Benchmarking the accuracy of structure-based binding affinity predictors on Spike-ACE2 deep mutational interaction...

Area of Science:

• Computational biology
• Structural biology
• Virology

Background:

• Understanding SARS-CoV-2 Spike-ACE2 interactions is crucial for pandemic response.
• Deep mutational scanning studies have generated extensive data on mutations affecting Spike and ACE2 binding.

Purpose of the Study:

• To benchmark the performance of commonly used structure-based binding affinity predictors.
• To evaluate predictors using experimental data from deep mutational scanning studies of Spike-ACE2 interface mutations.

Main Methods:

• Six structure-based binding affinity predictors (FoldX, EvoEF1, MutaBind2, SSIPe, HADDOCK, UEP) were selected and tested.
• Predictors were benchmarked against experimental binding data derived from deep mutational scanning.
• Performance was assessed using correlation coefficients and binary classification accuracy.

Main Results:

• No tested predictor showed a meaningful correlation with experimental binding affinity data.
• FoldX achieved the best correlation (R = -0.51) and binary classification accuracy (64%).
• Simple energetic scoring functions outperformed evolutionary-based methods, and AI approaches showed comparable performance to force field-based techniques.

Conclusions:

• Current structure-based binding affinity predictors require significant improvement for predicting variant-induced binding changes in host-pathogen systems.
• The study provides valuable benchmarking data and visualized mutant models to facilitate future predictor development.