Deep Learning Model Approach to Predict Diabetes Type 2 Based on Clinical, Biochemical, and Gut Microbiota Profiles

Abstract

Type 2 diabetes mellitus (T2DM) is a chronic metabolic disease. Gut microbiota plays a key role in metabolic homeostasis and the development of T2DM and its complications. With the advance of artificial intelligence (AI), it is possible to develop novel models based on machine learning (ML) that can predict the risk of developing certain diseases and facilitate their early diagnosis, or even take preventive measures in advance. This can be the case of T2DM, for example. Our objective was to develop a predictive model of the risk of developing T2DM based on clinical, biochemical, and intestinal microbiota parameters, which estimates the time margin for developing this disease. To this end, a Deep Learning Multilayer Perceptron (MLP) algorithm was developed and trained with data from real patients from a current large population epidemiological study. The data were normalised and augmented to increase their diversity and avoid overfitting. The neural network developed was optimised, and the best hyperparameters were chosen for model building by Bayesian optimisation. We succeeded in getting the model to return a numerical result corresponding to the number of months it will take for a particular individual to develop T2DM with an accuracy of 95.2%.