国际标准期刊号: 2169-0138
David G. Covell
A bioinformatic strategy was proposed for linking ligand Protein Data Bank (PDB) structural fragments to ChEMBL IC50 bioactivity of Protein Kinases (PKs). A bootstrap procedure, based on exhaustive enumeration, was used to assemble, and statistically evaluate, sets of fragments that were enriched for ligands that target PKs in separate branches of the kinome tree. Results found that probes comprised of six fragments return 84% correct predictions for branch-selective PKs ligands. Self-organizing maps were used to cluster the enriched six-fragment probes and, separately, the ChEMBL IC50 data, to identify branch-selective fragments and branch-chemoselective ligands. A contingency table, based on the co-occurrence of branch-chemoselective ligands possessing branch-selective fragments, used for Fisher’s exact tests of independence, found an average recovery of 44% for branch-chemoselective ligands. Seven percent (7%) of these cases represent exact structural matches to PDB ligands, inclusive of eight Food and Drug Administration (FDA) approved oncology compounds. Binding site analysis of enriched branch-selective fragments for these FDA ligands found roles for key hydrophobic and non-hydrophobic interactions. Global extension of these results found a subset of 402 branch-chemoselective ligands with enriched branch-selective fragments, but without crystallographic data, as candidates for selectively targeting PKs. These results extend the use of fragment-selective mining of chemical libraries aimed at discovering ligands that target PKs in separate kinome branches.