Predicting activatory and inhibitory drug-target interactions based on structural compound representations and...

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Drug discovery relies on identifying drug-target interactions (DTIs).
• Existing DTI prediction methods often focus on simple interactions, limiting mechanistic insights.
• Understanding drug mechanisms of action (MoA) is crucial for drug development.

Purpose of the Study:

• To develop a novel computational method, AI-DTI, for predicting both activatory and inhibitory drug-target interactions.
• To integrate mol2vec and genetically perturbed transcriptomes for enhanced DTI prediction.
• To improve the understanding of drug mechanisms of action through accurate DTI identification.

Main Methods:

• Utilized mol2vec for drug feature representation and genetically perturbed transcriptomes for target feature representation.
• Developed a machine learning model trained on large-scale DTIs with associated MoA data.
• Employed data augmentation techniques for target feature vectors to broaden model applicability.

Main Results:

• AI-DTI significantly outperformed previous models in predicting activatory and inhibitory DTIs.
• The model demonstrated strong performance on unseen targets and high-throughput screening datasets.
• AI-DTI successfully identified approximately half of the known DTIs for COVID-19 therapeutics.

Conclusions:

• AI-DTI is a powerful computational tool for accelerating drug discovery.
• The method provides valuable insights into drug mechanisms of action.
• AI-DTI can generate testable hypotheses for novel drug development and mechanistic studies.