Mycosub

Mycosub predicts the subcellular localization of proteins in Mycobacterium tuberculosis to inform protein function and support drug discovery.

Key Features:

• Optimal Tripeptide Composition Analysis: Utilizes tripeptide composition patterns to predict protein subcellular localization in mycobacteria.
• Feature Selection Strategy: Applies a binomial distribution-based strategy to select 219 tripeptide features most indicative of localization.
• Support Vector Machine (SVM) Integration: Employs a support vector machine classifier for localization prediction.
• Benchmark Dataset: Built and evaluated on a dataset of 272 non-redundant proteins sourced from the UniProt database.
• Validation and Performance: Evaluated by jackknife cross-validation with a reported maximum overall accuracy of 89.71% and an average accuracy of 81.12%.

Scientific Applications:

• Protein Function Prediction: Infers protein function in Mycobacterium tuberculosis by determining subcellular localization.
• Drug Discovery and Design: Provides localization-based insights to support identification and targeting of mycobacterial drug targets.

Methodology:

A curated dataset of 272 non-redundant Mycobacterium tuberculosis proteins from UniProt was assembled; tripeptide features were selected using a binomial distribution-based strategy to yield 219 features; an SVM model was trained and evaluated by jackknife cross-validation.