Improved protein structure prediction with trRosettaX2, AlphaFold2, and optimized MSAs in CASP15

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• Accurate protein structure prediction is crucial for understanding biological function and disease mechanisms.
• The Critical Assessment of protein Structure Prediction (CASP) is a community-wide experiment to assess the accuracy of protein structure prediction methods.
• Advancements in deep learning have revolutionized protein structure prediction, but challenges remain for complex targets like multimers.

Purpose of the Study:

• To present the performance of our novel protein structure prediction pipeline in the CASP15 competition.
• To evaluate the effectiveness of our method in predicting both monomer and multimer protein structures.
• To identify key factors contributing to prediction accuracy and areas requiring further development.

Main Methods:

• Developed an elaborate pipeline leveraging complementary sequence databases and advanced searching algorithms to generate high-quality multiple sequence alignments (MSAs).
• Utilized trRosettaX2 and AlphaFold2 for monomer structure prediction (Yang-Server) and AlphaFold-Multimer for multimer structure prediction (Yang-Multimer).
• Compared prediction results against default AlphaFold2 and AlphaFold-Multimer implementations.

Main Results:

• Yang-Server achieved top ranking for monomer structure prediction, with an average TM-score of 0.876, outperforming default AlphaFold2 (0.798).
• Yang-Multimer ranked fourth for multimer structure prediction, showing an average DockQ score of 0.464, exceeding default AlphaFold-Multimer (0.389).
• Improvements were attributed to enhanced MSAs, iterated modeling for large targets, and interplay between monomer and multimer prediction strategies.

Conclusions:

• Our pipeline demonstrates significant improvements in both monomer and multimer protein structure prediction accuracy.
• Enhanced MSAs and advanced modeling techniques are key drivers of prediction performance.
• Structure prediction for orphan proteins and complex multimers remains a challenging area requiring future breakthroughs.