TopoQA: a topological deep learning-based approach for protein complex structure interface quality assessment.

Area of Science:

• Computational Biology
• Structural Biology
• Machine Learning

Background:

• Protein complex structure prediction accuracy lags behind monomer prediction.
• Accurate quality assessment (QA) is crucial for selecting reliable protein complex models.
• Existing QA methods struggle with the complexity of protein interfaces.

Purpose of the Study:

• To develop an efficient and effective QA method for protein complex structures.
• To leverage topological features for improved accuracy estimation.
• To assess the performance of the new method against state-of-the-art models.

Main Methods:

• Utilized persistent homology (PH) to capture atomic-level topological information.
• Integrated PH with graph neural networks (GNNs) to create TopoQA.
• Validated TopoQA on benchmark datasets: DBM55-AF2, HAF2, and ABAG-AF3.

Main Results:

• TopoQA outperforms AF-Multimer-based AF2Rank on all tested datasets.
• TopoQA shows an advantage over AlphaFold3 (AF3) in nearly half of the targets.
• Achieved a 73.6% lower ranking loss on DBM55-AF2 compared to AF2Rank.
• Demonstrated superior Top 10 Hit-rate on DBM55-AF2 and lowest ranking loss on HAF2 among state-of-the-art methods.
• Ablation studies confirmed the significant contribution of topological features.

Conclusions:

• TopoQA offers a robust and accurate method for protein complex quality assessment.
• The integration of topological data analysis enhances deep learning models for structural biology.
• TopoQA provides a new paradigm for protein structure representation learning and QA.