GM-Pep

GM-Pep generates and evaluates peptide sequences using a Conditional Variational Autoencoder (CVAE) and a Deep-Multiclassifier to support discovery and prioritization of therapeutic peptides.

Key Features:

• Conditional Variational Autoencoder (CVAE): Generates novel, drug-like peptide sequences by sampling from a learned latent space using a conditional variational autoencoder architecture.
• Deep-Multiclassifier: Predicts bioactivity probabilities for generated peptides using a deep multiclass classification architecture.
• Multi-activity prediction: Produces simultaneous probability predictions for toxicity, antifungal, antihypertensive, and antibacterial activities.
• Training data: The Deep-Multiclassifier is trained on positive samples to evaluate multiple biological activities concurrently.
• Performance metrics: Reported overall accuracy is 96.41%, with class-specific accuracies of 94.48% (toxicity), 96.58% (antifungal), 97.18% (antihypertensive), and 96.91% (antibacterial).
• Empirical validation: Performance was evaluated using 12 synthesized antibacterial peptides compared against random peptide sequences.

Scientific Applications:

• Therapeutic peptide discovery: Supports generation and prioritization of candidate peptides for drug development.
• Early-stage candidate evaluation: Enables comprehensive evaluation of multiple bioactivities and toxicity risk during early discovery.
• Multi-activity profiling: Facilitates identification of peptides with desired antibacterial, antifungal, or antihypertensive properties while screening for toxicity.
• Antibacterial peptide validation: Informs selection of antibacterial candidates using empirical comparison to random sequences.

Methodology:

Conditional Variational Autoencoder (CVAE) for peptide sequence generation; Deep-Multiclassifier trained on positive samples to predict probabilities for toxicity, antifungal, antihypertensive, and antibacterial activities; combined generative and predictive modeling for candidate identification.