Advances in Small Molecule Representations and AI-Driven Drug Research: Bridging the Gap Between Theory and Application

AI scientists and cheminformatics experts work to identify effective drug precursors, optimizing costs and accelerating development. To reach this objective, digital molecular representation is of utmost importance. It enables molecules to be machine readable, thereby enhancing the accuracy of various molecule related prediction tasks and facilitating well informed decision making. In this study, we conduct a review of small molecular representations and AI-driven drug discovery related downstream tasks with these representations. First, we collect databases relevant to small molecules. Subsequently, we summarize fundamental molecular representations and the models employed to learn these representations from initial forms that capture patterns and salient features across extensive chemical spaces. Following that, we overview different drug discovery related downstream tasks, such as drug-target interaction prediction, drug-target affinity prediction, drug property prediction, and drug generation, all of which are based on the learned representations. Finally, we accentuate the challenges and opportunities associated with machine learning methods for molecular representation and for enhancing the performance of diverse downstream tasks. The representation of small molecules and AI-based downstream tasks are also crucial in uncovering Traditional Chinese Medicine (TCM) medicinal substances and TCM target discovery.