LocTree3

LocTree3 predicts protein subcellular localization across eukaryota, bacteria, and archaea using a profile-kernel support vector machine combined with homology-based inference to assign proteins to multiple cellular compartments for functional annotation.

Key Features:

• Multi-Class Prediction: Predicts native subcellular localization in 18 classes for eukaryotes, six classes for bacteria, and three classes for archaea.
• Machine Learning Approach: Employs a profile-kernel support vector machine that implements a cascading prediction mechanism to model cellular sorting and maintain accuracy for protein fragments.
• Homology-Based Inference: Incorporates homology-based inference to enhance prediction reliability and accuracy across organisms.
• Performance Metrics: On sequence-unique data achieves Q18 = 80±3% for eukaryotes and Q6 = 89±4% for bacteria.
• Input Scale and Throughput: Handles inputs from single protein sequences to entire proteomes, with reported runtimes of ~90 seconds for a 300-residue eukaryotic protein and a few hours for a full eukaryotic proteome, excluding alignment generation.
• Precomputed Predictions: Provides precomputed localization predictions for over 1000 fully sequenced organisms.

Scientific Applications:

• Functional Annotation: Supports assignment of protein function by mapping proteins to cellular compartments.
• Cellular Process Analysis: Enables investigation of cellular processes through compartment-specific protein distributions.
• Disease Mechanism Studies: Aids exploration of diseases associated with protein mislocalization.
• High-Throughput and Proteomics: Suitable for high-throughput studies and genome-scale proteomic analyses.

Methodology:

Combines a profile-kernel support vector machine implementing a cascading prediction architecture with homology-based inference; sequence alignment generation is handled separately.