MoaNet: Learning Protein–Ligand Mechanism of Action from Transcriptional Profiles and Sequence Information

Abstract

Understanding the functional interactions between molecules and their targets is critical in drug discovery. While current computational approaches achieve high accuracy in predicting drug–target interactions, they often fail to capture the functional consequences of these interactions, such as agonistic or antagonistic effects. Here, we present an end-to-end framework for predicting protein–ligand mechanism of action with transcriptional signature and molecular structural information (MoaNet). Computational evaluations demonstrate that MoaNet outperforms several representative molecular representation architectures adapted for MoA classification on our benchmark data set under scaffold-split evaluation. In a targeted experimental screen for estrogen receptor alpha (ERα) modulators, MoaNet successfully identified both ERα agonists and antagonists, highlighting its practical utility in functional drug discovery.