Protein phosphorylation is a reversible post-translational modification driving intracellular signalling networks and modifying protein function. During phosphorylation, a kinase catalyses the transfer of a phosphate group onto a target amino acid (a phosphosite). Despite its importance, kinase-phosphosite mapping remains challenging with less than 5% of known human phosphosites being mapped to a cognate kinase. A more tractable way to build phosphorylation networks is to closely investigate relationships between sequence motifs for all known phosphosites in a proteome. This can reveal closely related phosphosites that are likely to arise from the same kinase, and thus a range of regulatory relationships.
The analysis of amino acid sequences surrounding 11,224 phosphosites from the Saccharomyces cerevisiae proteome revealed 721 phosphosites with identical or near-identical sequence motifs on the same or different proteins. These sites will most likely be phosphorylated by the same kinase. Further, 78% of these motifs were strikingly unique, being only found at these specific sites within the S. cerevisiae proteome. Subsequent analyses of motifs using a protein-trained transformer model, the BLOSUM62 substitution matrix, an RBF kernel function and different clustering techniques identified non-identical motifs likely to be targeted by specific kinases, or by the same kinase family. This shows the potential to extract kinase motifs at different levels of kinase specificity. A draft AI-driven phosphosite network will be presented, predicting which proteins might be co-regulated by the same kinase, and helping to identify potential kinases for unmapped phosphosites.