conSSert: Consensus SVM Model for Accurate Prediction of Ordered Secondary Structure

Area of Science:

• * Computational biology
• * Bioinformatics
• * Structural biology

Background:

• * Accurate protein secondary structure prediction is vital for understanding protein folding.
• * Predicting ordered secondary structures, particularly beta-strands, remains a significant challenge.

Purpose of the Study:

• * To develop an enhanced consensus method for protein secondary structure prediction.
• * To improve the accuracy of beta-strand prediction.

Main Methods:

• * Developed conSSert, a consensus secondary structure prediction method.
• * Utilized support vector machines (SVM) as the core algorithm.
• * Integrated predictions from four leading methods: PSSpred, PSIPRED, SPINE-X, and RAPTOR.
• * Trained and tested on extensive datasets (PDBSelect25, PISCES) and validated with CASP targets.

Main Results:

• * conSSert achieved exceptional accuracy in beta-strand prediction (QE > 0.82, Q2-EH = 0.86).
• * Findings suggest training bias, not just non-local contacts, contributes to strand prediction difficulties.
• * The method demonstrates superior performance compared to existing approaches.

Conclusions:

• * conSSert offers a significant advancement in predicting protein secondary structures, particularly beta-strands.
• * The developed method addresses limitations in current prediction techniques.
• * conSSert is available as a free webservice for non-commercial use.