RingKin: portraying the vast macrocyclic chemical universe surrounding kinase drugs.

Cyclization strategies have emerged as a compelling approach to overcome persistent challenges in kinase drug development, notably poor selectivity and unfavorable pharmacological properties. Existing studies remain limited to individual kinases or retrospective analyses, while systematic exploration of the potentially vast and underexplored macrocyclic space for kinase modulation is still lacking. By leveraging artificial intelligence-based methods, we pioneered the creation of RingKin, an immense chemical universe encompassing 72.27 million macrocycles generated from 495 approved or clinical-stage kinase drugs. In addition to diverse macrocyclic scaffolds, RingKin offers approximately 1.8 billion model-estimated property annotations by 15 benchmarked deep learning or machine learning-based models, supporting the multidimensional characterization of generated macrocycles, including kinase selectivity profiles, physicochemical and ADMET features, and target associations. Systematic analyses suggest that these macrocycles have the potential to mitigate several major limitations associated with current kinase drugs. Our work portrays the prospective macrocyclic chemical landscape surrounding existing kinase drugs, establishing RingKin as a hypothesis-generating resource for macrocycle exploration and drug discovery. From macrocyclic derivatives of the approved pan-FGFR inhibitor erdafitinib, three preliminary FGFR2-preferring hit compounds were identified, demonstrating the utility of RingKin as an open-access resource for kinase-targeted drug design.