Novel computational approaches to predict drug response and combination effects

点击翻译按钮获取中文摘要

Summary Tailoring the most desired therapy to each individual patient is the primary goal of precision medicine. A reliable and robust predictive model of drug effectiveness based on patients' unique genomic background is the key. For decades, communities have been trying to establish the relationship between molecular characteristics and drug response in complex diseases. Over the last decade, a large amount of genomic and epigenomic data together with pharmacogenomics data and response to perturbations data has been generated for many human cell lines through collaborations in the research community. These projects have led to significant therapeutic discoveries and have provided unprecedented opportunities to predict drug response using molecular fingerprints. However, even with great interest and effort in developing computational methods for predicting drug response, the prediction accuracies are at best only moderate. A related but distinct question is to understand the mechanisms of action (MOA) of drugs. Understanding drug MOAs enables characterization of drug side effects and identification of old drugs for new uses (i.e. drug repositioning). The traditional experimental assays to identify MOAs of drugs are expensive and time- consuming. There are three key questions to be addressed in the study. 1. Can novel computational approaches largely improve prediction accuracy of response to single drugs using comprehensive genomic and chemical information? 2. Can computational approaches provide a systematic way to mine genomics and drug response data to generate biological insights into the mechanisms of actions of various drugs? 3. Is it possible to develop an interpretable and accurate computation model to predict drug combination effects using pharmacogenomics data? Inherent features make it very challenging to predict drug response accurately: High-dimensionality of input data, the complex relationship between input features and response data; and heterogeneous drug/compound response patterns across different genetic lineages. Recently, artificial intelligence (AI) has been making remarkable strides in various applications owing to the rapid progress of “deep learning. In Aim 1 of this study, we will develop novel AI-based approaches to address the computational challenges of improving the prediction accuracy of drug response. In Aim 2 of the study, we will develop a novel computation framework to study of MOA of drugs. In Aim 3, we will develop an interpretable deep-learning based computational framework to predict drug combination effects. In addition, we will develop a user-friendly web portal as an integrated research platform to share the methodology, algorithms and data generated from this proposed study to the research community.