tbpred

tbpred predicts the subcellular localization of mycobacterial proteins, classifying sequences as cytoplasmic, integral membrane, secretory, or membrane-attached by lipid anchor.

Key Features:

• Subcellular localization classes: Predicts four distinct localizations for mycobacterial proteins: cytoplasmic, integral membrane, secretory, and membrane-attached by lipid anchor.
• SVM models (amino acid composition): Support Vector Machine models trained on amino acid composition achieved overall accuracy of 82.51% and mean class-wise accuracy of 68.47%.
• PSSM-based SVM (PSI-BLAST): SVM models incorporating Position-Specific Scoring Matrix (PSSM) profiles derived from PSI-BLAST achieved overall accuracy of 86.62% and average accuracy of 73.71%.
• Hybrid models (PSSM + MEME/MAST): Combining PSSM-based SVM models with MEME/MAST motif analysis produced a maximum overall accuracy of 86.8% and an average accuracy reported as 89.00%.
• Training and validation: Models were trained and tested on a dataset of 852 mycobacterial proteins using five-fold cross-validation.
• Comparative benchmarking: Performance was benchmarked against existing methods for Gram-positive bacterial subcellular localization and reported superior accuracy.

Scientific Applications:

• Protein annotation: Predicts localization to support annotation of newly sequenced or hypothetical mycobacterial proteins.
• Immunogen discovery: Identifies membrane-attached and secretory proteins to aid selection of potential immunogens relevant to vaccine development and pathogenicity studies.

Methodology:

Support Vector Machine models using amino acid composition; SVM models incorporating PSSM profiles derived from PSI-BLAST; integration of MEME/MAST motif analysis with PSSM-based SVMs; training and testing on 852 mycobacterial proteins with five-fold cross-validation.