Network-based prediction of drug-target interactions using an arbitrary-order proximity embedded deep forest

Area of Science:

• Computational biology
• Bioinformatics
• Network science

Background:

• Systematic identification of molecular targets is crucial for drug repurposing and understanding drug side effects.
• Computational prediction of drug-target interactions (DTIs) is highly desirable compared to experimental methods.
• Large-scale biological networks from multiomics and systems biology offer opportunities for network-based target identification.

Purpose of the Study:

• To present AOPEDF, a network-based computational framework for predicting DTIs.
• To leverage heterogeneous biological networks for identifying new molecular targets of known drugs.
• To develop an accurate computational approach for drug-target interaction prediction.

Main Methods:

• Developed AOPEDF (arbitrary-order proximity embedded deep forest), a network-based computational framework.
• Constructed a heterogeneous network integrating 15 networks (chemical, genomic, phenotypic, network profiles).
• Employed a cascade deep forest classifier to infer novel DTIs from the integrated network.

Main Results:

• AOPEDF achieved high accuracy in predicting molecular targets for known drugs on external validation sets (DrugCentral AUROC=0.868, ChEMBL AUROC=0.768).
• The framework outperformed several state-of-the-art DTI prediction methods.
• A case study showed AOPEDF predicted targets linked to the mechanism of action for substance abuse disorder for drugs like aripiprazole.

Conclusions:

• AOPEDF provides an accurate and effective computational framework for DTI prediction.
• The method facilitates drug repurposing and the discovery of novel drug-target interactions.
• The findings highlight the utility of integrated biological networks in drug discovery and development.