IgPose: a generative data-augmented pipeline for robust immunoglobulin-antigen binding prediction.

Area of Science:

• Computational biology
• Structural biology
• Immunoinformatics

Background:

• Predicting immunoglobulin-antigen (Ig-Ag) binding is challenging due to limited experimental data and inaccurate de novo structure prediction.
• Accurate prediction of Ig-Ag complexes is crucial for antibody discovery and engineering.

Purpose of the Study:

• To develop a generalizable framework, IgPose, for accurate Ig-Ag pose identification and scoring.
• To address data scarcity in Ig-Ag complex prediction through a novel generative data-augmentation pipeline.

Main Methods:

• Constructed the Structural Immunoglobulin Decoy Database (SIDD) with synthetic decoys to mitigate data scarcity.
• Integrated equivariant graph neural networks, ESM-2 embeddings, and gated recurrent units to capture geometric and evolutionary features.
• Implemented interface-focused k-hop sampling and biologically guided pooling for enhanced generalization.

Main Results:

• IgPose framework comprises IgPoseClassifier for pose discrimination and IgPoseScore for DockQ score estimation.
• Achieved robust performance on internal test sets and the CASP-16 benchmark, outperforming existing baselines.
• Demonstrated accurate pose filtering and ranking capabilities for high-throughput antibody discovery.

Conclusions:

• IgPose provides a versatile computational tool for advancing antibody discovery pipelines.
• The framework's generative approach and integrated deep learning models enhance Ig-Ag binding prediction accuracy.
• IgPose facilitates efficient identification and scoring of potential antibody candidates.