Key points are not available for this paper at this time.
ABSTRACT Drug molecular interactions, including drug–drug interactions (DDIs) and drug–target interactions (DTIs), are critical for drug discovery and clinical safety, increasingly propelled by artificial intelligence (AI) technologies. Although previously treated as separate domains, DDIs and DTIs are highly interconnected in terms of biological mechanisms and model design. To foster their co‐evolution, this review provides a comprehensive landscape of drug molecular interaction modeling by first summarizing the advanced AI technologies across various prediction tasks in both domains. Then, the parallel development paths are examined in core architecture, feature engineering, and model learning paradigms, highlighting the convergence in patterns of feature engineering and trends of model design. Furthermore, the key challenges, such as insufficient generalizability and shortcut learning, are identified and evaluated through quantitative experiments, and future directions are proposed for building unified models to leverage AI in accelerating drug discovery and therapeutics design.
Sun et al. (Fri,) studied this question.