A Deep Learning Multi-omics Framework to Combine Microbiome and Metabolome Profiles for Disease Classification

Microbiome and metabolome contain information about host disease. Therefore, a multi-omics analysis of these data types can provide key constraints for disease classification. However, due to multi-omics data’s complex and high-dimensional nature, classical statistical methods struggle to capture the shared information between microbiome and metabolome. Deep learning represents a power framework to address this issue. We design a deep learning model for the integrated analysis of microbiome and metabolome that leverages the complementary information between the two datasets to perform a medical diagnosis of a given disease as a supervised classification task. We test our approach on six different matched microbiome/metabolome datasets, related to diverse pathologies. A comparative performance analysis shows that our proposed model called microBiome-metaBolome Network (BiBoNet) performs better than classical machine learning methods. In addition, we show that BiBoNet achieves better results than deep learning models based on individual or combined data. We highlight the importance of multi-omics integration through deep learning for improved medical diagnosis using microbiome and metabolome.