Limits and potential of combined folding and docking

Area of Science:

• Computational biology
• Structural biology
• Bioinformatics

Background:

• Deep learning (DL) has significantly improved de novo protein structure prediction by analyzing co-evolution information from multiple sequence alignments (MSAs).
• This co-evolutionary approach can be extended to predict contacts across protein-protein interfaces for quaternary structure determination.
• Previous studies often did not leverage the latest DL methods for inter-chain contact prediction.

Purpose of the Study:

• To introduce a novel fold-and-dock method for simultaneous prediction of tertiary and quaternary protein structures.
• To improve the accuracy of protein-protein docking using DL-predicted residue-residue distances.
• To develop a robust scoring function for evaluating the accuracy of predicted protein complex structures.

Main Methods:

• A fold-and-dock strategy was developed utilizing trRosetta for predicting residue-residue distances.
• Alternative methods for generating MSAs were explored to enhance inter-chain contact prediction.
• A novel scoring function, PconsDock, was introduced to differentiate correctly and incorrectly docked protein structures.

Main Results:

• The fold-and-dock method, when combined with improved MSA generation, significantly increased accurate protein docking.
• PconsDock demonstrated high accuracy, correctly classifying 98% of folded and docked proteins.
• The method's performance is comparable to traditional docking approaches, with complementary results suggesting combined pipelines could boost success rates.
• This methodology contributed to a top model for a CASP14 oligomeric target (H1065).

Conclusions:

• The developed fold-and-dock method, enhanced by advanced MSA techniques and PconsDock, offers a powerful approach for protein structure prediction.
• Integrating this method into a combined docking pipeline holds promise for significantly improving overall protein complex structure prediction.
• The study highlights the potential of DL-based co-evolutionary analysis for tackling complex quaternary structure prediction challenges.