TBpred: SVM-Based Prediction of Subcellular Localization of Mycobacterial Proteins Using Evolutionary Information and Motifs

TBpred: SVM-Based Prediction of Subcellular Localization of Mycobacterial Proteins Using Evolutionary Information and Motifs TBpred is a computational web server developed for predicting the subcellular localization of mycobacterial proteins. The tool predicts whether a mycobacterial protein belongs to cytoplasmic, integral membrane, secretory, or membrane-attached protein classes. TBpred uses support vector machine-based models, evolutionary information from PSSM profiles, HMM profiles, MEME/MAST motif search, and hybrid prediction strategies. Web Server: https://webs.iiitd.edu.in/raghava/tbpred/ Citation Rashid, M., Saha, S., and Raghava, G. P. S. Support Vector Machine-based method for predicting subcellular localization of mycobacterial proteins using evolutionary information and motifs. BMC Bioinformatics, 8, 337, 2007. https://doi.org/10.1186/1471-2105-8-337 About the Research Mycobacterium species include important pathogenic organisms such as Mycobacterium tuberculosis. Many mycobacterial proteins remain functionally unannotated, and knowing their subcellular localization can help understand their biological role, virulence, antigenic potential, and suitability as drug or vaccine targets. Existing subcellular localization methods were developed mainly for eukaryotic, Gram-positive, Gram-negative, or human proteins. However, mycobacteria have a unique and complex cell wall structure, so a dedicated organism-specific prediction method was needed. TBpred was developed to predict the subcellular location of mycobacterial proteins directly from protein sequence information. Data Compilation: The study used 852 mycobacterial proteins extracted from Swiss-Prot. These proteins were grouped into four major subcellular localization classes: 340 cytoplasmic proteins, 402 integral membrane proteins, 50 secretory proteins, and 60 membrane-attached proteins. Methodology: TBpred uses machine learning and motif-based approaches. The study developed models using amino acid composition, dipeptide composition, PSSM profiles generated from PSI-BLAST, HMM profiles, MEME/MAST motifs, and hybrid combinations of these methods.