UC San Diego

The human microbiota exhibit a highly dynamic composition over the course of life and changes in the human gut microbiota have been associated with human health or disease. Reprogramming of the gut microbiota by interventions that counter these changes and promote long-lasting health has been an emerging topic in microbiome research. Predicting changes in the gut microbiome is therefore crucial for the nature and design of these interventions. Here, we report on a new method based on deep learning to forecast changes in the microbiome. We processed and analyzed nine time-course datasets of the human gut microbiome, identifying the main microorganisms present in these microbial communities at any given time. We then used an encoder-decoder neural network to train a model that successfully predicts the progression of the microbiome composition over time given only five time points of context data. Our results demonstrate the ability to predict the fate of the human gut microbiome into the future, providing the foundation for rational intervention design.