High-throughput mapping of kinase-substrate interactions with protein language models and phospho-PCA

A lab-in-the-loop framework combining KINBERT with phospho-PCA identifies novel high-confidence kinase-substrate interactions for 23 kinase constructs spanning 21 kinases, including dark kinases ADCK1 and CaMKK1 (KKCC1).

• Why it matters: Many measured human phosphosites lack an assigned upstream kinase, and roughly a third of the human kinome remains functionally undercharacterized, limiting insight into disease signaling rewiring.
• What they did: The approach uses KINBERT to learn interaction rules from catalytic domains and substrate peptides, then applies phospho-PCA pooled screening against multiple kinases, analyzed with Bayesian hierarchical modeling.
• The result: Trained on an expansive literature-curated dataset, KINBERT generalizes strongly and outperforms existing models on standard and out-of-distribution benchmarks, with further gains when incorporating phospho-PCA data.