Protein Multiple Conformation Prediction Using Multi-Objective Evolution Algorithm.

Area of Science:

• Computational biology
• Structural bioinformatics
• Protein structure prediction

Background:

• AlphaFold2 and AlphaFold DB significantly advanced static protein structure prediction.
• Predicting multiple protein conformations remains a key challenge in structural biology.

Purpose of the Study:

• To develop a novel method, MultiSFold, for predicting multiple protein conformations.
• To evaluate MultiSFold's performance against state-of-the-art methods in conformational sampling.

Main Methods:

• Utilized a distance-based multi-objective evolutionary algorithm for multiple conformation prediction.
• Employed deep learning to generate competing constraints for energy landscape construction.
• Incorporated iterative modal exploration, multi-objective optimization, and geometric optimization for conformation sampling.

Main Results:

• MultiSFold achieved a 56.25% success rate in predicting multiple conformations, significantly outperforming AlphaFold2's 10.00%.
• MultiSFold improved TM-scores by 2.97% over AlphaFold2 and 7.72% over RoseTTAFold for static structure prediction.
• The method demonstrated potential in generating conformations across different conformational states.

Conclusions:

• MultiSFold effectively predicts multiple protein conformations, addressing a critical limitation of existing methods.
• Conformational sampling combined with deep learning shows promise for comprehensive protein structure modeling.
• MultiSFold offers enhanced accuracy for both dynamic and static protein structure predictions.