InMeRF

InMeRF predicts the pathogenicity of nonsynonymous single-nucleotide variants (nsSNVs) by training individual random forest models for each amino acid substitution.

Key Features:

• Individual Modeling Approach: Trains 150 independent random forest models, one for each possible amino acid substitution.
• Feature Set: Uses 34 rank scores from dbNSFP v4.0a as input feature values for each model.
• Algorithm: Implements Random Forest (RF) classifiers for variant pathogenicity prediction.
• Cross-validation Performance: Ten-fold cross-validation produced a ROC-AUC of 0.941 and a precision-recall AUC of 0.957.
• Comparative Benchmarking: Benchmarked against seven other tools using the same training dataset and three testing datasets, ranking first by ROC-AUC.
• Application to Specific Genes: Applied to nsSNVs in genes associated with congenital myasthenic syndromes and spina bifida (VANGL1), achieving sensitivity 0.942 and specificity 0.848.

Scientific Applications:

• Disease Variant Interpretation: Prioritizes and interprets disease-associated nsSNVs caused by amino acid substitutions.
• Genetic Disorder Analysis: Analyzes nsSNVs in genes linked to congenital myasthenic syndromes and spina bifida (VANGL1).
• Personalized Medicine Support: Provides variant-level pathogenicity scores to inform diagnosis and treatment planning in personalized medicine contexts.

Methodology:

Trains 150 independent RF models (one per amino acid substitution) using 34 rank scores from dbNSFP v4.0a as features; performance was assessed by ten-fold cross-validation and testing on three datasets with comparisons to seven other tools.