MHCVision

MHCVision estimates global and local false discovery rates (FDR) and posterior error probabilities (PEP) for MHC-peptide binding predictions to provide statistical confidence for NetMHCpan-derived IC50 and percentile rank scores.

Key Features:

• Global and local FDR estimation: Estimates both global and local false discovery rates for predicted peptide binding score distributions.
• Posterior error probability (PEP) estimation: Computes PEPs for individual peptide predictions to quantify confidence in each prediction.
• NetMHCpan score integration: Uses predicted binding affinities (IC50) and percentile rank scores from NetMHCpan as input.
• Two-component beta mixture model: Models score distributions with a two-component beta mixture to represent true and false prediction populations.
• Modified expectation maximization: Applies a modified expectation maximization framework to fit the mixture model and separate true versus false score distributions.
• Sensitivity–precision trade-off analysis: Enables evaluation beyond fixed thresholds (e.g., IC50 < 500 nM or percentile rank < 2%) to balance sensitivity and precision.

Scientific Applications:

• Immune response analysis: Provides statistically grounded evaluations of predicted MHC-peptide interactions to support studies of antigen presentation and immune recognition.
• Immunotherapeutic development: Aids selection and prioritization of candidate peptides for vaccine and immunotherapy design by supplying FDR and PEP estimates.
• Personalized immunotherapy: Supports evaluation and prioritization of patient-specific peptide predictions for personalized immunotherapies.

Methodology:

Applies a modified expectation maximization algorithm to a two-component beta mixture model to distinguish true and false score distributions and derive global/local FDR and PEP from NetMHCpan IC50 and percentile rank scores.