Discriminating activating, deactivating and resistance variants in protein kinases

Area of Science:

• Genomics
• Computational Biology
• Biochemistry

Background:

• Distinguishing gain-of-function (GOF) from loss-of-function (LOF) genetic variants in protein kinases is crucial for clinical genetics and personalized medicine.
• GOF variants, unlike LOF variants, are often treatable with inhibitors, highlighting the need for precise variant classification.
• Current computational tools primarily predict variant pathogenicity (damaging vs. benign) without providing specific functional insights.

Purpose of the Study:

• To develop and validate a data-driven computational approach for differentiating kinase variants into activating (GOF), deactivating (LOF), and resistance types.
• To provide functional insights into genetic variants beyond simple pathogenicity prediction.
• To facilitate the identification of therapeutically targetable kinase variants.

Main Methods:

• Curated a dataset of 2505 variants across 441 kinases, categorized as activating, deactivating, resistance, or neutral.
• Utilized sequence, evolutionary, and structural features to train machine learning models for variant classification.
• Achieved high predictive performance with a mean Area Under the Curve (AUC) of 0.941.
• Experimentally validated predictions using cell-based assays (over-expression, gene expression, biochemical tests).

Main Results:

• Observed significant enrichment of activating mutations in cancer genomes and deactivating variants in hereditary diseases.
• Experimentally validated predicted activating variants in cancer samples, observing increased kinase activity.
• Demonstrated experimentally that a predicted variant in MAP2K3 caused reduced mitochondrial function, contrary to the effect of deletions.
• Developed an online application for analyzing kinase-domain variants and exploring known nearby variants.

Conclusions:

• The developed predictors, combined with rapid experimental validation, offer a feasible strategy for timely identification of activating kinase variants.
• This approach can significantly aid in making clinical decisions for patients with kinase-related diseases.
• The method advances the potential for personalized medicine by enabling rapid functional characterization of genetic variants.