Predicting protein contact map using evolutionary and physical constraints by integer programming.

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Machine Learning

Background:

• Protein contact maps are crucial for 3D structure prediction, detailing residue relationships.
• Predicting contact maps solely from sequence remains challenging, with existing methods often neglecting contact correlations and physical feasibility.
• Current methods using mutual information require extensive sequence homologs and may yield physically infeasible results.

Purpose of the Study:

• To develop a novel method, PhyCMAP, for accurate protein contact map prediction.
• To integrate both evolutionary and physical restraints for improved prediction accuracy.
• To overcome limitations of existing methods that ignore contact correlations and physical constraints.

Main Methods:

• PhyCMAP employs a machine learning approach combined with integer linear programming.
• It integrates evolutionary restraints, which are more informative than mutual information.
• Physical restraints are incorporated to define concrete relationships among contacts, enhancing feasibility.

Main Results:

• PhyCMAP significantly reduces the solution space for contact map matrices.
• The method demonstrates improved prediction accuracy compared to existing popular methods.
• PhyCMAP's performance is robust, outperforming other methods irrespective of the number of available sequence homologs.

Conclusions:

• PhyCMAP offers a more accurate and physically feasible approach to protein contact map prediction.
• The integration of evolutionary and physical restraints is key to its improved performance.
• This method advances the field of protein 3D structure prediction, particularly when limited sequence homologs are available.