Philius

Philius predicts protein transmembrane topology and signal peptides using dynamic Bayesian networks to model membrane-spanning regions and signal peptide signals.

Key Features:

• Dynamic Bayesian Networks: Philius employs DBNs, building on the framework of Hidden Markov models (HMMs), to model sequence states for topology and signal peptides.
• Two-Stage Decoder: The method uses a two-stage DBN decoder that combines posterior decoding with Viterbi-style grammar constraints to balance flexibility and consistency in predictions.
• Integrated Submodels: Philius integrates a signal peptide submodel with a transmembrane submodel within its DBN framework to jointly analyze signal peptides and transmembrane segments.
• Confidence Metrics: The software provides confidence scores for protein type, segment, and topology predictions that correlate with observed precision.
• Performance Metrics: Philius reports a 13% improvement in transmembrane topology prediction over Phobius and achieves sensitivity 0.96 and specificity 0.96 for signal peptide detection.

Scientific Applications:

• Transmembrane Protein Topology Prediction: Predicts transmembrane topology with reported 13% improvement over Phobius.
• Signal Peptide Detection: Detects signal peptides with reported sensitivity 0.96 and specificity 0.96.
• Large-Scale Proteome Annotation: Applied to predict transmembrane topology and signal peptides for all 6.3 million proteins in the Yeast Resource Center (YRC) database to assess prevalence of signal peptides and transmembrane segments.

Methodology:

Philius uses dynamic Bayesian networks with a two-stage decoder that combines posterior decoding and Viterbi-style grammar constraints and integrates signal peptide and transmembrane submodels, building on the HMM-based model Phobius.