Combining pairwise structural similarity and deep learning interface contact prediction to estimate protein complex...

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Protein Structure Prediction

Background:

• Accurate estimation of quaternary structural models (EMA) is crucial for understanding protein function and interactions.
• Existing methods for estimating protein tertiary model quality are not always effective for quaternary models, especially when models are low quality and similar.
• The pairwise similarity approach has limitations in assessing the quality of protein complex models.

Purpose of the Study:

• To develop and evaluate a hybrid method for estimating the accuracy of protein quaternary structural models.
• To address the limitations of existing methods in handling low-quality and similar protein complex models.
• To identify key factors influencing the accuracy of quaternary structure model estimation.

Main Methods:

• Developed MULTICOM_qa, a hybrid method integrating a pairwise similarity score (PSS) and an interface contact probability score (ICPS).
• Utilized deep learning for inter-chain contact prediction within the ICPS component.
• Blindly participated in the 15th Critical Assessment of Techniques for Protein Structure Prediction (CASP15) to evaluate performance.

Main Results:

• MULTICOM_qa demonstrated strong performance in estimating the global structure accuracy of protein assembly models at CASP15.
• Achieved an average per-target correlation coefficient of 0.66 between predicted and true model quality scores.
• Showcased effective model selection capabilities with an average per-target ranking loss of 0.14.

Conclusions:

• The hybrid approach combining multi-model (PSS) and single-model (ICPS) methods is a promising strategy for EMA.
• MULTICOM_qa successfully estimated the accuracy of protein complex models in a large-scale assessment.
• Identified and analyzed critical factors affecting EMA, including target and model sampling difficulty, and model quality distribution.