Analysis of Kinase-Peptide Interactions: DGMM Latent Constructs and Clustering Towards New Groups and Stand-Out Structures.

Protein kinases are among the largest families of enzymes in eukaryotes and have been intensively studied as important cellular components and medicinal targets. In this study, interface properties for all experimental high-resolution human protein kinase domain-peptide complexes from the protein data bank (PDB) were calculated, and analyzed statistically. The structures encompassed all the six most populous human kinase groups, and the majority of peptides were found binding either solely to the kinase C lobe or both/between lobes. Analyses of the interface properties revealed numerous unexpected trends and correlations, including several unusual residue combinations and surface area features. The interface properties that differed most between human kinase group members or peptide-binding site location on the kinase domain were identified. Factor analysis also successfully separated the structures by human kinase group. The five factors or latent dimensions found are broadly interpretable as the following contrasts involving: number of interface residues vs contacts to interface surface area relationship, interface charge versus polarity, Kinase entity length versus neutral interface surface area relationship, and Peptide entity length versus interface surface area relationship. Deep Gaussian mixtures (DGMM) latent constructs and clustering determined that three traditional Kinase groups 3, 4, and 5, relating to CAMK, AGC, and CMGC, were each split into unique pairs of clusters derived by DGMM clusters and mapping onto 2D latent space. Based on latent space mapping and molecular visualisation various stand-out kinase peptide structures were identified as follows; 3ALO, 1UKI, 4NM7a, 4UBX, 5N4R, 5N52, 3FY2, 3CQW, 5BX7, 4G1W, and 4H3P-adding insights into structural differences with the potential to contribute to drug discovery innovation. These findings could help guide future experimental kinase-peptide interaction designs and studies, the engineering of kinase proteins, or peptide design in general. The results are particularly relevant in the absence of prior specific structural kinase information.