HLA-Clus

HLA-Clus clusters HLA class I alleles based on the three-dimensional structure of their peptide-binding grooves to characterize functional similarity relevant to peptide binding specificity.

Key Features:

• 3D Structure-Based Clustering: Classifies HLA class I alleles into supertypes and subtypes by analyzing the three-dimensional landscape of peptide-binding grooves.
• Structure Distance Metric: Implements a novel structure distance metric that integrates spatial and physicochemical properties to quantify inter-allelic similarity.
• Hierarchical and Nearest-Neighbor Clustering: Supports hierarchical clustering and a nearest-neighbor clustering method for grouping alleles.
• Integration with MHCnuggets: Interfaces with the peptide:HLA affinity predictor MHCnuggets to select optimal allele-specific deep learning models via nearest-neighbor clustering, improving peptide binding predictions for rare alleles.
• Coarse Graining and Transformation: Coarse-grains HLA Class I structural models into clouds of labeled points to facilitate efficient similarity assessment.
• Implementation: Provided as a Python package implementation.
• Improved Functional Correlation and Robustness: Shows enhanced correlation with peptide binding specificity, increased intra-cluster similarity, and robustness compared to existing approaches.

Scientific Applications:

• Peptide Affinity Prediction Development: Characterizes peptide-binding specificities across HLA alleles to support development of more accurate peptide affinity predictors.
• Disease Association Studies: Clarifies allele-specific binding properties to aid identification of associations between HLA types and diseases.
• HLA Matching for Transplantation: Improves HLA matching precision by grouping functionally similar alleles relevant to transplantation outcomes.

Methodology:

The pipeline comprises three stages: coarse-graining structural models into labeled point clouds, computing pairwise similarities using the novel structure distance metric that incorporates spatial and physicochemical features, and clustering alleles using hierarchical or nearest-neighbor algorithms.