TargetVita

TargetVita predicts vitamin-binding residues from protein sequence information to support functional annotation of proteins.

Key Features:

• Feature extraction (PSSM): Extracts Position-Specific Scoring Matrix (PSSM) features from protein sequences to capture evolutionary information.
• Predicted secondary structure: Incorporates predicted protein secondary structures to provide structural context for residue-level predictions.
• Vitamin binding propensity: Incorporates vitamin binding propensity scores to enhance specificity of vitamin-binding residue prediction.
• Feature selection: Applies various feature selection methods to identify optimal feature subspaces from the original feature set.
• Machine learning model: Trains heterogeneous Support Vector Machines (SVMs) on selected feature subspaces.
• Ensemble prediction: Combines an ensemble of SVMs to produce final residue-level binding predictions.
• Benchmarking and performance: Evaluated on four separate vitamin-binding benchmark datasets and reported an average ~10% improvement in Matthews Correlation Coefficient (MCC) on independent tests.

Scientific Applications:

• Residue-level prediction: Identification of protein vitamin-binding residues from sequence data.
• Functional annotation: Supporting functional annotation of proteins by labeling vitamin-binding sites.
• Annotation of uncharacterized proteins: Prioritizing annotation of unannotated or newly sequenced proteins in post-genomic datasets.

Methodology:

Computational steps explicitly include extracting PSSM features, incorporating predicted secondary structures and vitamin binding propensity, applying feature selection to obtain optimal feature subspaces, training heterogeneous SVMs on those subspaces, combining SVMs in an ensemble for final predictions, and evaluating performance on four vitamin-binding benchmark datasets.