Increasing the accuracy of protein loop structure prediction with evolutionary constraints

Area of Science:

• Structural biology
• Computational biology
• Bioinformatics

Background:

• Accurate prediction of protein loop structures, especially long ones, is a persistent challenge in structural biology.
• The vast conformational space and limited experimental data for long loops contribute to low prediction accuracy.
• Co-evolutionary contact predictors offer insights into residue proximity and have shown promise in improving whole-protein modeling.

Purpose of the Study:

• To investigate the efficacy of co-evolutionary constraints in enhancing the prediction accuracy of long protein loop structures.
• To assess the accuracy of predicted contacts specifically within loop regions.
• To evaluate different strategies for integrating contact information into loop structure prediction.

Main Methods:

• Assessed the accuracy of predicted contacts involving loop regions.
• Developed and examined two distinct strategies for incorporating co-evolutionary contact information.
• Tested the methods on a dataset of long loops (≥10 residues).
• Analyzed the impact on the proportion of near-native decoys and the root-mean-square deviation (RMSD) of predicted models.

Main Results:

• Predicted contacts involving loop regions were found to be less accurate than general contacts.
• A method was identified to filter out incorrectly predicted contacts that are never satisfied in generated conformations.
• Both tested strategies demonstrated improved prediction accuracy for long loops.
• Near-native decoys increased, and RMSD improved by over 3 Å in some cases for 97 out of 135 applicable loops.

Conclusions:

• Co-evolutionary contact predictions can significantly enhance the accuracy of long loop structure prediction.
• The integration of evolutionary constraints offers a viable approach to overcome limitations in predicting complex loop conformations.
• The developed strategies provide a valuable tool for improving protein structure modeling, particularly for challenging loop regions.