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

The human proteome contains approximately 1.8 million serine/threonine/tyrosine residues, with only a small fraction having experimental validation of phosphorylation. A recent study presented KinoPlex, a computational framework that integrates predicted protein structures and kinase recognition motifs to assign phosphorylation potential and kinase specificity to all serine/threonine/tyronine residues in the human proteome.

The authors used ~20,000 AlphaFold models and positive-unlabeled transfer learning to identify ~567,000 residues as structurally phospho-competent. They then intersected these with kinase position-specific scoring matrices to quantify motif specificity, yielding ~250,000 high-confidence candidates with sequence recognition potential and optimal structural presentation.

The study uncovered fundamental organizing principles guiding kinase substrate recognition and dynamics of phosphorylation, including a phenomenon called sequence–structure selective coupling, whereby kinases achieve specificity through structural scarcity or promiscuity rather than by motif discrimination alone. This finding has implications for understanding the complex interactions between kinases and their substrates in the human proteome.

While KinoPlex predictions were validated by deep phosphoproteomics in K562 cells, further research is needed to fully explore the potential of this computational framework and its applications in kinase research.

The study highlights the importance of integrating structural and functional information to better understand the complex mechanisms underlying kinase specificity and phosphorylation in the human proteome.