RAPID: Collaborative Research: Computational Drug Repurposing for COVID-19

RAPID：合作研究：针对 COVID-19 的计算药物再利用

• 批准号: 2030477
• 负责人: Jurij Leskovec
• 金额: $ 10万
• 依托单位: Stanford University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-05-01 至 2021-04-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2030477&HistoricalAwards=false
• 关键词: RAPID; Collaborative; Research; Computational; Drug

With the disruptive nature of the COVID-19 pandemic, effective treatments could save the lives of severely ill patients, protect individuals with a high risk of infection, and reduce the time patients spend in hospital beds. However, there are currently no effective treatments for COVID-19. Traditional methodologies take years to develop and test compounds from scratch. Machine learning provides promising new approaches to repurpose drugs that are safe and already approved for other diseases. This project will develop a machine learning toolset to expedite the development of safe and effective medicines for COVID-19. The toolset will rapidly identify safe repurposing opportunities for approved and experimental drugs. It will predict whether treatments may have therapeutic effects in COVID-19 patients, allowing the identification of drugs and drug cocktails that are safe and plentiful enough to treat a substantial number of patients. By putting tools in the hand of practitioners, the activities in this project will have an immediate impact. They will result in actionable predictions that are accurate and interpretable. Recently, the principal investigators have developed a series of machine learning tools to identify drug repurposing opportunities. Building on foundational previous work, in this project, the principal investigators will first build a large COVID-19 focused knowledge graph that will capture fundamental and COVID-19-specific biological knowledge. The graph learning methods will be adapted to identify safe drugs and drug cocktails for COVID-19. To predict the safety of cocktails with two or more drugs, the methods will generalize to an exponentially large space of high-order drug combinations. In addition to drug safety, efficacy is a crucial endpoint for drug development. The project will develop a novel graph neural network (GNN) method to identify efficacious drug repurposing opportunities, even for diseases, such as COVID-19, that do not yet have any drug treatments and thereby, no label, supervised information. The method will predict what drugs and drug combinations may have a therapeutic effect on COVID-19. Finally, the principal investigators will integrate the developed tools into a complete, explainable framework that will generate predictions, provide explanations, and incorporate human feedback into the machine learning loop. This project will provide new, open tools for rapid drug repurposing that will be relevant for COVID-19 and other emerging pathogens. Additionally, the project will provide unique opportunities for multi-disciplinary curriculum development, training and advising, and professional activities.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

鉴于COVID-19大流行的破坏性，有效的治疗可以挽救重症患者的生命，保护高感染风险的个人，并减少患者在医院病床上的时间。然而，目前并无有效治疗COVID-19的方法。传统的方法需要数年时间从头开始开发和测试化合物。机器学习提供了有前途的新方法来重新利用安全且已被批准用于其他疾病的药物。该项目将开发一套机器学习工具，以加快开发安全有效的新型冠状病毒药物。该工具集将快速识别已批准和实验性药物的安全再利用机会。它将预测治疗方法是否可能对COVID-19患者产生治疗效果，从而确定安全且足够治疗大量患者的药物和药物鸡尾酒。通过将工具交给实践者，该项目中的活动将立即产生影响。它们将导致准确和可解释的可操作的预测。最近，主要研究人员开发了一系列机器学习工具来识别药物再利用的机会。在此项目中，主要研究人员将在前期基础工作的基础上，首先构建一个大型的COVID-19重点知识图谱，以获取基础和COVID-19特异性生物学知识。图学习方法将被用于识别COVID-19的安全药物和药物鸡尾酒。为了预测含有两种或更多种药物的鸡尾酒的安全性，该方法将推广到指数级大的高阶药物组合空间。除了药物安全性外，疗效也是药物开发的关键终点。该项目将开发一种新的图形神经网络（GNN）方法，以识别有效的药物再利用机会，即使是对于尚未进行任何药物治疗的疾病，如COVID-19，也没有标签，监督信息。该方法将预测哪些药物和药物组合可能对COVID-19有治疗作用。最后，主要研究人员将把开发的工具整合到一个完整的、可解释的框架中，该框架将生成预测，提供解释，并将人类反馈纳入机器学习循环。该项目将为快速药物再利用提供新的开放工具，这些工具将与COVID-19和其他新兴病原体相关。此外，该项目将为多学科课程开发、培训和咨询以及专业活动提供独特的机会。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

QA-GNN: Reasoning with Language Models and Knowledge Graphs for Question Answering

• DOI: 10.18653/v1/2021.naacl-main.45
• 发表时间: 2021-04
• 作者: Michihiro Yasunaga;Hongyu Ren;Antoine Bosselut;Percy Liang;J. Leskovec

TEDIC: Neural Modeling of Behavioral Patterns in Dynamic Social Interaction Networks

• DOI: 10.1145/3442381.3450096
• 发表时间: 2021-04
• 期刊: Proceedings of the Web Conference 2021
• 作者: Yanbang Wang;Pan Li;Chongyang Bai;J. Leskovec

Modeling Heterogeneous Hierarchies with Relation-specific Hyperbolic Cones

• 发表时间: 2021-10
• 作者: Yushi Bai;Rex Ying;Hongyu Ren;J. Leskovec

Language-Agnostic Representation Learning of Source Code from Structure and Context

• 发表时间: 2021-03
• 期刊: ArXiv
• 作者: Daniel Zugner;Tobias Kirschstein;Michele Catasta;J. Leskovec;Stephan Gunnemann

Neural Distance Embeddings for Biological Sequences

• 发表时间: 2021-09
• 作者: Gabriele Corso;Rex Ying;Michal P'andy;Petar Velivckovi'c;J. Leskovec;P. Lio’