DECODE

DECODE extracts biochemical rules governing T cell receptor (TCR) binding from black-box machine learning models to elucidate TCR-epitope interaction mechanisms.

Key Features:

• Extraction of Binding Rules: Extracts and interprets binding rules from black-box machine learning models that predict TCR-epitope interactions.
• Customizable Computational Pipeline: Provides a configurable computational pipeline for tailoring analysis parameters and workflows for interpreting TCR-binding models.
• Analytical and Visualization Tools: Implements analytical methods and generates visualizations to assess and present computed binding rules and motif quality.
• Application to Existing Models: Demonstrated application to the TITAN TCR-binding prediction model to identify sequence motifs critical for TCR binding.

Scientific Applications:

• Immunotherapy safety assessment: Investigates cross-reactive events caused by off-target TCR binding to inform safety evaluations of T cell–based therapies.
• Sequence-motif discovery for TCR specificity: Facilitates identification of sequence motifs underlying TCR-epitope recognition to study specificity and cross-reactivity.
• Model interpretation and biological insight: Translates outputs of complex machine-learning predictors into biochemical rules linking computational predictions to mechanistic understanding.

Methodology:

Analyzes outputs of black-box machine-learning TCR–epitope binding predictors and leverages repertoire sequencing data to extract biochemical binding rules, with demonstration on the TITAN model.