eslpred2

eslpred2 predicts eukaryotic protein subcellular localization (cytoplasmic, mitochondrial, nuclear, and extracellular) to support protein function characterization.

Key Features:

• Predicted compartments: Predicts four eukaryotic subcellular localizations: cytoplasmic, mitochondrial, nuclear, and extracellular.
• Non-redundant dataset: Uses a highly non-redundant dataset containing 1198 fungal, 2597 animal, and 491 plant protein sequences.
• Feature extraction: Employs evolutionary information as profile composition together with whole and N-terminal sequence compositions as input features.
• Feature representation: Integrates the extracted features into a 440-dimensional feature vector.
• Machine learning model: Trains Support Vector Machine (SVM) models for localization prediction.
• Validation and reported accuracies: Evaluated by five-fold cross-validation with initial accuracies of 72.7% (fungal), 75.8% (animal), and 74.5% (plant), and improved accuracies of 75.9%, 80.8%, and 76.6% when similarity search results were combined with the features.
• Integration of methods: Combines sequence composition, profile composition, and similarity search results to produce the reported SVM prediction performance.

Scientific Applications:

• Protein function inference: Provides subcellular localization predictions to inform protein role and functional annotation studies.
• Protein deciphering studies: Aids rapid characterization of proteins by supplying likely cellular compartment assignments.
• Support for molecular biology and genetics: Supplies localization data useful for experimental planning and interpretation in molecular biology, genetics, and bioinformatics research.

Methodology:

Uses profile composition and whole plus N-terminal sequence compositions to form a 440-dimensional vector, trains Support Vector Machine models, and evaluates performance with five-fold cross-validation, including combination of similarity search results with these features.