How far can microbial monocultures predict growth in multi-strain communities?

Abstract

Previous research in microbes successfully predicted biculture growth based on monoculture growth curves. Still, the usual model-based approach does not seem to extend to communities involving more than two bacterial strains. Here, we use a model-blind machine-learning approach to predict community-wide yield, area under the growth curve, maximum relative growth rate and its timing in communities involving up to five strains of Escherichia coli. First, we identify the highest per-capita growth rate and its timing in monocultures as major predictors of multi-strain community growth. Next, we show that a random forest trained on communities involving a low number of strains is able to predict the outcomes of communities involving a higher number of strains. Finally, we reveal diminishing returns in using more and more complex communities in order to predict the behaviour of a higher-level community, because monoculture- and biculture-based growth predictions are often already accurate. This finding relativises the need for experiments involving high numbers of strains when studying the growth dynamics of multi-strain communities.