Boosting AlphaFold Protein Tertiary Structure Prediction through MSA Engineering and Extensive Model Sampling and...

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• AlphaFold2 and AlphaFold3 excel at predicting single-chain protein structures.
• Predicting structures for proteins with poor multiple sequence alignments (MSAs) or complex architectures remains difficult.

Purpose of the Study:

• To develop an integrative system, MULTICOM4, to enhance protein structure prediction accuracy for challenging targets.
• To improve the generation and ranking of protein models from AlphaFold2 and AlphaFold3.

Main Methods:

• MULTICOM4 utilizes diverse MSA generation strategies and extensive model sampling.
• An ensemble quality assessment (QA) approach combines multiple QA methods for improved model ranking.
• The system was evaluated in the Critical Assessment of Techniques for Protein Structure Prediction (CASP16).

Main Results:

• MULTICOM4 ranked among the top performers in CASP16, outperforming a standard AlphaFold3 predictor.
• The best MULTICOM4 predictor achieved an average TM-score of 0.902 for 84 CASP16 domains.
• High accuracy (TM-score > 0.9) was achieved for 73.8% of domains, and correct folds (TM-score > 0.5) for 97.6% in top-1 predictions.

Conclusions:

• MSA engineering and extensive model sampling are crucial for accurate protein structure modeling.
• Ensemble QA methods and model clustering enhance the reliability of structure prediction ranking.