Meta-SNP

Meta-SNP predicts the pathogenicity of non-synonymous single nucleotide variants (nsSNVs) by integrating outputs from PANTHER, PhD-SNP, SIFT, and SNAP to improve discrimination between disease-associated and benign amino acid substitutions.

Key Features:

• Integrated predictors: Integrates outputs from PANTHER, PhD-SNP, SIFT, and SNAP into a unified prediction framework.
• Machine learning-based ensemble: Combines constituent predictions using a machine learning-based approach to classify nsSNVs as disease-associated or benign.
• Benchmark dataset: Validated on 35,766 disease-annotated mutations from 8,667 proteins sourced from the SwissVar database.
• Overall performance: Achieved 79% accuracy and a Matthews correlation coefficient (MCC) of 0.59.
• Constituent method performance: The four methods individually achieved accuracies of 64%–76% and MCC values between 0.38 and 0.53.
• Improvement over best individual: Improved accuracy by approximately 3% and MCC by ~0.05 compared to the best-performing individual method.
• Discordant-prediction performance: On nsSNVs where constituent predictors disagreed, Meta-SNP showed an 8% higher accuracy than any single predictor.
• Consensus-prediction performance: For nsSNV subsets where all predictors agreed (46% of the dataset), Meta-SNP reached 87% accuracy and an MCC of 0.73.
• Biological scope: Targets non-synonymous SNVs that produce amino acid substitutions and addresses interpretation at the scale of variant catalogs such as dbSNP (over 50 million validated SNVs).

Scientific Applications:

• Prioritization of disease-associated nsSNVs: Prioritizes candidate nsSNVs for experimental validation based on predicted pathogenicity.
• Variant-disease association studies: Supports genomic research into variant-disease associations by providing integrated pathogenicity scores for nsSNVs.

Methodology:

Combines outputs from PANTHER, PhD-SNP, SIFT, and SNAP using a machine learning-based classifier and validates predictions against 35,766 disease-annotated mutations from 8,667 proteins in SwissVar.