Deep Learning Modeling of Androgen Receptor Responses to Prostate Cancer Therapies

摘要: Gain-of-function mutations in human androgen receptor (AR) are among the major causes of drug resistance prostate cancer (PCa). Identifying that cause resistant phenotype is critical importance for guiding treatment protocols, as well designing drugs do not elicit adverse responses. However, experimental characterization these time consuming and costly; thus, predictive models needed to anticipate guide discovery process. In this work, we leverage data collected on 68 AR mutants, either observed clinic or described literature, train a deep neural network (DNN) predicts response mutants currently used anti-androgens testosterone. We demonstrate use DNN, with general 2D descriptors, provides more accurate prediction biological outcome (inhibition, activation, no-response, mixed-response) mutant-drug pairs compared other machine learning approaches. Finally, developed approach was make predictions mutant latest inhibitor darolutamide, which were then validated by in-vitro experiments.