vitapred

vitapred predicts vitamin-interacting residues in protein sequences using Support Vector Machines (SVM) trained on binary and Position-Specific Scoring Matrix (PSSM) features to identify residues interacting with vitamins including vitamin A, vitamin B, and pyridoxal-5-phosphate.

Key Features:

• Machine Learning Approach: vitapred uses Support Vector Machines (SVM) trained on binary and Position-Specific Scoring Matrix (PSSM) features derived from protein sequences.
• Comprehensive Prediction Modules: The method provides modules for Vitamin Interacting Residues (VIRs), Vitamin-A Interacting Residues (VAIRs), Vitamin-B Interacting Residues (VBIRs), and Pyridoxal-5-phosphate (Vitamin B6) Interacting Residues (PLPIRs).
• Evolutionary Information Utilization: vitapred leverages PSSM-based evolutionary information and reports Matthews Correlation Coefficient (MCC) scores of 0.53 for VIRs, 0.48 for VAIRs, 0.61 for VBIRs, and 0.81 for PLPIRs.
• Comparative Analysis: Two Sample Logo (TSL) analysis was used to compare amino acid preferences for vitamin interactions against other ligands such as ATP, GTP, NAD, FAD, and mannose.
• Training and Validation: Models were trained and tested on non-redundant datasets using five-fold cross-validation and further evaluated on balanced and independent datasets.

Scientific Applications:

• Residue-level annotation: Predicting vitamin-binding residues to map vitamin interaction sites on protein sequences.
• Inhibitor design: Informing design of inhibitors that target specific vitamin-protein interactions by identifying interacting residues.
• Structural and functional analysis: Supporting investigations of structural requirements for vitamin binding in studies of enzymatic cofactors.
• Drug discovery and biochemical research: Contributing predictive data useful for medicinal chemistry, drug discovery, and broader research in biochemistry and molecular biology.

Methodology:

Support Vector Machine classification using binary and PSSM features derived from protein sequences, Two Sample Logo (TSL) comparative analysis, training/testing on non-redundant datasets with five-fold cross-validation, evaluation on balanced and independent datasets, and reporting performance by Matthews Correlation Coefficient (MCC).