Bio-knowledge-based filters improve residue-residue contact prediction accuracy

Area of Science:

• Structural Biology
• Bioinformatics
• Computational Biology

Background:

• Direct coupling analysis (DCA) is a powerful method for predicting residue-residue contacts in proteins, crucial for structural modeling.
• Current DCA accuracy, while impressive, needs improvement for routine protein structure modeling.

Purpose of the Study:

• To enhance the accuracy of residue-residue contact prediction by developing and applying filters for false positive removal.
• To leverage general knowledge of protein structures to refine contact predictions.

Main Methods:

• Development and evaluation of a set of filters designed to identify and remove false positive contact predictions.
• Investigation of filter performance based on one or a few cut-off parameters.
• Application of combined filters using default parameters on a test set of 851 protein domains.

Main Results:

• A combined filter approach successfully removed 29% of predicted contacts from the test set.
• Out of the removed predictions, 92% were confirmed as false positives, indicating high filter efficacy.
• The filters effectively reduce noise in contact predictions, improving the reliability of downstream modeling.

Conclusions:

• Filtering predicted residue-residue contacts using general protein structure knowledge significantly improves prediction accuracy.
• The developed filtering strategy offers a practical approach to enhance the reliability of contact predictions for protein structure modeling.
• The availability of data and scripts facilitates the adoption and further development of these filtering techniques.