Vasor

Vasor predicts the effects of missense variants on the multidrug resistance protein 3 (MDR3), classifying single amino acid substitutions as benign or pathogenic for evaluation of MDR3-associated liver disease.

Key Features:

• Machine learning classification: Classifies MDR3 missense variants into benign or pathogenic categories using a supervised machine learning approach.
• Integrated predictors: Incorporates outputs from Evolutionary Models of Variant Effects (EVE), EVmutation, PolyPhen-2, I-Mutant2.0, MUpro, MAESTRO, and PON-P2 as input features.
• Variant-level features: Utilizes variant properties including half-sphere exposure and posttranslational modification sites to inform predictions.
• Training data: Developed using the largest MDR3-specific variant dataset reported in the source description.
• Sequence-space coverage: Provides predictions covering the entire MDR3 sequence space for all possible single-site amino acid substitutions.
• Performance metrics: Achieved an F1-score of 0.90 and a Matthew's correlation coefficient (MCC) of 0.80 on an external test set.
• Comparative performance: Demonstrated superior performance compared to integrated general predictors and the external tool MutPred2.

Scientific Applications:

• Variant interpretation in liver disease: Assess pathogenicity of MDR3 missense variants associated with progressive familial intrahepatic cholestasis, intrahepatic cholestasis of pregnancy, and familial gallstone disease.
• Prioritization for experimental validation: Prioritize MDR3 variants for functional testing based on predicted benign versus pathogenic classification.
• Systematic in silico mutagenesis: Enable comprehensive in silico evaluation of all single-site amino acid substitutions across MDR3.
• Benchmarking predictive methods: Serve as a reference for comparing MDR3-specific predictive performance against general predictors and MutPred2.

Methodology:

Uses a machine learning classifier trained on an MDR3-specific variant dataset that integrates outputs from EVE, EVmutation, PolyPhen-2, I-Mutant2.0, MUpro, MAESTRO, and PON-P2 and includes features such as half-sphere exposure and posttranslational modification sites; performance was evaluated on an external test set reporting F1-score and MCC and compared to MutPred2.