AOPEDF

AOPEDF predicts drug-target interactions by embedding arbitrary-order proximities from heterogeneous biological networks and classifying them with a cascade deep forest to support drug repurposing and side-effect analysis.

Key Features:

• Network Integration: Integrates 15 diverse networks encompassing chemical, genomic, phenotypic, and network profiles linking drugs, proteins (targets), and diseases into a heterogeneous biological network.
• Multi-omics and Systems Biology Integration: Leverages multiomics technologies and systems biology concepts to incorporate diverse biological data types into the network representation.
• Arbitrary-Order Proximity Embedding: Learns low-dimensional vector representations that preserve arbitrary-order proximities to capture complex relationships within the integrated network.
• Cascade Deep Forest Classifier: Uses a cascade deep forest machine-learning model to predict new drug-target interactions from the embedded features.
• Performance Evaluation: Reports AUROC values of 0.868 on a DrugCentral validation set and 0.768 on ChEMBL for systematic assessment of predictive performance.

Scientific Applications:

• Drug Repurposing: Identifies potential new targets for existing drugs to inform repurposing efforts.
• Side Effect Analysis: Predicts off-target interactions that can help explain unexpected adverse drug reactions.
• Mechanism-of-Action Studies: Aids elucidation of drug mechanisms, illustrated by case studies involving aripiprazole, risperidone, and haloperidol in substance abuse disorder contexts.

Methodology:

Constructs a heterogeneous network integrating 15 diverse networks; learns feature representations via arbitrary-order proximity embedding; applies a cascade deep forest classifier to predict DTIs; and evaluates performance using AUROC on DrugCentral and ChEMBL.