
      "start": 1567.466,
      "end": 1569.8,
      "text": "也是科学真正前进的方式",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 650,
      "start": 1570.0,
      "end": 1571.966,
      "text": "AI把搜索空间压缩",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 651,
      "start": 1572.133,
      "end": 1573.7,
      "text": "人类负责判断证据",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 652,
      "start": 1573.933,
      "end": 1575.733,
      "text": "两者各自守住边界",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 653,
      "start": 1575.966,
      "end": 1578.333,
      "text": "alphagenome Atlas才可能成为",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    },
    {
      "id": 654,
      "start": 1578.333,
      "end": 1581.266,
      "text": "基因组学里的下一座alphafold数据库",
      "source": "video_source_subtitle",
      "authority": "video_source_subtitle"
    }
  ],
  "transcript_selection": {
    "schema": "videonote.transcript_selection.v1",
    "status": "SELECTED",
    "requested_language": "zh",
    "duration": 1582.0,
    "reference_path": "/Users/sagawa/AI_WORK/video_notes/url/20260910_193708__Google最強AI誕生_全人類90億種基因病因已被_預先計算_完成/reference_subtitle.vtt",
    "reference_kind": "manual",
    "reference_language": "zh",
    "selected_source": "video_source_subtitle",
    "source_caption_kind": "manual",
    "source_caption_processing": "verbatim_direct",
    "youtube_overlap_clamps": [],
    "general_high_confidence_repairs": [],
    "decision_reason": "complete_source_subtitle_fast_path_no_whisper",
    "whisper_rejected": false,
    "source_profile": {
      "segment_count": 655,
      "duration_coverage": 0.926573
    }
  }
}