ZNGEA: ZINB-NMF Integrated Graph Embedding Autoencoder for Metabolite-Disease Association Identification

Metabolism, a series of complex and orderly chemical reactions within a biological organism, has a significant impact on sustaining life activities. Disease development is often linked to alterations in the types or levels of metabolites; however, traditional biological experiments typically involve repeated trials, which are both time-intensive and labor-demanding. Therefore, it is crucial to develop computational approaches for efficiently identifying potential associations between metabolites and diseases. In this paper, we propose a novel deep learning algorithm, named ZNGEA, which integrates the Zero-Inflated Negative Binomial distribution (ZINB) and Non-negative Matrix Factorization (NMF). First, we employ a nonlinear method to combine multiple similarity networks of diseases and metabolites with the goal of extracting important information from various perspectives. Subsequently, to obtain linear features and capture the global probabilistic structure of the data, we applied NMF and a ZINB-based graph convolutional autoencoder to extract features related to metabolites and diseases, respectively. Finally, we feed the extracted features into a bilinear decoder for training. The results of the experiments show that ZNGEA achieves the area under the Receiver Operating Characteristic (AUC) curve of 0.9859 and the area under the Precision-Recall curve (AUPR) of 0.9820 in 5-fold cross-validation, both surpassing existing methods. Case studies show that the majority of the newly identified links are validated, further confirming the reliability of ZNGEA in predicting relationships between metabolites and diseases. This method can become a valuable tool for biomedical research in exploring potential links between metabolites and diseases. The source code and data set are available at https://github.com/1521250466LYP/ZNGEA and https://zenodo.org/records/17394093.