Title:	Can microbes distinguish friend from foe?
DNr:	Berzelius-2025-10
Project Type:	LiU Berzelius
Principal Investigator:	Laura Carroll <laura.carroll@umu.se>
Affiliation:	Umeå universitet
Duration:	2025-02-18 – 2025-09-01
Classification:	10610
Keywords:

Abstract

Microbes live in complex social environments where interactions range along a spectrum from competition to cooperation. Even interactions between the same partners can shift as environmental resources change. Within this context, there can be significant fitness benefits for microbes that correctly predict or infer the type of interaction(s) and express an appropriate behavior. How microbes approach this inference problem ultimately determines their interactions and the resulting ecological structure of the community---and yet, much is unknown about the nature of this inference problem. To this end, we will introduce a novel interpretable deep learning approach for microbial metabolic models, which utilizes an improved batch ensemble and interpretation inspired by coalition game theory. Our method treats microbes (in the form of metabolic models) as agents playing in a particular environment and lets them learn a beneficial policy, which would underline whether agents should cooperate or compete. To interpret the corollaries, we will adapt Shapley values for the reinforcement learning explanation framework. We will also train a generative model to produce fake information that could affect the oracle and indicate a case of fake strategy response. In addition, we will use metabolic simulations for our framework’s final validation. This project is expected to answer the following questions: -What is the most optimal strategy for an agent given a certain environment? -How do intracellular and extracellular compounds define the agent’s strategy? -What microbial flux will maintain the stability of the system (e.g. ensuring the agents grow after initialization) and by what margin? -Which pairs of microbes are more likely to compete/cooperate? -How does the phenomenon of switching from competitive to cooperative behavior occur? This project will help us understand how microorganisms adjust their behaviors based on what they can detect around them. Further, results from this project can be leveraged to create well-functioning, cooperative communities of microbes, e.g., within the realms of synthetic biology and microbiome engineering.