IDEA #0LUPE1 CODE-AE: Context-aware deconfounding autoencoder

Accurate and robust prediction of patient-specific responses to drug treatments is critical for developing personalized medicine. However, it is difficult to obtain a sufficient amount of patient data for training a generalized machine learning model. Although cell line data are abundant, few existing computational methods can reliably predict individual patient clinical responses to new drugs from in vitro screens due to data distribution shift and confounding factors. We develop a novel Context-aware Deconfounding Autoencoder (CODE-AE) that can extract common biological signals masked by context-specific patterns and confounding factors. Extensive studies demonstrate that CODE-AE significantly improves accuracy and robustness over conventional methods in both predicting patient-specific ex vivo and in vivo drug responses purely from cell line screens and disentangle intrinsic biological signal from confounding factors. CODE-AE can be immediately applied to personalized medicine, precision medicine, and drug discovery