PSLDoc: Protein subcellular localization prediction based on gapped-dipeptides and probabilistic latent semantic...

Area of Science:

• Bioinformatics
• Computational Biology
• Proteomics

Background:

• Protein subcellular localization (PSL) is crucial for understanding protein function, genome annotation, and drug discovery.
• Accurate computational methods for PSL prediction are essential, particularly for Gram-negative bacteria.
• Existing methods often rely on sequence homology, which can limit performance on diverse datasets.

Purpose of the Study:

• To develop a novel computational method, PSLDoc, for predicting protein subcellular localization.
• To leverage gapped-dipeptides and probabilistic latent semantic analysis (PLSA) for improved feature representation.
• To evaluate PSLDoc's performance against established methods like CELLO II and PSORTb v.2.0.

Main Methods:

• Proteins represented as term strings of gapped-dipeptides (residues separated by one or more positions).
• Gapped-dipeptide weighting incorporates sequence evolutionary information via position-specific score matrices.
• Probabilistic latent semantic analysis (PLSA) for feature reduction, followed by five one-versus-rest support vector machine classifiers.

Main Results:

• PSLDoc achieved high accuracy: 86.84% on low-homology and 98.21% on high-homology datasets.
• Demonstrated superior performance compared to CELLO II.
• Achieved 97.89% precision at a 0.7 confidence threshold, outperforming PSORTb v.2.0.

Conclusions:

• PSLDoc effectively predicts protein subcellular localization in Gram-negative bacteria using a novel feature representation.
• The method shows significant improvements in prediction accuracy and precision.
• The approach is generalizable and can be extended to eukaryotic proteomes.