Predicting Drug-Disease Associations via Using Gaussian Interaction Profile and Kernel-Based Autoencoder.

Area of Science:

• Computational biology
• Pharmacology
• Machine learning

Background:

• Drug repositioning identifies new therapeutic uses for existing drugs, reducing development costs and timelines.
• Accurate prediction of drug-disease associations is crucial for successful drug repositioning strategies.
• Machine learning approaches are increasingly utilized for predicting drug-disease relationships.

Purpose of the Study:

• To propose a novel feature learning method, Gaussian interaction profile kernel and autoencoder (GIPAE), for predicting drug-disease associations.
• To enhance computational efficiency by incorporating batch normalization and fully-connected layers.
• To evaluate the performance and reliability of the GIPAE model.

Main Methods:

• Developed a novel feature learning method named GIPAE, integrating Gaussian interaction profile kernel and autoencoder.
• Implemented batch normalization and fully-connected layers to reduce computational complexity and training time.
• Conducted 10-fold cross-validation on Fdataset and Cdataset to assess predictive performance.

Main Results:

• GIPAE achieved high performance with AUCs of 93.30% on Fdataset and 96.03% on Cdataset, outperforming previous computational models.
• Case studies on obesity and Alzheimer's disease showed significant validation of predicted drug-disease associations in the CTD database.
• The model successfully identified 14 obesity-related and 11 Alzheimer's disease-related drugs among top predictions.

Conclusions:

• GIPAE is a reliable and accurate computational model for predicting drug-disease associations.
• The proposed method offers a significant advancement in computational drug repositioning.
• GIPAE's validated predictions can accelerate the identification of new therapeutic indications for existing drugs.