Evaluation and comparison of mammalian subcellular localization prediction methods

Area of Science:

• Bioinformatics
• Computational Biology
• Proteomics

Background:

• Determining protein subcellular localization is crucial for understanding protein function, especially for novel proteins identified through genome sequencing.
• Experimental determination is challenging, driving the development of computational prediction methods.
• These methods vary in accuracy and performance depending on the input sequences.

Purpose of the Study:

• To comprehensively survey and compare the performance of publicly available computational methods for predicting protein subcellular localization.
• To evaluate these methods using both established and independent datasets to assess their reliability.

Main Methods:

• Selected five prediction methods (CELLO, MultiLoc, Proteome Analyst, pTarget, WoLF PSORT) based on batch processing, public availability, and prediction of at least nine major subcellular locations.
• Evaluated methods using 3763 mouse proteins from SwissProt and an independent set of 2145 mouse proteins from LOCATE.
• Calculated sensitivity and specificity for each method.

Main Results:

• No single method demonstrated sufficient sensitivity across both datasets for reliable prediction of hypothetical protein localization.
• All evaluated methods performed less effectively on the independent LOCATE dataset compared to the SwissProt dataset.
• Performance varied across different subcellular locations, with secretory pathway proteins being the most difficult to predict accurately.

Conclusions:

• Current computational methods are insufficient for the reliable prediction of subcellular localization in hypothetical proteins.
• Further development is needed to improve accuracy, particularly for proteins in the secretory pathway.
• Nuclear and extracellular proteins were predicted with the highest sensitivity among the evaluated locations.