AI-Powered Microbial Mapping Paves the Way for Designer Gut Microbiomes

By Brittany Wade 

July 19, 2022 | The gut microbiome–trillions of bacteria, archaea, protists, and fungi–is often collectively referred to as a supporting organ. Its role in stimulating the immune system, metabolizing toxic substances, maintaining blood glucose and cholesterol levels, and influencing the nervous system demonstrates its significance to human health. 

Recognizing its vitality, a team of biologists and engineers from the Universities of Michigan and Wisconsin used a computerized human gut microbiome to study the dynamics between various intestinal microbial communities.  

By discerning a species' relationship to its host and each other, scientists predict how exposure to new microbes affects a patient’s propensity for disease. Additionally, understanding these relationships lays the groundwork for developing new microbial communities with designer target functions. 

Every healthy human maintains a collection of symbiotic and pathogenic gut microbes. However, in the event of an infectious illness, pathogenic microorganisms outnumber their symbiotic counterparts. Similarly, the microbiome becomes unbalanced–a condition called dysbiosis–after a course of relatively strong antibiotics or the sudden dominance of a new species.  

Recently, microbial studies have become attractive to interdisciplinary research groups. As probiotics become all the rage in consumer markets, challenges arise in determining how incoming microorganisms will affect an individual’s existing microbial community and overall health. In other words, not every probiotic is precision medicine. 

A New Microbial Model 

Machine learning is a subset of artificial intelligence that makes predictions based on data-extrapolated patterns. Deep learning offers a more sophisticated approach using an algorithmic structure to mimic human organ behavior. The team’s study, published in eLife (DOI: 10.7554/eLife.73870), used a long short-term memory deep learning framework to anticipate intestinal microbial behaviors. 

This framework mapped approximately 33 million microbial communities in minutes, compared to conventional models that take hours. Furthermore, the network outperformed existing models like the generalized Lotka-Volterra–an ecological model describing the dynamics between two species–which may be too rigid and straightforward to capture the higher-order interactions and dynamic nature of such a vast ecosystem. 

"Problems of this scale required a complete overhaul in terms of how we model community behavior," said Mayank Baranwal, co-first author and Indian Institute of Technology Adjunct Professor of Systems and Control Engineering, in a press release. With deep learning, the team successfully mapped the microbiome’s complex and nuanced evolution, including the emergence of often neglected but significant species in lower concentrations. 

Designer Synthetic Microbiomes 

A key component of understanding the microbiome is deciphering its relationship to metabolites–chemical substances produced when a host or microbe breaks down food or other molecules. Human health is intricately tied to a microorganism’s role in creating and consuming metabolites, as these molecules influence energy conversion, cell signaling, and epigenetic modifications. 

"Metabolites are produced in very high concentrations in the intestines," said Ophelia Venturelli, co-corresponding author, University of Wisconsin Assistant Professor of Biochemistry, and head of an engineering laboratory studying live spatiotemporal microbial behaviors. "Some are beneficial to the host, like butyrate. Others have more complex interactions with the host and gut community."  

The team studied four health-relevant metabolites: butyrate, lactate, acetate, and succinate. Butyrate and acetate are short-chain fatty acids linked to gut and brain health, muscular function, and the prevention of chronic diseases such as cancer and bowel disorders. The brain and heart use lactate as an energy source, while the liver and kidneys convert it to glucose. Lastly, succinate is a critical component of the citric acid cycle, an intermediary of cellular respiration. 

The model accurately captured microbial growth as well as metabolite production and consumption, emulating healthy human intestines and the live microbes in Venturelli’s lab. Then, measuring microbe-microbe and microbe-metabolite interactions, the team extracted detailed and identifying behavioral information. According to the model, phyla Actinobacteria, Firmicutes, and Proteobacteria demonstrated a heavy influence in metabolite production. 

The model also predicted each species’ effect on neighboring communities. These predictions, along with the data gleaned from metabolite examination, led to the construction of synthetic microbiomes with target metabolite functions. 

"This new modeling approach, coupled with the speed at which we could test new communities in the Venturelli lab, could enable the design of useful microbial communities," said Ryan Clark, co-first author and former postdoctoral researcher in Venturelli's lab. "It was much easier to optimize for the production of multiple metabolites at once." 

By controlling metabolite concentration and function, the team hopes to influence the progression of health and mitigation of disease. For example, increasing butyrate production through designer microbiomes could improve gut and brain health while reducing the risk of chronic illnesses like ulcerative colitis and Crohn’s disease.  

Metabolite production was not the only focus. The team also discovered that while the genus Bacteroides did not play a significant role in metabolite synthesis, the model suggests that it dramatically influences community growth and broader community dynamics. Therefore, future designer communities should include Bacteroides species for optimal growth. 

Understanding the purpose of each microbe and its use of metabolites helps scientists diagnose and predict the course of many microbiome-related diseases. Given the team’s results, it may be safe to assume that designer microbiomes and their intended functions could emerge as a promising new subset of treatments.