Machine Learning Reveals Missing Edges and Putative Interaction Mechanisms in Microbial Ecosystem Networks

Abstract

Different organisms in a microbial community may drastically affect each other’s growth phenotypes, significantly affecting the community dynamics, with important implications for human and environmental health. Novel culturing methods and the decreasing costs of sequencing will gradually enable high-throughput measurements of pairwise interactions in systematic coculturing studies. However, a thorough characterization of all interactions that occur within a microbial community is greatly limited both by the combinatorial complexity of possible assortments and by the limited biological insight that interaction measurements typically provide without laborious specific follow-ups. Here, we show how a simple and flexible formal representation of microbial pairs can be used for the classification of interactions via machine learning. The approach we propose predicts with high accuracy the outcome of yet-to-be performed experiments and generates testable hypotheses about the mechanisms of specific interactions.

Phylogenetic Responses of Marine Free-Living Bacterial Community to Phaeocystis globosa Bloom in Beibu Gulf, China

“…This evidence suggests that the application of bacterial bioindicators could systematically reflect and be used to record phytoplankton blooms. Random forest analysis is a good approach to investigate a highly relevant set of microbial biomarkers and for classification purposes (Dimucci et al, 2018). Our results indicated that Bacteriovorax, Marinobacter, Glaciecola, Sulfitobacter, Lokanella, Aquiluna, and Roseivirga were potential bacterial bioindicators for forecasting P. globosa blooms, with higher Gini values and significant correlations with P. globosa density after blooms (Figure 5).…”

Machine and deep learning meet genome-scale metabolic modeling

“…The algorithm seeks to determine the EFMs that differ the most in terms of time evolution. Expanding the analysis of fluxes to an ecological scale, DiMucci and colleagues developed an approach to predict interactions among bacterial species starting from temporal simulations of cocultures through dynamic FBA (dFBA) [72]. A random forest classifier was trained on binary vectors representing the exchange reactions in each GSMM, using dFBA relative yield predictions of cocultures with respect to independent cultures.…”

“…From this point of view, the generation of flux data from a condition-specific GSMM can be regarded as an elaborate but transparent feature engineering step, in which a fluxome is the result of combining available omics with expert knowledge and mathematical optimization. Therefore, we believe that CBM could be the key to building more interpretable machine learning models-for instance, by providing variables of clear meaning [68,72].…”

Dynamic simulations of microbial communities under perturbations: opportunities for microbiome engineering

“…Bauer et al [32] and DiMucci et al [42] apply two dynamic GEM modeling methods -BacArena [43] and COMETS [44], respectively; however, they nevertheless analyse only static data by only taking into account the final time point. BacArena, using an agent based approach, is able to manage hundreds of GEMs, while COMETS is limited to only a few.…”

“…Apart from biomass and metabolite concentration, solving GEMs with mathematical methods (FBA) also returns data about fluxes through metabolic reactions, known as "fluxomics" data. Simulated fluxomics data from GEM simulations have been proposed as the input to supervised Machine Learning (ML) classifiers to investigate, for example, the prediction of ecological niches (endosphere or rhizosphere) for isolated microbes (one GEM) [61] or the prediction of ecological interactions between pairs of microbes in communities (multiple GEMs) [42]. Similarly, unsupervised ML methods (mainly clustering) have been applied to simulated fluxomics data from GEMs.…”