Identifying Gut Bacteria and Their Interactions using Deep Learning Based Image Analysis and Gnotobiotic Experiments
Loading...
Date
2020-02-27
Authors
Hay, Edouard
Journal Title
Journal ISSN
Volume Title
Publisher
University of Oregon
Abstract
The microbial communities of animal intestines are composed of dozens to hundreds of species and play important roles in host development, health and disease. Due to the complexity of these communities, the determinants of the microbial composition, which may include physical characteristics or biochemical interactions, remain largely unknown. Understanding the spatial structure and the effect of bacterial interactions are paramount to learning more about how these communities are formed.
In this dissertation, we develop the use of a deep convolutional neural network for identification of individual bacteria in 3D images of the intestines of larval zebrafish which contain fluorescently labeled bacteria taken using light sheet fluorescent microscopy. This network achieves human expert level accuracy and we extend its use to multiple bacterial species through transfer learning. Next we show the application of U-net in segmentation of the intestine in phase contrast microscopy images. These two techniques can be used in the future to study the spatial structure of microbes in the zebrafish intestine.
Lastly, we present an experiment in which explore bacterial interactions within larval zebrafish. We consider commensal intestinal microbes in larval zebrafish, initially raised germ-free to allow introduction of controlled combinations of 1-5 bacterial species. Using dissection and plating assays, we find strong pairwise interactions between certain bacteria. In the 4 or 5 bacterial species communities, we find weaker interactions and a much higher than expected level of coexistence suggesting that the pairwise interactions are not sufficient to predict the composition of multispecies gut communities and that higher-order interactions may dampen strong competition.
Description
Keywords
bacterial interactions, deep learning, image processing, larval zebrafish, machine learning, microbiome