IDEA #31L4AA 23A0012 TransProt: High Throughput Predictive Modeling of Proteomics Profiles Perturbed by Novel Chemicals

For the first time, we develop an end-to-end deep learning framework TransProt to predict proteomics profiles and cellular phenotypes perturbed by novel unseen chemicals. Our comprehensive evaluations in anti-cancer drug sensitivity and drug adverse reaction predictions suggest that the accuracy of TransProt predictions is comparable to that of experimental data. Thus, TranProt could a useful tool for proteomics data imputation and phenotype-based drug discovery.