Ordering Protein Contact Matrices.

Area of Science:

• Biophysics
• Computational Biology
• Structural Bioinformatics

Background:

• Biophysical methods yield protein residue spatial proximity data.
• Assigning protein folds from proximity data is challenging without primary sequence residue identification.

Purpose of the Study:

• To develop an algorithm for assigning residue numbers from raw protein contact matrices.
• To enable protein fold prediction using noisy spatial proximity information.

Main Methods:

• Formulating residue ordering as a graph trail search problem.
• Employing dynamic programming to find the longest trail in the residue contact graph.
• Utilizing Self-Organizing Maps (SOM) for clustering ordered matrices.

Main Results:

• The algorithm successfully orders residues from proximity data, even with up to 40% noise.
• Achieved average gaps of ~20% between ordered and initial matrices.
• Demonstrated potential for fold prediction with TM scores > 0.5 at 10% noise levels.

Conclusions:

• The combined dynamic programming and SOM approach offers an innovative solution for residue assignment.
• This method facilitates protein fold prediction from noisy proximity data.
• Validated on a large protein dataset and experimental cases, showing satisfactory performance.