Identifying potential drug-target interactions based on ensemble deep learning.

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Drug-target interaction (DTI) prediction is crucial for drug discovery and development.
• Experimental DTI identification methods are often time-consuming and labor-intensive.
• Novel computational approaches are needed to accelerate DTI prediction.

Purpose of the Study:

• To develop and validate a novel computational method, EnGDD, for accurate drug-target interaction prediction.
• To compare EnGDD's performance against existing state-of-the-art DTI prediction methods.
• To identify potential therapeutic targets for neurodegenerative diseases using the EnGDD model.

Main Methods:

• EnGDD combines feature acquisition, dimensionality reduction, and classification using Gradient Boosting Neural Networks, Deep Neural Networks, and Deep Forest.
• Performance evaluation involved cross-validation across drug, target, and drug-target pair datasets.
• Comparative analysis was conducted against seven established DTI prediction methods.

Main Results:

• EnGDD demonstrated superior performance, achieving the best recall, accuracy, F1-score, AUC, and AUPR across multiple datasets and validation strategies.
• The method identified several novel potential drug-target interactions, including D00002 (Nadide) with hsa10935 (Mitochondrial peroxiredoxin3).
• EnGDD predicted potential therapeutic agents for Parkinson's disease (e.g., D01277 targeting hsa1813) and Alzheimer's disease (e.g., D02173 targeting hsa5743).

Conclusions:

• The EnGDD model offers a powerful and efficient approach for drug-target interaction prediction.
• EnGDD has the potential to significantly accelerate the identification of novel therapeutic strategies for various diseases.
• Further biomedical validation is warranted for the predicted drug-target interactions and therapeutic applications.