KinMutRF

KinMutRF classifies pathogenic and neutral sequence variants in the human protein kinase superfamily using a random forest classifier to link sequence variation to mechanistic molecular traits.

Key Features:

• Random Forest Classifier: Employs a random forest ensemble comprising 26 decision trees for prediction.
• Comprehensive Feature Set: Integrates gene-level features (KinBase group membership, Gene Ontology terms), PFAM domain-level features, and residue-level features (amino acid types, biochemical property changes, and functional annotations from UniProt, Phospho.ELM, and FireDB).
• Performance Metrics: Reports accuracy 0.88, precision 0.82, recall 0.75, F-score 0.78, and Matthews correlation coefficient (MCC) 0.68 from training and cross-validation on 3,689 UniProt-annotated human kinase variants.
• Benchmarking: Predictions were compared to SIFT, Polyphen-2, MutationAssessor, MutationTaster, LRT, CADD, FATHMM, and VEST.
• Kinase-Specific Feature Importance: Identifies kinase-specific features as the largest source of information gain, supporting the use of family-specific classifiers.
• Unclassified Variant Predictions: Provides predictions for 848 previously unclassified protein kinase variants in UniProt.

Scientific Applications:

• Pathogenic Variant Identification: Classifies kinase variants as pathogenic or neutral to support variant interpretation in research datasets.
• Research and Development: Enables investigation of molecular mechanisms and signal-processing alterations associated with kinase variants.
• Clinical Implications: Supports discrimination of disease-associated kinase variants relevant to targeted therapies and precision medicine research.

Methodology:

Trained a random forest (26 trees) on a curated set of 3,689 human kinase variants annotated as neutral or pathogenic in UniProt (excluding unclassified variants), evaluated by cross-validation, validated against independent kinase-specific mutation sets Kin-Driver (643 variants) and Pon-BTK (1,495 variants), and used to generate predictions for 848 previously unclassified UniProt kinase variants.