SVM-PB-Pred: SVM based protein block prediction method using sequence profiles and secondary structures.

Area of Science:

• Bioinformatics
• Computational Biology
• Structural Bioinformatics

Background:

• Protein structure prediction is crucial for understanding protein function.
• Accurate prediction of protein blocks can aid in determining protein structure and function.
• Existing methods for protein block prediction have limitations.

Purpose of the Study:

• To develop a novel web server, SVM-PB-Pred, for predicting protein blocks.
• To evaluate the performance of support vector machine (SVM) models using various input features and datasets.
• To provide a freely accessible tool for predicting protein block letters from amino acid sequences.

Main Methods:

• Developed a support vector machine (SVM) based web server, SVM-PB-Pred.
• Utilized input features including sequence profiles (PSSM) and secondary structures (SS) from DSSP, NPS@, and GOR4.
• Trained and tested SVM models on four datasets (RS90, DB433, LI1264, SP1577) using self-consistency, cross-validation, and independent tests.

Main Results:

• Maximum prediction accuracy of ~70% was achieved in self-consistency tests using PSSM+SS(DSSP) features on LI1264 and SP1577 datasets.
• Prediction accuracies decreased in independent case tests to ~53% for PSSM+SS(NPS@) and ~43% for PSSM+SS(GOR4).
• The SVM-PB-Pred server demonstrated a prediction accuracy of ~53% using the SP1577 dataset and NPS@ predicted secondary structures.

Conclusions:

• SVM-PB-Pred provides a valuable tool for predicting protein blocks from amino acid sequences.
• The choice of input features and datasets significantly impacts prediction accuracy.
• The developed web server offers a practical solution for researchers in structural bioinformatics and related fields.