An AI-enabled structural atlas decodes kinase specificity across the human proteome

A new computational framework called KinoPlex has been developed to map phosphorylation potential and kinase specificity across the entire human proteome. KinoPlex utilizes AlphaFold models, positive-unlabeled transfer learning, and kinase position-specific scoring matrices to identify 567,000 structurally phospho-competent residues and 250,000 high-confidence candidates with the ability to be recognized by kinases.

The atlas reveals fundamental principles guiding kinase substrate recognition and the dynamics of phosphorylation, including a phenomenon called sequence–structure selective coupling. This principle explains how kinases achieve specificity through structural scarcity or accessibility of their preferred motifs. Deep phosphoproteomics in K562 cells confirmed the accuracy of KinoPlex predictions and its ability to accurately predict kinase enrichment.

The entire structural atlas and proteome-wide PSSM percentiles are available at doi.org/10.5281/zenodo.20633591.