Predicting transcriptional responses to novel chemical perturbations using deep generative model

Abstract Understanding transcriptional responses to chemical perturbations is central for drug discovery, but exhaustive experimental high-throughput screening of disease and compound combinations is unfeasible. To overcome this limitation, here we present a perturbation-conditioned deep generative model named PRnet for predicting transcriptional responses to novel chemical perturbations that were never experimentally perturbed at bulk and single-cell levels. Evaluation indicated that PRnet outperformed alternative methods in predicting responses across novel compounds, pathways, and cell lines. PRnet enables gene-level response interpretation and novel compounds screening for diseases based on gene signatures. PRnet further identified and experimentally tested novel compounds candidates against small cell lung cancer and colorectal cancer. Lastly, PRnet generated a large-scale integration atlas of perturbation profiles, covering 88 cell lines and 52 tissues perturbed by various screening compound libraries. PRnet provided a robust and scalable candidate recommendation workflow and has successfully recommended drug candidates for 233 different diseases based on the atlas. Overall, PRnet is an effective and valuable tool for cell- and gene-based therapeutics screening.