Protein contact prediction by integrating joint evolutionary coupling analysis and supervised learning

Area of Science:

• Computational Biology
• Bioinformatics
• Structural Biology

Background:

• Protein contact prediction is crucial for understanding protein structure and function.
• Existing methods like evolutionary coupling (EC) analysis and machine learning have limitations, particularly for proteins with few homologous sequences.
• Accurate contact prediction remains a challenge in structural biology.

Purpose of the Study:

• To develop an improved method for protein contact prediction, especially for proteins lacking numerous sequence homologs.
• To enhance the accuracy of contact prediction by integrating diverse evolutionary and machine learning information.
• To provide a novel approach for analyzing residue coevolution across related protein families.

Main Methods:

• A group graphical lasso (GGL) method was developed, integrating joint multi-family EC analysis with supervised learning.
• Joint EC analysis leverages residue coevolution information from both the target protein family and related families with similar folds.
• Gaussian graphical models were used to coestimate parameters across related protein families, enforcing similarity constraints.

Main Results:

• The GGL method demonstrated superior performance compared to existing methods for proteins with limited sequence homologs.
• The approach showed improved accuracy in predicting both conserved and family-specific protein contacts.
• Integration of supervised learning further enhanced the predictive accuracy of the GGL method.

Conclusions:

• The GGL method offers a significant advancement in protein contact prediction, particularly for challenging cases with sparse sequence data.
• Joint multi-family EC analysis provides a powerful strategy to exploit evolutionary information from related protein families.
• The developed method has implications for protein structure prediction and functional studies.