对超过90,000例癌症患者的临床测序队列中正构激酶错义突变的分析

Being among the most well-studied family in the human proteome, the approximately 500 kinases that constitute the human kinome possess ample sequence, structural, and functional annotations within both pan-proteome and kinase-specific resources. Analyzing mutations in homologous residues, particularly in the orthosteric binding pocket, can provide insights into recurrent mechanisms of pathogenicity and drug resistance. However, harmonizing these resources poses a non-trivial task: identifier disambiguation, mapping between disparate sequence representations, and even concordance between resources as to what constitutes a kinase enzyme all pose significant challenges. Here, we describe the creation of a Python package to reliably and reproducibly curate and harmonize kinase sequences. We use this toolkit to interrogate missense mutations within the orthosteric kinase binding site from Memorial Sloan Kettering Cancer Center's (MSKCC) clinical sequencing cohort. We mapped the 85-residue binding site from the Kinase-Ligand Interaction Fingerprints and Structures (KLIFS) database to the canonical UniProt sequence for a superset of kinases from various resources. Among the 126,930 samples from the 93,249 patients in MSKCC's clinical sequencing cohort as of mid-November 2025, we identified 99 kinase genes assessed for somatic mutations. Within the 87 kinases that mapped to the canonical UniProt sequence, we detected 6,198 unique single amino acid substitutions in the ATP-binding pocket in at least one sample in 11,071 (11.9%) of patients. KLIFS regions with significantly higher percentages of kinases containing one or more such alterations included the gatekeeper (median: 65.9%), linker (65.3%), catalytic loop (61.8%), activation loop/xDFG motif (61.2%), and glycine-rich loop (56.5%), while beta-sheet regions (47.1%) had significantly lower percentages. Kinase families that possessed a significantly higher percentages of KLIFS residues containing one or more such alterations included tyrosine kinase (61.9%) and STE (59.5%), while tyrosine kinase-like (54.8%), AGC (51.8%), calcium/calmodulin-dependent protein kinase (51.2%), and atypical kinases (44.0%) had significantly lower percentages. Since our toolkit harmonizes sequences across the human kinome, it provides a multiple sequence alignment to evaluate species-level protein family conservation. We identified highly conserved, frequently mutated residues in the glycine residues of the glycine-rich loop, the arginine of the HRD motif, and the aspartic acid of the xDFG motif. Our analysis provides a unique pan-cancer assessment of missense mutations across the human kinome, suggesting shared functional mechanisms of pathogenicity for further characterization. Our tools can be used to interrogate missense kinase mutations in any cBioPortal cohort in a similar manner.