Multi contact-based folding method for de novo protein structure prediction.

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• Meta contact methods combine contact maps for improved protein contact prediction accuracy.
• Existing meta contact approaches do not fully leverage information from original contact maps.

Purpose of the Study:

• To propose MultiCFold, a novel multi-contact-based protein structure prediction method.
• To enhance protein folding by directly utilizing diverse contact map information and noncontact data.

Main Methods:

• Developed MultiCFold, integrating evolutionary algorithms for protein structure folding.
• Employed populations to guide folding using detailed information from multiple contact maps.
• Incorporated noncontact information as a supplementary feature for protein folding.

Main Results:

• MultiCFold achieved an average TM-score of 0.617 and average RMSD of 5.815 Å on 120 nonredundant proteins.
• Demonstrated a 6.62% improvement in TM-score and 8.82% in RMSD compared to MetaCFold.
• Noncontact information addition boosted the average TM-score by 6.30%.

Conclusions:

• MultiCFold significantly improves protein structure prediction accuracy by fully exploiting contact map data.
• The inclusion of noncontact information further enhances prediction performance.
• MultiCFold shows superior performance against state-of-the-art methods on CASP13 FM targets.