Predicting targets and off-targets of small molecule drugs

Target identification for small-molecule ligands is a pivotal step in drug discovery for mechanism deconvolution of phenotypic hits, identifying off-targets, discovering new druggable targets, and designing analogs to improve safety. Early algorithmic approaches for target prediction, spanning ligand-centric chemical fingerprints and target-focused 3D-structural pattern analyses, have established the foundation for building the next generation of improved methods. Increased availability of high-throughput omics data and their multimodal integration with knowledge graphs have added another dimension to the range of computational methods, enabling systems-wide screening for targets and off-targets. In this review, we discuss the current advances, especially the structure-based methods integrated with deep learning frameworks. Together, the methods are generating predictive power for simultaneously uncovering mechanisms, targets, off-targets, and re-purposable drugs.