Precision oncology in the age of AI: lessons from AI-driven drug discovery and clinical translation

Drug discovery has been constrained by extended timelines and high costs, as the cumulative requirements of preclinical validation, multi-phase clinical trials, and regulatory approval have been imposed. Recently, computational modeling has been explored as a supportive approach to accelerate the identification and refinement of therapeutic candidates. Proof-of-concept was provided in a phase 2a trial of a de novo-designed TNIK inhibitor in idiopathic pulmonary fibrosis, in which safety, tolerability, and pharmacodynamic target engagement were demonstrated, with a trend toward reduced functional decline. This study showed that AI-derived molecules can advance into human testing, but broader validation, mechanistic understanding, and regulatory alignment remain essential. In oncology, where tumor heterogeneity, clonal evolution, and therapeutic resistance continue to constrain durable clinical benefit, there is an increasing need for adaptive and data-informed drug discovery strategies. This Perspective reviews recent progress and limitations in AI-driven drug discovery and early clinical translation. It emphasizes how the clinical evaluation of an AI-generated TNIK inhibitor serves as an early translational reference and outlines practical strategies for integrating multi-omics data, federated model validation, and adaptive trial design to advance precision oncology-oriented therapeutics.