This repo contains code for USP-ddG: A Unified Structural Paradigm with Data Efficacy and Mixture-of-Experts for Predicting Mutational Effects on Protein-Protein Interactions by Guanglei Yu, Xuehua Bi, Qichang Zhao, and Jianxin Wang*.

In this work, we present USP-ddG, a unified structural paradigm for $\Delta \Delta G$ prediction. USP-ddG integrates a dual-channel architecture with three complementary components: (i) inverse folding-based log-odds ratio, (ii) empirical energy terms from FoldX, and (iii) a geometric encoder with Gaussian noise to capture relaxed conformations. To enhance representation power, we introduce a framework that integrates feed-forward network (FFN) and Mixture-of-Experts (MoE) to model domain-invariant and -specific features, respectively. We further propose CATH-guided Folding Ordering (CFO), a data efficacy strategy that organizes samples to mitigate catastrophic forgetting and data distribution bias. USP-ddG consistently outperforms existing state-of-the-art (SoTA) methods on the SKEMPI v2.0 benchmark, including the challenging hold-out CATH test set. It achieves superior accuracy on both single- and multi-point mutations and demonstrates strong performance in antibody affinity optimization against H1N1 and HER2, and in predicting the impact of SARS-CoV-2 variants on hACE2 binding.

Overview of our USP-ddG architecture is shown below.

To install USP-ddG, first clone this repository

git clone https://github.com/ak422/USP-ddG.git

To install the necessary python dependencies, then use the file env.yml to create the environment

conda env create -f env.yml
conda activate USP-ddG

or

mamba env create -f env.yml
conda activate USP-ddG

The default PyTorch version is 2.1.2, python version is 3.10, and cudatoolkit version is 11.8. They can be changed in env.yml.

To generate mutant structures and prepare the processed datasets for SKEMPI v2.0, CR6261, HER2, and S285, respectively, execute the following commands from the code directory:

python skempi_parallel.py --reset --subset skempi_v2
python skempi_parallel.py --reset --subset CR6261
python skempi_parallel.py --reset --subset HER2
python skempi_parallel.py --reset --subset S285

1. The 5 trained model weights of USP-ddG for the hold-out CATH test set are available at: USP-ddG-sampler;
2. The trained weights for case study is located in: case_study

1. Download the trained weights from case_study and place case_study.pt in the ./trained_models/ directory.
2. Download the Demo dataset from Demo_cache and place Demo_cache in the ./data/SKEMPI2/ directory.
3. Execute the following command:

python case_study.py ./configs/inference/demo.yml --device cuda:0

python test.py ./configs/train/CATH-MoE.yml  --device cuda:0

python case_study.py ./configs/inference/case_study_CR6261.yml --device cuda:0

python case_study.py ./configs/inference/case_study_HER2.yml --device cuda:0

python case_study.py ./configs/inference/case_study_S285.yml --device cuda:0

python train.py ./configs/train/CATH-MoE.yml --device cuda:0

python zero_shot.py ./configs/inference/zero_shot.yml  --device cuda:0

If you use this code or the models in your research, please cite the following paper:

@article{yu2025usp,
  title={USP-ddG: A Unified Structural Paradigm with Data Efficacy and Mixture-of-Experts for Predicting Mutational Effects on Protein-Protein Interactions},
  author={Guanglei Yu and Xuehua Bi and Qichang Zhao and Jianxin Wang},
  journal={bioRxiv},
  pages={2025--11},
  year={2025},
}