A deep learning method for predicting metabolite-disease associations via graph neural network.

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• Metabolism is crucial for life, involving the continuous exchange of substances.
• Altered metabolite concentrations are observed in various diseases.
• Traditional experimental methods for identifying metabolite-disease links are time-consuming and expensive.

Purpose of the Study:

• To develop a novel computational method for predicting potential metabolite-disease associations.
• To address the urgent need for efficient tools in biomedical research.

Main Methods:

• A heterogeneous network was constructed using known metabolite-disease associations, and metabolite-metabolite and disease-disease similarities.
• A deep learning algorithm, graph convolutional network with graph attention network (GCNAT), was employed.
• Metabolite and disease features were encoded using graph convolutional neural networks, and a graph attention layer integrated embeddings from multiple layers.

Main Results:

• GCNAT achieved an area under the receiver operating characteristic curve of 0.95 and a precision-recall curve of 0.405 in 5-fold cross-validation.
• Performance surpassed five existing state-of-the-art predictive methods.
• Case studies validated the predicted metabolite-disease correlations through experimental evidence.

Conclusions:

• GCNAT demonstrates high reliability and accuracy in predicting metabolite-disease associations.
• The developed algorithm offers a promising computational tool for accelerating biomedical research.
• GCNAT can aid in discovering novel insights into the metabolic underpinnings of diseases.