Comparative study of the effectiveness and limitations of current methods for detecting sequence coevolution.

Area of Science:

• Bioinformatics
• Computational Biology
• Protein Science

Background:

• Extracting information from multiple sequence alignments (MSAs) is vital due to rapid sequence data accumulation.
• Coevolutionary analysis aims to infer protein structure and function from coupled evolutionary changes.
• Existing methods face challenges with false positives due to insufficient MSA size, phylogeny, and indirect couplings.

Purpose of the Study:

• To assess the effectiveness and limitations of various coevolutionary analysis methods.
• To identify strategies for improving the accuracy of detecting structural and functional relationships from MSAs.
• To provide guidelines for selecting appropriate methods based on MSA size and computational resources.

Main Methods:

• Evaluation of 16 pairs of non-interacting proteins using different coevolutionary analysis methods.
• Application of traditional methods (e.g., mutual information with shuffling) and computationally intensive methods.
• Validation using 2,330 protein families from the Negatome database and a training set of 162 protein families.

Main Results:

• Computationally expensive methods excel at detecting tertiary structural contacts and minimizing false positives from indirect couplings.
• Refined traditional methods like mutual information with shuffling are highly efficient.
• A novel combined method, developed from a training dataset, demonstrates superior performance over individual existing methods.

Conclusions:

• The study offers practical guidelines for selecting coevolutionary analysis methods and strategies.
• The choice of method should consider the available multiple sequence alignment size and computational constraints.
• Accurate coevolutionary analysis is essential for robust inferences of protein structure and function.