PRED-TMBB2

PRED-TMBB2 predicts the transmembrane β-barrel topology of outer membrane proteins (OMPs) and discriminates β-barrel OMPs from water-soluble proteins to support proteome-scale identification and topology annotation.

Key Features:

• Hidden Markov Model (HMM) framework: Based on an HMM architecture optimized for the sequential organization of protein structure and refined to capture characteristics specific to β-barrel outer membrane proteins (OMPs).
• Topology prediction improvements: Incorporates a defined end state for β-barrel domain termination, distinct emission probabilities for adjacent residues within strands to account for asymmetric amino-acid distributions, and retraining with algorithms that optimize sequence labeling to better match experimentally determined structures.
• Enhanced training and dataset: Trained on a larger, non-redundant dataset that includes more recently solved atomic-resolution OMP structures and incorporates evolutionary information derived from multiple sequence alignments.
• Decoding methods: Provides newly developed decoding strategies in addition to existing decoding options for flexible prediction workflows.
• Performance metrics: Strict cross-validation reports 76% correct topology prediction success (≈7% higher than the best available predictors) with SOV of 0.9, and detection using only query sequences yields a Matthews correlation coefficient (MCC) of 0.92 with reported improvements in accuracy and speed relative to specialized predictors.
• Discrimination capability: Distinguishes β-barrel OMPs from water-soluble proteins with reported classification success rates of 88.8% for outer membrane proteins and 89.2% for water-soluble proteins.

Scientific Applications:

• Proteome-scale OMP identification: Enables whole-proteome scans (e.g., E. coli) to identify novel β-barrel outer membrane protein candidates.
• Topology annotation: Predicts transmembrane strands and global topology to support membrane protein structure–function studies.
• Membrane versus soluble protein classification: Differentiates β-barrel membrane proteins from water-soluble proteins for large-scale screening and classification.

Methodology:

PRED-TMBB2 trains a Hidden Markov Model using conditional maximum likelihood (discriminative) criteria on a curated, non-redundant set of non-homologous outer membrane proteins with known atomic-resolution structures, incorporates evolutionary information from multiple sequence alignments, uses retraining algorithms that optimize sequence labeling, and applies multiple decoding strategies.