BetAware-Deep

BetAware-Deep predicts the topology of transmembrane beta-barrel (TMBB) proteins in Gram-negative bacteria to support structural analysis and identification of membrane-associated drug targets.

Key Features:

• Topology Prediction: Identifies the number and orientation of membrane-spanning segments from protein sequences to assign TMBB topologies.
• Discrimination from Non-TMBB Proteins: Distinguishes TMBB proteins from non-TMBB proteins for accurate classification.
• Innovative Feature Computation: Incorporates non-canonical hydrophobic moment computations and sequence-profile weighting based on the White&Wimley hydrophobicity scale.
• Advanced Methodology: Employs a two-step computational strategy that integrates deep learning techniques with probabilistic graphical models.
• Training and Benchmarking: Trained on a dataset of 58 TMBB proteins and benchmarked on a novel set of 15 TMBB proteins, correctly predicting topologies for 10 of the 15 and outperforming two recent methods.
• Extensive Validation: Validated on datasets containing 1009 TMBB and 7571 non-TMBB proteins, achieving performance comparable to leading methods.
• Residue-level Outputs: Produces residue-level annotations and prediction probabilities as analytical outputs.

Scientific Applications:

• Drug Target Identification: Enables identification of potential membrane-associated drug targets in Gram-negative bacteria through topology information.
• Structural Biology Research: Supports structural studies of TMBBs by providing topology annotations that inform structure and function analyses.
• Complementary Computational Analysis: Complements experimental techniques by supplying computational topology predictions for analyses of bacterial outer-membrane proteins.

Methodology:

BetAware-Deep applies a two-step pipeline combining deep learning and probabilistic graphical models, incorporates non-canonical hydrophobic moment calculations and sequence-profile weighting based on the White&Wimley hydrophobicity scale, and was trained on 58 TMBB proteins with benchmarking on 15 TMBB proteins and validation on datasets of 1009 TMBB and 7571 non-TMBB proteins.