Prediction of heterotrimeric protein complexes by two-phase learning using neighboring kernels

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• Protein complexes are vital in biological processes like gene regulation and metabolism.
• Existing prediction methods often focus on larger complexes, overlooking smaller ones which are prevalent.
• Previous work established a successful method for heterodimeric complex prediction.

Purpose of the Study:

• To develop and evaluate novel methods for predicting heterotrimeric protein complexes.
• To extend previous techniques for heterodimeric complex prediction to heterotrimeric complexes.
• To improve the accuracy of identifying small protein complexes.

Main Methods:

• Proposed a two-phase prediction strategy building on prior work.
• Designed novel feature space mappings for the second prediction phase.
• Utilized support vector machines (SVMs) and relevance vector machines (RVMs) as classifiers.
• Incorporated a domain composition kernel.

Main Results:

• The proposed two-phase methods demonstrated superior performance compared to the existing NWE method.
• The enhanced SVM with extended features outperformed other established methods for heterotrimeric complex prediction.
• Computational experiments using 10-fold cross-validation validated the findings.

Conclusions:

• The developed two-phase prediction methods, incorporating extended features and a domain composition kernel, are effective for heterotrimeric protein complex identification.
• The combination of extended features, domain composition kernel, and SVM as the second classifier offers a particularly powerful approach.