MSGCA: Drug-Disease Associations Prediction Based on Multi-Similarities Graph Convolutional Autoencoder.

Area of Science:

• Bioinformatics
• Computational Biology
• Pharmacology

Background:

• Identifying drug-disease associations (DDAs) is crucial for drug development but traditional methods are costly and inefficient.
• Existing computational methods for DDA prediction often suffer from performance limitations due to the sparsity of the initial DDAs matrix.
• There is a need for advanced computational approaches to enhance the accuracy and efficiency of DDA prediction.

Purpose of the Study:

• To propose a novel computational method, the multi-similarities graph convolutional autoencoder (MSGCA), for accurate drug-disease association prediction.
• To overcome the challenge of data sparsity in DDAs prediction by integrating multiple similarity measures and reconstructing the DDAs matrix.
• To validate the effectiveness of the proposed MSGCA method against existing approaches using multiple datasets.

Main Methods:

• Integration of multiple drug and disease similarities using centered kernel alignment-based multiple kernel learning (CKA-MKL).
• Improvement of integrated similarities via linear neighborhood and reconstruction of the DDAs matrix using weighted K-nearest neighbor profiles.
• Construction of a heterogeneous network incorporating reconstructed DDAs and enhanced similarities, followed by prediction using a graph convolutional autoencoder with an attention mechanism.

Main Results:

• The MSGCA method demonstrated superior performance compared to existing DDA prediction methods across three independent datasets.
• The integration of multiple similarities and the graph convolutional autoencoder approach effectively addressed the issue of data sparsity.
• Case studies provided further evidence supporting the reliability and accuracy of the MSGCA model for predicting drug-disease associations.

Conclusions:

• The proposed MSGCA method offers a significant advancement in computational drug-disease association prediction by effectively handling data sparsity.
• MSGCA provides a more accurate and reliable approach for identifying potential drug-disease relationships, aiding in drug discovery and development.
• The study highlights the potential of integrating multiple similarity measures and advanced deep learning techniques for complex biological network analysis.