PA-SUB

PA-SUB predicts protein subcellular localization across animals, plants, fungi, Gram-negative bacteria and Gram-positive bacteria using Naïve Bayes classifiers trained on database text annotations from homologous proteins to support functional annotation and proteome-scale analyses.

Key Features:

• Machine Learning-Based Predictions: Uses Naïve Bayes classifiers trained on database text annotations from homologous proteins rather than relying solely on sequence information.
• High Accuracy Across Diverse Organisms: Reports prediction accuracies of 81% for fungi and 92–94% for animals, plants, Gram-negative bacteria and Gram-positive bacteria.
• Comprehensive Coverage: Covers a wide array of organisms and protein types for broad subcellular localization prediction.
• Custom Classifier Creation: Supports creation of custom classifiers trained on user-provided labeled training data to predict new properties or labels.
• Sophisticated Explanation Feature: Provides an interactive explanation mechanism that elucidates why particular localization predictions are made.
• Integration with Proteome Analyst (PA): Implemented as the subcellular localization component of the Proteome Analyst (PA) framework.

Scientific Applications:

• Functional Annotation: Infers potential protein functions by predicting subcellular localization to aid annotation efforts.
• Proteome Analysis: Enables high-throughput, proteome-scale analysis of subcellular localization patterns.
• Custom Research Needs: Allows targeted investigations using custom classifiers for specialized ontologies or protein families, such as potassium-ion channels.

Methodology:

Naïve Bayes classifiers trained on database text annotations from homologous proteins; support for user-trained custom classifiers using labeled training data; interactive explanation of prediction decisions.