(A-4) Construction of an effective model for engineered microbial dynamics against complex background microbiome

Once introduced in vivo, engineered microbes and microbial consortia usually face the challenge from the resident background microbiome. While empirical evidence found the dynamics of these engineered microbes can be predicted by low dimensional representations of the resident microbiome composition, there is no systematic identification and quantification of this low dimensionality. In this work, we use machine learning to guide the design of semi-mechanistic models which enable the reconstruction of target microbial population dynamics. We show that the essential information content is linearly correlated with the target population size, and can be implemented via a low-dimensional general Lotka-Volterra model.