Predicting drug-target interaction network using deep learning model.

Area of Science:

• Computational drug discovery
• Bioinformatics
• Machine learning in pharmacology

Background:

• Traditional drug discovery is costly and slow, driving interest in drug repurposing.
• Predicting drug-target interactions (DTIs) computationally is crucial for efficient drug repurposing.
• High-dimensional data in drug discovery challenges traditional machine learning models.

Purpose of the Study:

• To develop advanced computational models for predicting DTIs.
• To enhance drug repurposing strategies, particularly for breast cancer treatment.
• To overcome limitations of traditional machine learning in analyzing complex biological data.

Main Methods:

• Utilized drug descriptors, protein sequence, and domain data from Drugbank and NCBI.
• Developed novel similarity-based approaches for negative DTI dataset creation.
• Proposed and compared multiple LASSO (Least absolute shrinkage and selection operator) models and a LASSO-Deep Neural Network (DNN) model.

Main Results:

• The LASSO-DNN model significantly outperformed standard logistic regression, LASSO, SVM, and standard DNN models.
• LASSO models incorporating protein tripeptide composition (TC) and domain features showed superior predictive power.
• Top DTIs identified by LASSO-DNN show potential for repurposing existing drugs for breast cancer treatment.

Conclusions:

• Effective feature representation of drugs and targets is critical for accurate DTI prediction models.
• Disease-associated risk genes identified through genomic studies serve as valuable targets for drug repurposing efforts.