Efficient global accuracy estimation for protein complex structural models using multi-view representation learning.

Area of Science:

• Computational biology
• Structural bioinformatics
• Machine learning in structural biology

Background:

• Existing estimation of model accuracy (EMA) methods face challenges in balancing computational efficiency and performance with the growing volume of predicted protein structural data.
• Accurate assessment of protein complex model quality is crucial for advancing structural biology and drug discovery.

Purpose of the Study:

• To develop a computationally efficient and high-performing single-model EMA method for protein complex models.
• To improve the selection of accurate protein complex models from large-scale datasets.

Main Methods:

• MViewEMA utilizes a multi-view representation learning framework.
• It integrates residue-residue interaction features from micro, meso, and macro-environment levels.
• The method focuses on global accuracy assessment of protein complex models.

Main Results:

• MViewEMA outperforms state-of-the-art EMA methods in global accuracy assessment.
• It achieves over a 10-fold improvement in computational efficiency compared to DeepUMQA3.
• MViewEMA demonstrated top performance in model selection during the CASP16 blind test.

Conclusions:

• MViewEMA offers an efficient solution for selecting high-quality protein complex models.
• The method has the potential to enhance protein complex structure prediction accuracy when integrated with tools like AlphaFold-Multimer, AlphaFold3, and DiffDock-PP.