Prediction of cis/trans isomerization using feature selection and support vector machines

Area of Science:

• * Structural Biology
• * Bioinformatics
• * Computational Biology

Background:

• * Peptide bonds in proteins predominantly adopt a trans conformation.
• * Cis conformation is rare but significant in biological processes.
• * Most cis peptide bonds involve proline residues (X-Pro imides).

Purpose of the Study:

• * To develop a reliable method for predicting peptide bond conformation (cis/trans isomerization).
• * To evaluate various features for predicting both proline and non-proline cis/trans isomerization.
• * To identify key features and residue contributions influencing peptide bond conformation.

Main Methods:

• * Utilized evolutionary profiles, secondary structure, solvent accessibility, and physicochemical properties.
• * Explored a modified feature vector including condensed position-specific scoring matrices (PSSMX).
• * Employed a wrapper feature selection algorithm and a Support Vector Machine (SVM).

Main Results:

• * Achieved 70% accuracy, 75% sensitivity, and 71% positive predictive value (PPV) for peptide bond conformation prediction.
• * The best performance was obtained using evolutionary profiles, secondary structure, solvent accessibility, and physicochemical properties.
• * Feature selection identified discriminatory features and the contribution of neighboring residues.

Conclusions:

• * The developed method accurately predicts peptide bond conformation, advancing understanding of cis/trans isomerization.
• * Identified key features that contribute to the prediction accuracy.
• * Provides insights into the role of neighboring residues in determining peptide bond conformation.