AI models of multi-omic data integration for ming longevity core signaling pathways

项目摘要 异常长寿（EL）与异常健康跨度、低风险和延迟发病密切相关。 与年龄有关的疾病。此外，EL是一种复杂的遗传性状，就像与衰老相关的疾病一样，受以下因素的影响： 多基因靶点和其他因素，如性别、种族、生活方式选择、社会和环境因素。 因此，在EL研究中，单一的保护性遗传靶点通常对极端年龄的生存具有较弱的影响。 然而，基因靶点以及其他因素的正确组合可以产生更强的效果。 因此，发现这些保护性因子、遗传靶点和后续信号传导具有重要意义 EL的途径，这是指导新药开发和管理的重要基础 用于疾病预防/治疗，以延长健康和寿命。大规模和多组学数据集，如 基因组、表观基因组、转录组、蛋白质组、代谢组、微生物组、表型组， 百岁老人和特别长寿的人，已经在多个EL项目中产生。 然而，整合和解释复杂的多组学数据集仍然具有挑战性。 为了响应NIH RFA-AG-23-033，我们建议改进和开发新型人工智能（AI） 模型，可以有效地集成和解释EL多组学数据集，并识别风险和保护 靶点和药物，以纠正疾病风险信号传导途径， 健康寿命延长。深度学习（DL）和AI模型已广泛应用于医疗保健领域 并超越传统的机器学习模型，从而为这个关键问题提供解决方案。我们 在开发用于目标排名的多组学数据分析的可解释AI模型方面拥有丰富的经验， 核心信令网络推理。在这项研究中，我们将（目标1）开发两个（GNN）AI模型，PathFormer和 使用多组学数据进行无偏核心信号通路推断（无偏/无指导 推理）;（目标2）：开发一种新的GNN AI模型，模块化k-Hop DeepNetFlow，用于假设引导核心 使用多组学数据的信号通路推断（半引导推断）;（目的4）：开发新的 DeepDrugMap知识图谱和知识驱动的多模块多证据（M3 E）模型， 预测可以增强保护信号并抑制疾病风险信号通路的药物 预防/治疗;开发一种新颖的开源可视化编程工具LongevityOmicNet，以支持 利用不同的支持性数据集对人工智能模型进行传播和可再现的分析， EL或老化研究社区。此外（目标3）：与Michael Province博士（主要研究者之一）合作，领导LLFS 在华盛顿大学的一个项目中，我们将应用这些人工智能模型来识别与EL相关的保护因素，比如性别， 遗传学、胰岛素抵抗、环境因素（SGIE因子）和相关信号通路/生物学 过程，使用EL研究的大规模多组学数据，即，LLFS、LG和劳工组织的研究。

PROJECT SUMMARY Exceptional longevity (EL) is strongly correlated with exceptional health span, lower risk and delayed onset of age-related diseases. Moreover, EL is a complex genetic trait, like aging-related diseases, affected by polygenic targets, and other factors, like sex, ethnicity, lifestyle choices, social and environmental factors. Thus, in EL studies, single protective genetic targets usually have weaker effects upon survival to extreme age. Whereas, the right combination of genetic targets, as well as other factors, can have a stronger effect. Therefore, it is important to discover these protective factors, genetic targets and subsequent signaling pathways of EL, which are the critical basis to guide the development of novel medications and management for disease prevention/treatment to extend health and life span. Large-scale and multi-omics datasets, like genome, epigenome, transcriptome, proteome, metabolome, microbiome, phenome, of large-scale cohorts of centenarians and exceptional long-lived individuals, have been being generated in multiple EL projects. Whereas, it remains challenging to integrate and interpret complex multi-omics datasets. In response to the NIH RFA-AG-23-033, we propose to improve and develop novel artificial intelligence (AI) models that can efficiently integrate and interpret the EL multi-omics datasets, and identify risk and protective targets and medications to correct the disease risk signaling pathways for disease prevention and long and healthy life span extension. Deep learning (DL) and AI models have been widely used in the healthcare field and outperform traditional machine learning models, and thus offering solutions to this critical problem. We have rich experience in developing interpretable AI models of multi-omics data analysis for target ranking and core signaling network inference. In this study, we will (Aim 1) develop two (GNN) AI models, PathFormer and PathFinder, for unbiased core signaling pathways inference using multi-omics data (unbiased/unguided inference); (aim 2): develop a novel GNN AI model, modular k-Hop DeepNetFlow, for hypothesis guided core signaling pathway inference using multi-omics data (semi-guided inference); (Aim 4): develop novel DeepDrugMap knowledge graph, and Knowledge-driven, Multi-Module, Multi-Evidence (M3E) models to predict drugs that can boost protective signaling and inhibit the risk signaling pathways for disease prevention/treatment; develop a novel, open-source visual programming tool, LongevityOmicNet, to support the dissemination and reproducible analysis of the AI models with diverse supportive datasets, to the broader EL or aging study community. Also (Aim 3): collaborating with Dr. Michael Province (Co-PI), leading the LLFS project in WashU, we will apply these AI models to identify EL-associated protective factors, like the Sex, Genetics, Insulin resistance, Environment factors (SGIE-factors), and associated signaling pathways/biological processes, using large-scale multi-omics data of EL studies, i.e., LLFS, LG and ILO studies.