Extending Protein Domain Boundary Predictors to Detect Discontinuous Domains

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Protein Domain Prediction

Background:

• Most protein domain predictors struggle with discontinuous domains, despite their prevalence in multidomain proteins.
• Discontinuous domains are crucial structural and functional units in nearly 40% of multidomain proteins.

Purpose of the Study:

• To develop a novel method, DomEx, for predicting discontinuous protein domains.
• To improve the accuracy and recall of discontinuous domain prediction by assembling continuous domain segments.

Main Methods:

• DomEx assembles continuous domain segments to predict discontinuous domains.
• It matches sequence profiles of concatenated segments against a single-domain library (SCOP, CATH, Pfam).
• Matches are filtered using similarity scores, a symmetric index, and profile-profile alignment.

Main Results:

• DomEx achieved 32.3% recall and 86.5% precision on a test set.
• It demonstrated superior performance compared to ThreaDom, recalling 26.7% with 72.7% precision.
• The combined approach with ThreaDom ranked first among 10 predictors.

Conclusions:

• DomEx effectively predicts discontinuous protein domains by leveraging continuous segments.
• The method offers significant improvements over existing tools and enhances prediction accuracy when combined with other predictors.