Demystifying gut bacteria with AI

Gut bacteria are known to be a key factor in many health-related concerns. However, the number and variety of them is vast, as are the ways in which they interact with the body’s chemistry and each other. For the first time, researchers from the University of Tokyo used a special kind of artificial intelligence called a Bayesian neural network to probe a dataset on gut bacteria in order to find relationships that current analytical tools could not reliably identify.

The human body comprises about 30 trillion to 40 trillion cells, but your intestines contain about 100 trillion gut bacteria. Technically, you’re carrying around more cells that aren’t you than are. Food for thought. And speaking of food, these gut bacteria are of course responsible for some aspects of digestion, though what’s surprising to some is how they can relate to many other aspects of human health as well. The bacteria are incredibly varied and also produce and modify a bewildering number of different chemicals called metabolites. These act like molecular messengers, permeating your body, affecting everything from your immune system and metabolism to your brain function and mood. Needless to say, there’s much to gain by understanding gut bacteria.

“The problem is that we’re only beginning to understand which bacteria produce which human metabolites and how these relationships change in different diseases,” said Project Researcher Tung Dang from the Tsunoda lab in the Department of Biological Sciences. “By accurately mapping these bacteria-chemical relationships, we could potentially develop personalized treatments. Imagine being able to grow a specific bacterium to produce beneficial human metabolites or designing targeted therapies that modify these metabolites to treat diseases.”

This sounds good, so what’s the problem? As mentioned, there are uncountably many and varied bacteria and metabolites, and therefore far more relationships between these things. Gathering data on this alone is a monumental undertaking but unpicking that data to find interesting patterns that might betray some useful function is even more so. To do this, Dang and his team decided to explore the use of state-of-the art artificial intelligence (AI) tools.

“Our system, VBayesMM, automatically distinguishes the key players that significantly influence metabolites from the vast background of less relevant microbes, while also acknowledging uncertainty about the predicted relationships, rather than providing overconfident but potentially wrong answers,” said Dang. “When tested on real data from sleep disorder, obesity and cancer studies, our approach consistently outperformed existing methods and identified specific bacterial families that align with known biological processes, giving confidence that it discovers real biological relationships rather than meaningless statistical patterns.”

As VBayesMM can handle and communicate issues of uncertainty, it gives researchers more confidence than a tool which does not. Even though the system is optimized to cope with heavy analytical workloads, mining such huge datasets still comes with high computational cost; however, as time goes on, this will become less and less of a barrier to those wishing to use it. Other limitations at present include that the system benefits from having more data about the gut bacteria than the metabolites they produce; when there’s insufficient bacteria data, the accuracy drops. Also, VBayesMM assumes the microbes act independently, but in reality, gut bacteria interact in an incredibly complex number of ways.

“We plan to work with more comprehensive chemical datasets that capture the complete range of bacterial products, though this creates new challenges in determining whether chemicals come from bacteria, the human body or external sources like diet,” said Dang. “We also aim to make VBayesMM more robust when analyzing diverse patient populations, incorporating bacterial ‘family tree’ relationships to make better predictions, and further reducing the computational time needed for analysis. For clinical applications, the ultimate goal is identifying specific bacterial targets for treatments or dietary interventions that could actually help patients, moving from basic research toward practical medical applications.”

###

Journal: Tung Dang, Artem Lysenko, Keith A. Boroevich and Tatsuhiko Tsunoda, “VBayesMM: Variational Bayesian neural network to prioritize important relationships of high-dimensional microbiome multiomics data”, Briefings in Bioinformatics, DOI: 10.1093/bib/bbaf300

Funding: This work was partly supported by JSPS KAKENHI Grant Numbers JP20H03240 and JP24K15175, and JST CREST Grant Number JPMJCR2231.

Useful links:

Graduate School of Science - https://www.s.u-tokyo.ac.jp/en/

Tsunoda Lab - https://mesm.bs.s.u-tokyo.ac.jp/index_Eng.html

Research Contacts:

Professor Tatsuhiko Tsunoda

Department of Biological Sciences, The University of Tokyo,

7-3-1 Hongo, Bunkyo-ku, Tokyo 113-0033, JAPAN
tsunoda@bs.s.u-tokyo.ac.jp

Press contact:
Mr. Rohan Mehra
Public Relations Group, The University of Tokyo,
7-3-1 Hongo, Bunkyo-ku, Tokyo, 113-8656, Japan
press-releases.adm@gs.mail.u-tokyo.ac.jp
 

About The University of Tokyo:

The University of Tokyo is Japan's leading university and one of the world's top research universities. The vast research output of some 6,000 researchers is published in the world's top journals across the arts and sciences. Our vibrant student body of around 15,000 undergraduate and 15,000 graduate students includes over 5,000 international students. Find out more at www.u-tokyo.ac.jp/en/ or follow us on X (formerly Twitter) at @UTokyo_News_en.