DTI-CDF

DTI-CDF predicts drug-target interactions by integrating heterogeneous graph-derived similarity features for drugs and target proteins and classifying them with a cascade deep forest (CDF) model to support target-based drug discovery.

Key Features:

• Cascade Deep Forest (CDF) Model: Uses a cascade of decision-tree ensembles arranged in multiple layers to classify potential drug-target interactions.
• Heterogeneous Graph Representation: Constructs a heterogeneous graph containing known drug-target interactions to inform feature derivation.
• Similarity-based and Hybrid Features: Extracts similarity-based features for drugs and target proteins and integrates heterogeneous graph-derived features to capture complex drug-target relationships.
• Improved Precision and Reduced False Positives: Demonstrates higher prediction precision and lower false-positive rates relative to several baseline methods.

Scientific Applications:

• Target Identification: Predicts potential protein targets for small molecules to support identification of therapeutic targets.
• Drug Repurposing: Identifies novel target interactions for existing drugs to enable repositioning opportunities.
• Lead Optimization: Characterizes interaction profiles to inform optimization of lead compounds.

Methodology:

Constructs a heterogeneous graph of known drug-target interactions, extracts similarity-based features for drugs and target proteins, inputs these features into a cascade deep forest model (multiple layers of decision trees in cascades), evaluates performance using five replicates of 10-fold cross-validation across three experimental settings (SD, ST, SP) where some drug-target information is omitted from training, compares results to random forest, XGBoost, deep neural networks and DDR, and validates 1,352 newly predicted DTIs against KEGG and DrugBank.