Validating subcellular localization prediction tools with mycobacterial proteins.

Area of Science:

• * Bioinformatics
• * Molecular Biology
• * Computational Biology

Background:

• * Accurate prediction of mycobacterial protein subcellular localization is crucial for proteome annotation.
• * This aids in identifying novel drug targets and vaccine candidates.
• * Various computational classifiers exist, including general and feature-based approaches.

Purpose of the Study:

• * To validate the accuracy of different bioinformatics tools in predicting secreted bacterial proteins.
• * To compare the performance of tools like SignalP 2.0, TatP 1.0, LipoP 1.0, Phobius, PA-SUB 2.5, PSORTb v.2.0.4, and Gpos-PLoc.
• * To assess sensitivity, specificity, and Matthew's Correlation Coefficient (MCC) using experimentally validated mycobacterial proteins.

Main Methods:

• * A validation set of 272 experimentally confirmed mycobacterial proteins with <40% identity was curated.
• * Seven distinct bioinformatics tools were employed for prediction.
• * Performance was evaluated using sensitivity, specificity, and MCC metrics.

Main Results:

• * All evaluated tools demonstrated specificity above 0.90.
• * Sensitivity and MCC values generally exceeded 0.22.
• * PA-SUB 2.5 yielded the highest metrics, though potential bias was noted.
• * PSORTb v.2.0.4 excluded 56 proteins, while Gpos-PLoc excluded only one.

Conclusions:

• * Both general and feature-based localization prediction approaches exhibited high specificity.
• * Gpos-PLoc performed best among general localization tools not relying on SWISS-PROT homology.
• * SignalP 2.0 was the top-performing feature-based tool.
• * PA-SUB 2.5's high metrics may be influenced by its training data including the tested proteins.