Relation path feature embedding based convolutional neural network method for drug discovery.

Area of Science:

• Biomedical Informatics
• Computational Drug Discovery
• Artificial Intelligence in Medicine

Background:

• Drug development is a costly and lengthy endeavor.
• Literature-based discovery offers a supplementary approach to expedite drug discovery.
• Identifying novel therapeutic agents remains a critical challenge in medicine.

Purpose of the Study:

• To develop an advanced computational model for drug discovery using biomedical literature.
• To accelerate the identification of potential drugs for treating diseases, particularly incurable ones.
• To enhance the efficiency of the drug discovery pipeline through artificial intelligence.

Main Methods:

• A convolutional neural network model with an attention mechanism, termed PACNN, was developed.
• Biomedical knowledge graphs were constructed from abstract predictions.
• Drug-disease relation path features were extracted using a path ranking algorithm and embedded for model training.

Main Results:

• The PACNN model demonstrated promising performance in drug discovery tasks.
• Experimental results showed superior performance compared to traditional random walk algorithms.
• The model effectively mined potential drugs for various diseases from literature.

Conclusions:

• The proposed PACNN model serves as an effective auxiliary tool for drug discovery.
• This approach can significantly speed up the discovery of novel therapeutics.
• The method holds potential for identifying treatments for currently incurable diseases.