DTI-Voodoo

DTI-Voodoo predicts drug-target interactions by integrating molecular features, ontology-encoded phenotypic effects, and protein-protein interaction networks with graph convolutional neural networks to identify candidate drugs for specific proteins and mitigate biases in DTI dataset evaluation.

Key Features:

• Integration of molecular and phenotypic data: Combines molecular features with ontology-encoded phenotypic effects for joint analysis.
• Bottom-up and top-down approaches: Leverages both direct (bottom-up) and indirect (top-down) signals from molecular and phenotypic information.
• Protein-protein interaction networks: Incorporates PPI networks to capture broader biological context around targets.
• Graph Convolutional Neural Network: Uses graph convolutional neural networks to model and predict interactions on network-structured data.
• Bias mitigation and evaluation scheme: Employs a modified evaluation scheme to address intrinsic biases in common DTI datasets.
• Candidate identification for proteins: Prioritizes potential drug candidates for specific protein targets.

Scientific Applications:

• Drug Discovery: Predicts potential drug–protein interactions to support identification of novel therapeutic targets.
• Drug Repurposing: Identifies existing drugs with predicted interactions to enable repurposing for new protein targets.

Methodology:

Integrates molecular features, ontology-encoded phenotypic effects, and protein-protein interaction networks and applies graph convolutional neural networks using both bottom-up (direct) and top-down (indirect) approaches, together with a modified evaluation scheme to mitigate dataset biases.