Predicting gut metabolites from gut microbiome and their interpretability analysis of IBD prediction based on LIME

Abstract The pathophysiology of inflammatory bowel disease (IBD) is influenced by the gut microbiome and gut metabolite, but understanding how IBD is affected remains challenging. It is crucial to understand which features affect IBD in order to effectively diagnose the disease. Traditional technology for measuring metabolite features is time-consuming and costly. The abundance of metabolite features in IBD patients is altered depending to changes in the abundance of gut microbiome. LSTM-VAE is proposed to predict gut metabolite features using gut microbiome of IBD patients. The pathogenesis of IBD is investigated by LSTM-VAE without gut metabolite data. In order to explore IBD is affected by the features, GBDT-LR is used to predict IBD disease using the gut microbiome and the generated gut metabolites. GBDT-LR achieved high-precision prediction, with an accuracy of 0.97 at the genus level and 0.95 at the species level. It is noteworthy that LIME is used to explain the prediction process of GBDT-LR, solving the prediction of the 'black box model'. The cost of measuring intestinal metabolites were reduced in this study and the researches were assisted in the diagnosis and drug research of IBD diseases. Insight Box Dysbiosis of the gut microbiota and the resulting abnormal metabolites were influenced in the IBD, promoting inflammatory responses and damaging intestinal barrier function. LSTM-VAE was proposed to predict changes in gut metabolite features in IBD patients without the need for direct measurement of costly and time-consuming metabolite data. Furthermore, high-precision prediction of IBD based on gut microbiome data was demonstrated and metabolite features were generated in the GBDT-LR, achieving accuracy rates of 0.97 at the genus level and 0.95 at the species level. Additionally, the LIME is employed to interpret the "black box" prediction process of GBDT-LR. The cost of measuring gut metabolites was reduced, but also strong support for the diagnosis and drug development of IBD was provided.