Mechanism-guided drug repurposing for host-directed therapy of infectious diseases using interpretable and integrative ML

摘要 我们治疗传染病的能力受到两个主要问题的阻碍。一个是快速增长的 一个是抗菌素耐药性，另一个是发现新药所需的高昂成本和时间。一个 克服这些问题的潜在方法是专注于改变现有药物的用途，以用于宿主导向的药物 心理治疗。然而，这是一个新兴的应用领域。虽然有几项研究使用了这一广泛的方法来 寻找针对特定病毒和细菌感染的候选药物，缺乏系统的计算 可用于重新调整治疗任何传染病的药物用途的框架，特别是那些专注于药物的框架 以及疾病机制，而不是个别药物和靶向特性。此外，还缺少以下框架 可以利用可用于非传染性疾病的海量数据和知识来应对 传染病治疗。在这个项目中，我们将开发一个综合框架，使用机制- 指导的，可解释的机器学习(ML)模型，以重新调整药物的用途，以增强宿主对感染的反应。 我们的框架利用了大量的转录组数据收集和基因组规模的人类基因交互 网络；这是关于与此相关的分子机制的两个互补信息来源 改变用途的努力。它还使用关于数百种非传染性疾病和数千种非传染性疾病的数据和知识 小分子药物(包括FDA批准的药物)，以创造大量的再利用机会。要求 只有宿主应对感染的转录组数据，我们的通用ML框架将适用于 新的、新兴的和未被充分研究的传染病。该项目还将产生高置信度药物 几种传染病的候选者以及对宿主定向新途径的机械性见解 治疗学。

ABSTRACT Our ability to treat infectious diseases is impeded by two major problems. One is the rapid increase of antimicrobial resistance, and the other is the prohibitive cost and time required for discovering new drugs. A potential approach to overcome these problems is to focus on repurposing existing drugs for host-directed therapy. However, this is an emerging application area. While several studies have used this broad approach to find drug candidates for specific viruses and bacterial infections, there is a dearth of systematic computational frameworks that can be used to repurpose drugs for any infectious disease, especially ones that focus on drug and disease mechanisms rather than individual drug and target properties. Also missing are frameworks that can leverage the massive amounts of data and knowledge available for non-infectious diseases to tackle infectious disease treatment. In this project, we will develop an integrative framework that uses mechanism- guided, interpretable machine learning (ML) models to repurpose drugs to bolster host response to infection. Our framework leverages massive transcriptome data collections and genome-scale human gene interaction networks; these are two complementary sources of information about molecular mechanisms relevant for this repurposing effort. It also uses data and knowledge about hundreds of non-infectious diseases and thousands of small molecules (including FDA-approved drugs) to create numerous repurposing opportunities. Requiring only host transcriptome data in response to infection, our general-purpose ML framework will be applicable to new, emerging, and understudied infectious diseases. This project will also result in high-confidence drug candidates for several infectious diseases along with mechanistic insights into new avenues for host-directed therapeutics.