A Computational-Based Method for Predicting Drug-Target Interactions by Using Stacked Autoencoder Deep Neural Network

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Accurate drug-target interaction (DTI) prediction is crucial for efficient drug development.
• Traditional experimental methods for DTI screening are time-consuming, costly, and prone to errors.

Purpose of the Study:

• To develop an effective computational method for predicting DTIs using drug molecular structure and protein sequences.
• To leverage deep learning for automated feature extraction and accurate DTI prediction.

Main Methods:

• Utilized a stacked autoencoder deep learning model to extract information from raw drug and protein data.
• Generated representative features by combining molecular substructure fingerprints and protein sequence information.
• Employed a rotation forest classifier for accurate DTI prediction.

Main Results:

• Achieved high accuracy in predicting DTIs on gold standard datasets: enzymes (0.9414), ion channels (0.9116), G-protein-coupled receptors (GPCRs) (0.8669), and nuclear receptors (0.8056).
• Demonstrated superior performance compared to other feature extraction algorithms and state-of-the-art classifiers.
• Validated the method's competitiveness in predicting drug-target interactions.

Conclusions:

• The proposed deep learning method offers a highly accurate and efficient approach for predicting drug-target interactions.
• This computational strategy can significantly aid in accelerating the drug discovery pipeline.
• The method shows strong potential for identifying novel drug-target relationships with reduced experimental burden.