PIQLE: protein-protein interface quality estimation by deep graph learning of multimeric interaction geometries

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Machine Learning

Background:

• Accurate protein-protein interaction interface modeling is crucial for protein complex structure prediction.
• Current methods often overlook evolutionary information and multimeric geometries, focusing only on physicochemical properties.

Purpose of the Study:

• To introduce PIQLE, a deep graph learning approach for protein-protein interface quality estimation.
• To improve the accuracy of protein complex structure prediction by incorporating novel features.

Main Methods:

• PIQLE utilizes a multi-head graph attention network to analyze multimeric interaction geometries and evolutionary information.
• It combines sequence- and structure-derived features for residue-level interaction quality estimation.
• Probabilistic combination of scores refines the overall interface quality assessment.

Main Results:

• PIQLE demonstrates superior performance compared to state-of-the-art methods like DProQA, TRScore, GNN-DOVE, and DOVE.
• Ablation studies confirm the effectiveness of integrating multimeric geometries and evolutionary information.
• The method shows consistent outperformance across multiple independent test datasets and evaluation metrics.

Conclusions:

• PIQLE offers a significant advancement in protein-protein interface quality estimation.
• The deep graph learning approach effectively leverages complex structural and evolutionary data.
• The open-source implementation facilitates broader application in structural biology.