CATHEDRAL: a fast and effective algorithm to predict folds and domain boundaries from multidomain protein structures

Area of Science:

• Structural bioinformatics
• Computational biology
• Protein structure analysis

Background:

• Accurate identification of protein domains and folds is crucial for understanding protein function and evolution.
• Existing methods for protein domain identification in multidomain structures face challenges in accuracy and automation.

Purpose of the Study:

• To develop and benchmark CATHEDRAL, an iterative protocol for identifying known protein folds in novel multidomain protein structures.
• To improve the accuracy and efficiency of automated protein structure classification.

Main Methods:

• CATHEDRAL utilizes a secondary-structure-based graph theory approach for fold localization.
• A residue-based, double-dynamic programming algorithm aligns target folds to query structures.
• Support vector machines optimize assignment scoring, and an iterative excision process refines domain identification.

Main Results:

• CATHEDRAL achieves superior performance in fold recognition and alignment accuracy compared to existing methods.
• It successfully locates known folds in 90% of novel multidomain structures with <1% false positives.
• Domain boundaries are correctly delineated within ten residues for nearly 80% of assigned domains.

Conclusions:

• CATHEDRAL significantly enhances the automation of protein structure classification by accurately identifying known domains in novel structures.
• Its high accuracy and efficiency offer a substantial improvement over traditional sequence-based methods.
• This protocol will facilitate the analysis of the growing number of newly determined protein structures.