ADP-Fuse

ADP-Fuse predicts and classifies antidiabetic peptides (ADPs) to identify peptides that target β-cells or T-cells and modulate insulin production.

Key Features:

• Two-Layer Prediction Framework: Employs a two-layer machine-learning framework that integrates multiple feature descriptors and classifiers to improve prediction accuracy.
• 22 Peptide Sequence-Derived Features: Evaluates 22 peptide sequence-derived features for comprehensive sequence characterization.
• Multiview Information Integration: Embeds multiview information into single-feature models to leverage diverse data perspectives.
• Machine Learning Evaluation and Optimization: Assesses eight notable machine-learning algorithms and selects optimal feature descriptors and classifiers for each predictive layer.
• SHAP Analysis: Applies SHapley Additive exPlanation (SHAP) to quantify individual feature contributions to prediction outcomes.
• Validation: Uses comprehensive cross-validation and independent testing to assess and validate model performance.

Scientific Applications:

• Peptide Discovery: Accelerates identification of candidate antidiabetic peptides for therapeutic development.
• Type Differentiation: Categorizes peptides into type 1 and type 2 ADPs to inform targeted strategies.
• Functional and Target Research: Supports analysis of peptide functionalities and interactions with β-cells and T-cells relevant to insulin modulation.

Methodology:

Evaluates 22 peptide sequence-derived features using eight machine-learning algorithms, selects the most effective feature descriptors and classifiers for each predictive layer, embeds multiview information into single-feature models, applies SHAP for feature contribution analysis, and validates models via comprehensive cross-validation and independent testing.