DBTpred

DBTpred predicts peptide–MHC class I binding affinities using a differential boundary tree–based model to detect single-residue mutations that alter peptide–MHC interactions for neoantigen identification in personalized cancer immunotherapy.

Key Features:

• Differential Boundary Tree-Based Model: Employs differential boundary trees to predict MHC class I peptide binding affinities and to identify single-residue mutations that significantly change binding.
• High Accuracy and Low False Positives: Demonstrates superior performance relative to state-of-the-art deep learning models with reduced false-positive rates in MHC–peptide interaction prediction.
• Parallel Training Algorithm: Implements a parallel training algorithm to accelerate training and inference and to enable scalable application to large datasets.
• Interpretability via Statistical Analysis: Analyzes statistical properties of differential boundary trees and prediction paths to provide interpretable clues about critical residue mutations affecting peptide–MHC interactions.

Scientific Applications:

• Neoantigen Identification: Supports identification of neoantigens by precisely predicting changes in peptide–MHC binding affinity caused by somatic mutations.
• Immunological Research: Facilitates investigation of the molecular basis of immune recognition by providing detailed insights into peptide–MHC interactions.

Methodology:

Uses a differential boundary tree–based model and analysis of statistical properties of boundary trees and prediction paths; incorporates a parallel training algorithm to accelerate training and inference; implemented in Python.