RRCRank: a fusion method using rank strategy for residue-residue contact prediction

Area of Science:

• Structural biology
• Computational biology
• Bioinformatics

Background:

• Protein residue-residue contacts are vital for protein structure prediction.
• Existing methods for contact prediction have advanced, with fusion methods showing recent promise.
• Constraining conformational search space is crucial for de novo protein structure prediction.

Purpose of the Study:

• To introduce a novel fusion method for predicting protein residue-residue contacts.
• To treat contact prediction as a ranking task rather than regression or classification.
• To leverage learning-to-rank algorithms for improved contact prediction accuracy.

Main Methods:

• Feature extraction from correlated mutations and ensemble machine-learning classifiers.
• Application of a learning-to-rank algorithm to predict contact probabilities for residue pairs.
• Development of the RRCRank fusion method based on rank strategy.

Main Results:

• The RRCRank method demonstrated superior performance compared to existing methods on CASP11 and CASP12 datasets, particularly for medium and short-range contacts.
• The ranking strategy outperformed traditional regression and classification strategies for all contact types, especially long-range contacts.
• RRCRank achieved comparable prediction precisions and outperformed several state-of-the-art methods in benchmark assessments.

Conclusions:

• A novel rank-based method (RRCRank) was developed using a learning-to-rank algorithm for protein residue-residue contact prediction.
• The proposed method achieves state-of-the-art performance.
• The study highlights the effectiveness of a ranking strategy in contact prediction.