Deep Transfer Learning of Drug Responses by Integrating Bulk and Single-cell RNA-seq data

Abstract

Massively bulk RNA sequencing databases incorporating drug screening have opened up an avenue to inform the optimal clinical application of cancer drugs. Meanwhile, the growing single-cell RNA sequencing (scRNA-seq) data contributes to improving therapeutic effectiveness by studying the heterogeneity of drug responses for cancer cell subpopulations. There is a clear significance in developing computational biology approaches to predict and interpret cancer drug response in single cell data from clinical samples. Here, we introduce scDEAL, a deep transfer learning framework for cancer drug response prediction at single-cell level by integrating large-scale bulk cell line data. The true innovation of scDEAL is to translate cancer cell line drug responses into predicting clinical drug responses via learning relations of gene expressions and drug responses at bulk-level and transfer to predict drug responses in scRNA-seq. Another innovation is the integrated gradient feature interpretation to infer a comprehensive set of signature genes to reveal potential drug resistance mechanisms. We benchmarked scDEAL on six scRNA-seq datasets and indicate its model interpretability through these case studies. We believe that this work may help study cell reprogramming, drug selection, and repurposing for improving therapeutic efficacy.