A model for predicting drug-disease associations based on dense convolutional attention network.

Area of Science:

• Computational biology
• Pharmacogenomics
• Bioinformatics

Background:

• Drug discovery is lengthy and costly, necessitating computational approaches for identifying new therapeutic uses of existing drugs.
• Drug-disease association prediction is crucial for drug repurposing but existing methods overlook valuable prior knowledge and detailed network features.

Purpose of the Study:

• To propose a novel deep learning model, DCNN, for enhanced drug-disease association prediction.
• To integrate diverse similarity measures and advanced neural network architectures for improved predictive accuracy.

Main Methods:

• Developed DCNN, a deep learning model incorporating Gaussian interaction profile kernel similarity, drug structural similarity, and disease semantic similarity.
• Employed Dense Convolutional Neural Network (DenseCNN) for feature extraction and a Convolutional Block Attention Module (CBAM) for adaptive feature optimization.

Main Results:

• DCNN demonstrated superior performance compared to existing drug-disease association predictors in ten-fold cross-validation.
• Case study results confirmed the model's effectiveness in accurately predicting drug-disease associations.

Conclusions:

• The proposed DCNN model offers a significant advancement in computational drug-disease association prediction.
• Integrating multiple similarity metrics and attention mechanisms enhances the identification of potential therapeutic relationships.