Loc3D

Loc3D predicts the subcellular localization of eukaryotic proteins from their three-dimensional (3D) structures to support structural genomics and functional annotation.

Key Features:

• Multiple prediction methods: Implements PredictNLS to identify nuclear localization signals, LOChom to infer localization by sequence homology, LOCkey to analyze SWISS-PROT keywords via automatic text analysis, and LOC3Dini which uses neural networks and vector support machines for ab initio predictions.
• Database content: Contains over 8,700 eukaryotic protein chains sourced from the Protein Data Bank (PDB).
• Prediction from predicted structures: Supports localization prediction for proteins with predicted structures derived from threading servers.
• Hierarchical LOCtree system: Integrates support vector machines (SVMs) and other prediction methods using sequence and predicted structural features to mimic cellular sorting mechanisms.
• Performance metrics: LOCtree accuracy reported as 74% for non-plant eukaryotes, 70% for plants, and 84% for prokaryotes.
• Specialized systems for structure status: Provides separate systems tailored for proteins of known and unknown structures to address limitations when applied to PDB sequences.
• Motif and evolutionary analysis: Leverages evolutionary information from multiple alignments, motif analysis, and protein structure aspects to improve localization predictions, with improved accuracy for extracellular and nuclear proteins and lower effectiveness for mitochondrial proteins compared to TargetP.

Scientific Applications:

• Structural genomics: Supports target selection and interpretation in structural genomics by assigning subcellular localization to proteins with experimental or modeled 3D structures.
• Functional annotation and large-scale studies: Aids functional annotation and target selection in large-scale genomic and proteomic studies by providing localization predictions for eukaryotic proteins.

Methodology:

Combines PredictNLS, LOChom (sequence homology), LOCkey (SWISS-PROT keyword automatic text analysis), LOC3Dini (neural networks and vector support machines), hierarchical LOCtree (support vector machines and other methods), multiple alignments for evolutionary information, motif analysis, protein structural feature analysis, and inputs from threading-derived predicted structures.