ATSE

ATSE predicts peptide toxicity by integrating structural and evolutionary information using graph neural networks and an attention mechanism to classify peptides as toxic or non-toxic.

Key Features:

• Sequence Processing Module: Transforms peptide sequences into molecular graphs and generates evolutionary profiles reflecting biological history.
• Feature Extraction Module: Applies graph neural networks (GNNs) to extract discriminative features from structural and evolutionary data.
• Attention Module: Employs an attention mechanism to prioritize and optimize the most relevant features for prediction.
• Output Module: Classifies peptides as toxic or non-toxic using the optimized feature representations.
• Integrated structural and evolutionary representation: Combines molecular graph-based structural information with evolutionary profiles as complementary inputs.
• Interpretable, data-driven features: Learns features that are data-driven and interpretable, enabling visualization and further analysis.
• Validation: Performance has been evaluated through comparative studies showing superior performance over existing methods.

Scientific Applications:

• Toxicity screening: Assess potential toxicity among peptide candidate libraries.
• Peptide therapeutic development: Support selection and safety assessment of peptide-based therapeutics.
• Regulatory and safety assessment: Inform safety evaluation and risk assessment of peptide compounds.

Methodology:

Peptide sequences are converted into molecular graphs and evolutionary profiles, features are extracted via graph neural networks, an attention mechanism refines these features, and a classifier predicts toxic versus non-toxic labels.