ProtMutMap: Predicting Protein Binding Free Energy Changes Using Multiple-Mutation Networks
- 1Department of Computer Science, School of Computing
- 2Institute of Science Tokyo
- 3AI-Science Nexus Center, Institute of New Industry Incubation
Abstract
Predicting changes in binding free energy ΔΔ G caused by multiple mutations at protein−protein interfaces requires accounting for nonadditive interactions between mutations, which an additive approximation based on single mutations cannot capture, and for the dependence on the computational pathway in stepwise approaches that introduce mutations sequentially. Here, we propose ProtMutMap, a method that integrates information from multiple pathways in a network whose nodes represent target variants and variants containing subsets of their mutations, and whose edges represent free energy perturbation (FEP) calculations for residue substitutions. ProtMutMap simultaneously estimates the binding ΔΔ G of each variant from all edge values using Huber loss. We compared ProtMutMap with the additive approximation and the stepwise method at 30 evaluation points, including intermediate variants, using a multiple-mutation dataset comprising 7 protein complex systems derived from SKEMPI2. ProtMutMap achieved a root mean squared error (RMSE) of 1.13 kcal/mol, lower than that of either comparison method. ProtMutMap also yielded lower prediction errors than Rosetta flex ddG. These comparisons with experimental values demonstrate that integrating multiple pathways through variants containing subsets of the target mutations improves binding ΔΔ G predictions for protein variants with multiple mutations.
Related
- A lipoprotein-biomimetic nanoplatform for oncogenic KRAS mutation-specific lysosome-targeting protein degradation.
- RipplePLM: Structural and Property Decoupling for Protein Mutation Effect Generation
- An inverse-protein-folding pre-trained model for predicting the mutational effects in protein-protein interactions
- Towards Accurate Prediction of Mutation-Induced Changes in Protein Structure
- Computational Investigation of S781F Mutation-Induced Remodeling of the Staphylococcal Nuclease Domain-Containing Protein 1–Argonaute 2 Interface
BibTeX
@misc{furui2026protmutmap,
title = {{ProtMutMap: Predicting Protein Binding Free Energy Changes Using Multiple-Mutation Networks}},
author = {Kairi Furui and Masahito Ohue},
howpublished = {ChemRxiv},
year = {2026},
month = sep,
doi = {10.26434/chemrxiv.15009337/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009337/v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - ProtMutMap: Predicting Protein Binding Free Energy Changes Using Multiple-Mutation Networks
AU - Kairi Furui
AU - Masahito Ohue
PY - 2026
DA - 2026/09/23
PB - ChemRxiv
DO - 10.26434/chemrxiv.15009337/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009337/v1
AB - Predicting changes in binding free energy ΔΔ G caused by multiple mutations at protein−protein interfaces requires accounting for nonadditive interactions between mutations, which an additive approximation based on single mutations cannot capture, and for the dependence on the computational pathway in stepwise approaches that introduce mutations sequentially. Here, we propose ProtMutMap, a method that integrates information from multiple pathways in a network whose nodes represent target variants and variants containing subsets of their mutations, and whose edges represent free energy perturbation (FEP) calculations for residue substitutions. ProtMutMap simultaneously estimates the binding ΔΔ G of each variant from all edge values using Huber loss. We compared ProtMutMap with the additive approximation and the stepwise method at 30 evaluation points, including intermediate variants, using a multiple-mutation dataset comprising 7 protein complex systems derived from SKEMPI2. ProtMutMap achieved a root mean squared error (RMSE) of 1.13 kcal/mol, lower than that of either comparison method. ProtMutMap also yielded lower prediction errors than Rosetta flex ddG. These comparisons with experimental values demonstrate that integrating multiple pathways through variants containing subsets of the target mutations improves binding ΔΔ G predictions for protein variants with multiple mutations.
ER -