EquiRank: Improved protein-protein interface quality estimation using protein language-model-informed equivariant...

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Artificial Intelligence in Biology

Background:

• Accurate quality estimation of protein complex structural models is crucial for model selection and protein-protein docking.
• Existing methods struggle to fully leverage advances in symmetry-aware deep learning and protein language models (pLMs) for interface quality estimation.

Purpose of the Study:

• To develop an improved method, EquiRank, for estimating the quality of protein-protein interaction interfaces.
• To integrate the strengths of E(3) equivariant graph neural networks (EGNNs) and pLM embeddings for enhanced prediction accuracy.

Main Methods:

• Utilized a symmetry-aware E(3) equivariant deep graph neural network (EGNN).
• Integrated embeddings from the pretrained ESM-2 protein language model.
• Represented protein-protein interfaces using a graph-based approach of interacting residue pairs with diverse features.

Main Results:

• EquiRank demonstrated superior ranking performance on diverse datasets compared to state-of-the-art methods like VoroIF_GNN and AlphaFold-Multimer's self-assessment module.
• Ablation studies confirmed the significant contributions of both pLMs and the equivariant nature of EGNN to performance improvements.
• The method achieved better results across various performance evaluation metrics.

Conclusions:

• EquiRank offers a significant advancement in protein-protein interface quality estimation.
• The integration of advanced deep learning architectures and pLMs provides a powerful framework for evaluating protein complex models.
• The developed method is freely available, facilitating further research and application in structural bioinformatics.