Discriminative HMMs

Discriminative HMMs predict protein subcellular localization from sequence by identifying compartment-specific targeting motifs using a discriminative Hidden Markov Model framework.

Key Features:

• Discriminative Motif Identification: Identifies motifs that are present in one subcellular compartment but absent in nearby compartments using a discriminative framework.
• Hierarchical Structure: Employs a hierarchical model that mirrors natural protein sorting to focus motif discovery on compartment-specific signals.
• Improved Localization Prediction: Demonstrates improved localization prediction accuracy on benchmark datasets, including yeast protein datasets.
• Conservation and Mapping of Motifs: Finds motifs that are more conserved than average protein sequence and maps identified motifs to known targeting motifs.
• Identification of Annotation Errors: Detects potential protein localization annotation errors in public databases by comparing motif-based predictions with existing annotations.

Scientific Applications:

• Proteomics: Enhances assignment of subcellular localization to proteins for proteome annotation and analysis.
• Cellular Biology: Supports investigation of protein targeting mechanisms and compartment-specific function within cells.
• Database Curation: Aids refinement of protein localization entries in public databases by highlighting discrepancies and potential annotation errors.

Methodology:

Uses discriminative Hidden Markov Models with a hierarchical structure to identify compartment-specific targeting motifs, map them to known targeting motifs, and benchmark predictions on yeast protein datasets.