Phobius (EBI)

Phobius predicts transmembrane protein topology and signal peptides using a hidden Markov model to distinguish hydrophobic transmembrane helices from signal peptides and infer N‑terminal orientation.

Key Features:

• Hidden Markov Model (HMM): Employs an HMM to represent sequence properties of signal peptides and transmembrane proteins.
• Interconnected state modeling: Models different sequence regions through interconnected HMM states that represent signal peptide segments and transmembrane helices.
• Signal peptide vs transmembrane discrimination: Distinguishes between signal peptides and transmembrane helices to reduce cross-prediction errors.
• N‑terminal orientation inference: Recognizes that a predicted signal peptide implies the mature protein N‑terminus is non-cytoplasmic.
• Performance improvement: Demonstrates reduced false classifications versus TMHMM and SignalP, lowering signal peptide false calls from 26.1% to 3.9% and transmembrane helix false calls from 19.0% to 7.7%.
• Proteome-scale applicability: Applied to whole proteomes including Homo sapiens and Escherichia coli with marked reduction in false classifications.

Scientific Applications:

• Whole-genome annotation: Improves annotation of transmembrane topology and signal peptides across complete proteomes.
• Proteome-scale prediction: Enables large-scale identification of signal peptides and transmembrane helices for proteome analyses.
• Protein sorting and localization studies: Assists studies that require accurate discrimination between secretory signals and membrane-spanning regions.
• Structural and functional inference: Provides topology information used in downstream structural modeling and functional annotation.

Methodology:

Phobius implements a hidden Markov model with interconnected states that explicitly model sequence regions of signal peptides and transmembrane proteins, using these states to discriminate features and infer N‑terminal topology.