A benchmark for evaluation of structure-based online tools for antibody-antigen binding affinity

Area of Science:

• Immunology
• Computational Biology
• Structural Biology

Background:

• Predicting binding affinity changes from missense mutations aids understanding of antigen-antibody interactions.
• Numerous structure-based computational tools exist, but selecting appropriate software for specific research, particularly concerning SARS-CoV-2 spike protein and antibodies, remains challenging.
• Benchmarking these tools with mutation-diverse datasets is critical for SARS-CoV-2 research.

Purpose of the Study:

• To benchmark the performance of seven structure-based online computational tools for predicting binding affinity changes in SARS-CoV-2 spike protein-antibody complexes.
• To provide a comprehensive evaluation using a curated dataset of 1216 variants from 22 complexes.
• To guide the selection of suitable tools for research on antigen-antibody interactions and therapeutic antibody design.

Main Methods:

• Collected and curated datasets of binding affinity changes for 1216 variants across 22 SARS-CoV-2 spike protein and monoclonal antibody complexes.
• Applied seven different binding affinity prediction tools to evaluate their performance.
• Assessed performance using Pearson correlation for continuous affinity changes and accuracy for classification tasks (predicting affinity increase/decrease).

Main Results:

• Pearson correlations between predicted and measured binding affinity changes ranged from -0.158 to 0.657.
• Classification accuracy for predicting affinity changes varied from 0.444 to 0.834.
• Tools performed better on single mutations, especially at epitope sites, but showed poor performance for extremely decreasing affinities. Tool performance was insensitive to the experimental techniques used for complex structures.

Conclusions:

• The developed datasets and benchmarking results provide valuable insights into the capabilities of current structure-based prediction tools for SARS-CoV-2 antigen-antibody interactions.
• These findings will aid researchers in selecting appropriate computational tools for analyzing mutation effects on binding affinity.
• The evaluation supports the enhancement of computational design strategies for therapeutic monoclonal antibodies targeting SARS-CoV-2.