IDDF2023-ABS-0078 Integration of gut microbiome and exposome data through artificial intelligence enables accurate diagnosis of pancreatic cancer with high fidelity and interpretability

Background

Non-invasive detection of pancreatic ductal adenocarcinoma (PDAC) remains a formidable challenge, while emerging evidence suggests the role of the gut microbiome in PDAC evolution and progression. Moreover, although the human exposome can mediate the gut microbiome’s influence on host variables, it is intriguing to investigate whether the exposome plays a role in concert with the gut microbiome during the development of PDAC. In this study, we aimed to achieve state-of-the-art performance in PDAC diagnosis by integrating exposome and gut microbiome data, and exploring the potential interplay between these factors in PDAC development.

Methods

We presented a robust, accurate, and non-invasive classifier based on the combination of exposome and microbiome for PDAC diagnosis, namely Multi-Omics Graph Convolutional Networks (MOGONET) to predict PDAC (IDDF2023-ABS-0078 Figure 1a). MOGONET consists of two main components: a Graph Convolutional Network (GCN) module that learns different omics data features and a View Correlation Discovery Network (VCDN) module that integrates multi-omics data. We tested our model on a cohort of 57 PDAC patients and 50 controls.

Results

Our results demonstrated that MOGONET achieved high Accuracy (0.82±0.09), F1 score (0.84±0.08), and AUROC (0.87±0.09) in our case-control cohort. Compared to traditional machine learning methods and previous studies that predicted PDAC using gut microbiome alone, MOGONET outperformed these approaches (IDDF2023-ABS-0078 Figure 1b). Additionally, we discovered that the exposome might impact PDAC development through its complex interplay with the gut microbiome (IDDF2023-ABS-0078 Figure 1c). For example, we observed a depletion of Bacteroides coprocola and Clostridium species in samples with high exposome feature ‘alcohol consumption’, who are at an increased risk of developing PDAC. These same species were also found to be depleted in PDAC patients, indicating their potential role in resisting PDAC but having low alcohol tolerance.

Conclusions

Our study sheds light on how exposome and microbiome in concert could contribute to the development of PDAC, and demonstrates how the combination of microbiome and exposome can enable the diagnosis of PDAC with high fidelity and interpretability.