CrossDome

CrossDome predicts off-target T-cell cross-reactivity risk for immunotherapies by comparing peptide sequences and TCR-contact features against immunopeptidomics databases to assess potential molecular mimicry and toxicity.

Key Features:

• Multi-Omics Integration: Integrates multi-omics data to generate comprehensive off-target toxicity risk assessments.
• Peptide-Centered Prediction: Evaluates cross-reactivity likelihood based on peptide sequences and ranks potential off-targets among large candidate pools.
• TCR-Centered Prediction: Applies a contact map (CM) penalty system focused on T-cell receptor (TCR) hotspots to refine peptide risk rankings.
• Proof-of-Principle Validation: Evaluated 16 well-known cross-reactivity cases and identified a high-risk TITIN-derived peptide at >99th percentile among ~36,000 candidates.
• Monte Carlo Simulation: Employs Monte Carlo simulations to assess relatedness scores across >5 million putative peptide pairs and derive cut-off p-values for risk.
• Enrichment Analysis: Quantifies enrichment of validated cross-reactive peptides, with the TCR-centered approach achieving up to 82% accuracy among top-ranked candidates.
• Functional Characterization: Integrates expression data, HLA binding affinity, and immunogenicity predictions to contextualize off-target assessments.

Scientific Applications:

• Risk Identification: Identifying and mitigating risks associated with T-cell cross-reactivity in immunotherapies.
• Therapy Development: Informing the development and prioritization of targeted and safer cancer immunotherapies by ranking off-target peptide risks.
• Antigen Discovery Support: Supporting antigen discovery and selection within immunopeptidomics-driven pipelines.

Methodology:

Analyzes peptide sequences and TCR interactions, applies a contact map (CM) penalty system based on TCR hotspots, runs Monte Carlo simulations to compute relatedness scores across >5 million peptide pairs and establish p-value cut-offs, performs enrichment analysis and ranking among large candidate sets (~36,000), and integrates expression data, HLA binding affinity, and immunogenicity predictions; validated on 16 cross-reactivity cases including a TITIN-derived peptide example.