Convolutional Neural Network and Bidirectional Long Short-Term Memory-Based Method for Predicting Drug-Disease...

Area of Science:

• Computational biology
• Pharmacology
• Artificial intelligence in medicine

Background:

• Identifying new drug indications accelerates development and cuts costs.
• Previous methods often used shallow models, missing complex drug-disease path information.
• A need exists for advanced deep learning to predict drug-disease associations.

Purpose of the Study:

• To develop a novel deep learning method, CBPred, for predicting drug-disease associations.
• To deeply integrate drug-disease similarities, associations, and path information.
• To improve the accuracy and efficiency of discovering potential drug repurposing candidates.

Main Methods:

• Proposed CBPred, a hybrid model combining Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (BiLSTM).
• CNN framework learns representations from drug-disease similarities and associations.
• BiLSTM framework learns path representations with an attention mechanism for differential path contributions.

Main Results:

• CBPred demonstrated superior performance in predicting drug-disease associations compared to existing methods.
• The method successfully retrieved more true associations at the forefront of prediction lists.
• Case studies validated CBPred's capability in discovering potential drug-disease relationships.

Conclusions:

• CBPred offers a powerful deep learning approach for identifying novel drug-disease associations.
• The integration of diverse data types and path analysis enhances prediction accuracy.
• This method can significantly aid researchers in drug repurposing and discovery efforts.