Enabling AI-based Mouse Genetic Discovery

实现基于人工智能的小鼠基因发现

• 批准号: 10724522
• 负责人: GARY A PELTZ
• 金额: $ 77.97万
• 依托单位: STANFORD UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2023
• 资助国家: 美国
• 起止时间: 2023-08-01 至 2027-07-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10724522
• 关键词: Acceleration; Affect; Alleles; Amino Acid Sequence; Artificial Intelligence; Candidate Disease Gene; Clustered Regularly Interspaced Short Palindromic Repeats; Communities; DNA Sequence Alteration; Data; Data Analyses; Data Set; Data Sources; Databases; Development; Diabesity; Diabetes Mellitus; Disease; Disease model; Engineering; Evaluation; Exhibits; Exons; Foundations; Generations; Genes; Genetic; Genetic Enhancement; Genetic Models; Genetic Variation; Genome engineering; Genomics; Health; Healthcare; Hodgkin Disease; Homologous Gene; Human; Inbred Strain; Individual; Knock-in Mouse; Knock-out; Laboratory mice; Lymphoma; Machine Learning; Malignant Neoplasms; Maps; Measures; Methods; Mus; Network-based; Obesity; Paper; Pattern; Phenotype; Public Health; Publishing; Research; Tandem Repeat Sequences; Training; Validation; Variant; candidate identification; causal variant; computational pipelines; computerized tools; disease phenotype; genetic variant; genome wide association study; graph neural network; human disease; human model; improved; innovation; knockout gene; model organism; mouse genetics; mouse genome; mouse model; novel; protein protein interaction; public database; response; trait

Abstract No model organism has contributed more than the laboratory mouse to improving human health. Many genetic factors and therapies for human diseases were initially discovered or characterized in mice, before they were transitioned to human use. Large-scale efforts are underway to integrate recent advances in artificial intelligence (AI) into human healthcare, but very few AI advances have been used for analysis of the data produced using the model organism that has formed the foundation for many healthcare innovations. We recently developed an AI-based computational pipeline that could identify causative genetic factors for murine genetic models of human biomedical traits and diseases. After assessing the strength of allelic associations with the phenotypic response pattern exhibited by the inbred strains; this AI pipeline uses a machine-learning trained method to analyze 29M published papers and assess candidate gene-phenotype relationships; and the information obtained from assessment of their protein-protein interaction network and protein sequence features of the candidate genes are also incorporated into the graph neural network-based analysis. This project will produce a markedly enhanced AI pipeline (AIv2) that will greatly accelerate the pace of genetic discovery using murine genetic models. First, long read genomic sequencing (LRS) and computational tools are used to produce a more complete map of the pattern of genetic variation among the inbred strains, which also includes alleles for two major types of genetic variation (structural variants, tandem repeats), which are poorly characterized using conventional sequencing methods. Second, we develop two additional computational tools for the AI, which facilitate candidate gene prioritization through the evaluation of: (i) the phenotypes exhibited by 8200 mouse lines with individual gene knockouts (KOs); and (ii) the results of 5700 human GWAS covering many biomedical phenotypes to determine if alleles within the human homologues of candidate murine genes affect an analyzed trait. The ability of AIv2 to accelerate genetic discovery will be demonstrated by using it to identify new genetic factors through analysis of a public database with >10,307 datasets, which measure biomedical or disease-related responses in panels of inbred strains. Since it is critical to experimentally confirm some of the computational findings, genetic factors for two murine models of human diseases that are major public health problems (cancer, diabetes/obesity), which were identified by the AI pipeline, will be experimentally validated. CRISPR engineering is used to revert the causative mutation(s) to wildtype on the genetic background of the strain exhibiting the disease phenotype, and the genome engineered mice are analyzed to assess the contribution of the genetic factor to the disease phenotype.

抽象的 没有任何模型生物体比实验室小鼠对改善人类健康的贡献更大。很多遗传 人类疾病的因素和疗法最初是在小鼠身上发现或表征的，然后才被研究出来。 转变为人类使用。 正在进行大规模的努力来整合人工技术的最新进展 人工智能（AI）进入人类医疗保健领域，但很少有人工智能进步被用于数据分析 使用模型生物体生产，该生物体已成为许多医疗保健创新的基础。我们 最近开发了一种基于人工智能的计算管道，可以识别小鼠的致病遗传因素 人类生物医学特征和疾病的遗传模型。评估等位基因关联的强度后 具有近交系表现出的表型反应模式；这个人工智能管道使用机器学习 训练有素的方法来分析 2900 万篇发表的论文并评估候选基因-表型关系；和 通过评估蛋白质-蛋白质相互作用网络和蛋白质序列获得的信息 候选基因的特征也被纳入基于图神经网络的分析中。 该项目将产生显着增强的人工智能管道（AIv2），这将大大加快遗传研究的步伐 使用小鼠遗传模型进行发现。一、长读长基因组测序（LRS）和计算工具 用于生成近交系遗传变异模式的更完整图谱， 还包括两种主要遗传变异类型（结构变异、串联重复）的等位基因，它们是 使用传统测序方法很难表征。其次，我们开发了另外两个 人工智能的计算工具，通过评估以下内容促进候选基因优先排序：（i） 8200 个个体基因敲除 (KO) 小鼠品系表现出的表型； (ii) 5700 的结果 人类 GWAS 涵盖许多生物医学表型，以确定等位基因是否属于人类同源物 候选小鼠基因影响分析的性状。 AIv2 加速基因发现的能力将是 通过对包含 >10,307 个公共数据库的分析，使用它来识别新的遗传因素，从而证明了这一点 数据集，用于测量近交菌株组中的生物医学或疾病相关反应。既然很关键 为了通过实验证实一些计算结果，人类的两种小鼠模型的遗传因素 人工智能识别出的重大公共卫生问题疾病（癌症、糖尿病/肥胖） 管道，将进行实验验证。 CRISPR 工程用于将致病突变恢复为 表现出疾病表型的菌株遗传背景的野生型，以及基因组工程 对小鼠进行分析以评估遗传因素对疾病表型的贡献。

项目成果

Genetic Discovery Enabled by A Large Language Model.

由大型语言模型实现的基因发现。

• 发表时间: 2023-11-12
• 作者: Tu, Tao;Fang, Zhouqing;Cheng, Zhuanfen;Spasic, Svetolik;Palepu, Anil;Stankovic, Konstantina M;Natarajan, Vivek;Peltz, Gary
• 通讯作者: Peltz, Gary