ProLoc-GO

ProLoc-GO predicts protein subcellular localization from amino-acid sequences by mining informative Gene Ontology (GO) annotations and using them as features for Support Vector Machine (SVM) classification.

Key Features:

• Gene Ontology Utilization: Uses Gene Ontology (GO) annotations as functional features for localization prediction.
• Sequence-Based Prediction: Operates on input amino-acid sequences to predict localization, enabling analysis of proteins lacking known accession numbers.
• BLAST Homology Search: Applies BLAST to find homologous proteins and retrieve their accession-linked GO annotations.
• GOmining Genetic Algorithm: Implements GOmining, a genetic algorithm that selects a small informative subset of GO terms from the full GO term pool.
• Targeted Subcellular Compartment Annotation: Selected GO terms annotate compartments such as nucleus (GO:0005634), cytoplasm (GO:0005737), and cytoskeleton (GO:0005856).
• SVM Classifier: Uses the selected GO terms as input features for a Support Vector Machine (SVM) classifier to distinguish subcellular localizations.
• Efficiency in Feature Selection: Reduces the GO term set from n to m (m << n), for example using 44 informative GO terms for SCL12 and 60 for SCL16.

Scientific Applications:

• Localization of Novel Proteins: Predicts subcellular localization for uncharacterized or novel proteins without known accession numbers.
• Benchmarking on SCL12 and SCL16: Demonstrates test accuracies of 88.1% on SCL12 (human, 12 locations) and 83.3% on SCL16 (eukaryotic, 16 locations).
• Improved Performance with Accession Numbers: With known accession numbers, achieves 90.6% (SCL12) and 85.7% (SCL16), outperforming ensemble classifiers such as Hum-PLoc and Euk-OET-PLoc.
• Comparison to Composition-Based Methods: Substantially outperforms SVM classifiers based on amino-acid composition, which achieve under 35% accuracy on the same datasets.

Methodology:

Performs BLAST homology search to retrieve GO annotations, applies the GOmining genetic algorithm to select informative GO terms, and trains a Support Vector Machine (SVM) classifier on those GO-term features, with example feature sets of 44 terms for SCL12 and 60 terms for SCL16.