BaCelLo

BaCelLo predicts the subcellular localization of eukaryotic proteins to support functional annotation and interpretation of proteomes.

Key Features:

• Subcellular Localization Classes: Predicts five classes: secretory pathway, cytoplasm, nucleus, mitochondrion, and chloroplast.
• Machine Learning Approach: Employs Support Vector Machines (SVMs) organized in a decision tree structure to classify protein sequences.
• Sequence and Evolutionary Information: Utilizes residue sequence information together with evolutionary data from alignment profiles.
• Sequence Composition Analysis: Examines whole-sequence composition and specific compositions at the N- and C-termini.
• Curated Training Set: Trains on a curated, non-redundant dataset and applies a balancing procedure to reduce training biases.
• Kingdom-Specific Predictors: Provides separate predictors for animals, plants, and fungi.
• Performance Metrics: Reported accuracies are 74% for animal proteins, 76% for fungal proteins, and 67% for plant proteins for the distributed classes.
• Whole Proteome Analysis: Applied to predict proteome-wide localizations for Homo sapiens, Mus musculus, Caenorhabditis elegans, Saccharomyces cerevisiae, and Arabidopsis thaliana.

Scientific Applications:

• Functional Genomics: Supports assignment of cellular compartments to proteins for genome annotation.
• Proteomics: Enables large-scale analysis of protein distribution across cellular compartments in proteomes.
• Systems Biology: Provides compartment localization data useful for modeling cellular processes and networks.
• Drug Discovery: Informs target localization considerations relevant to drug targeting and mechanism studies.
• Disease Modeling: Assists interpretation of protein mislocalization and its potential role in disease mechanisms.
• Synthetic Biology: Aids design and evaluation of protein targeting and compartment-specific constructs.

Methodology:

Prediction uses SVMs arranged in a decision tree, combining residue sequence information and evolutionary profiles from alignments, analyzing whole-sequence and N-/C-terminal compositions, trained on a curated non-redundant dataset with a balancing procedure and implemented as kingdom-specific predictors for animals, plants, and fungi.