PNNL

Abstract

Microorganisms in terrestrial habitats self-organize to form specific spatial patterns through their interactions with each other and with the environment. Spatial patterns of microorganisms can therefore be viewed as a key ecological phenotype. However, their use for predicting microbial interactions is currently lacking, primarily due to an intrinsic limitation of current network inference methods that are based on population-level analyses. To address this gap, here we propose supervised deep learning as a new tool for predicting interspecies interactions from spatial patterns of microbial assembly. In a case study of two interacting organisms, we developed an agent-based model to simulate their spatiotemporal evolution under diverse growth and interaction scenarios. These data were subsequently used as a primary source to train, validate, and test deep neural networks. To do so, we first tested the effectiveness of the deep learning method for small-size domains (100 µm × 100 µm) over which interaction coefficients are assumed to be invariant. We obtained fairly accurate predictions, as indicated by an average R2 value of 0.84. We further tested how deep learning models can be extended to relatively larger domains (450 µm × 450 µm) where interaction coefficients are varying in space. Notably, without any additional training, our deep neural network model was able to successfully recover the spatial distribution of interaction coefficients. Lastly, we evaluated the applicability of our computational approach to real biological data through imaging experiments of binary interactions between Pseudomonas fluorescens and a mutant of Escherichia coli. These co-culture experiments were designed to control interactions as a function of environmental conditions. For example, P. fluorescens and E. coli develop a degrader-cheater relationship when growing on chitin polymers, which however turns into competition when growing on chitin hydrolysis products (i.e., N-acetyl glucosamine). Consistent with our expectations, the analysis of real images using the deep neural network model correctly predicted the dramatic shifts in interactions of the two organisms in the two different environmental contexts mentioned above. The combined use of the agent-based model and machine learning algorithm developed in this work successfully demonstrates how to infer microbial interactions from spatially distributed data, presenting itself as a useful tool for the analysis of more complex microbial community interactions.