AtbPpred

AtbPpred predicts anti-tubercular peptides (AtbPs) using a two-layer machine learning framework to identify peptide candidates active against Mycobacterium tuberculosis.

Key Features:

• Two-Layer Prediction Framework: AtbPpred employs a two-layer machine learning approach to classify anti-tubercular peptides.
• Feature Encodings: Models are developed using nine different feature encodings.
• Two-Step Feature Selection: A two-step feature selection procedure is applied in the first layer to identify optimal feature sets for each encoding.
• Probabilistic Integration: Predicted probabilities from the first-layer models are used as input features for the second-layer model.
• Machine Learning Algorithm: Extremely randomized trees (ERT) are used for both first- and second-layer models.
• Performance Metrics: Reported average accuracy is 88.3% in cross-validation and 87.3% on independent evaluation, corresponding to approximate improvements of 8.7% and 10.0% over prior methods, respectively.

Scientific Applications:

• In silico screening: Enables high-throughput computational screening and prioritization of candidate anti-tubercular peptides prior to in vitro and in vivo validation.
• Therapeutic discovery: Facilitates identification and prioritization of novel peptide-based therapies against Mycobacterium tuberculosis.
• Probability-guided selection: Provides probability estimates to guide selection of high-confidence AtbP candidates.

Methodology:

AtbPpred applies extremely randomized trees (ERT) in a two-layer framework: first-layer ERT models are trained on nine feature encodings with a two-step feature selection, and the first-layer predicted probabilities are integrated as input features to a second-layer ERT model; performance was evaluated by cross-validation and independent testing.