Meet MeLSI: A Machine Learning Tool to Detect Hidden Disease Signals in Gut Bacteria
Metric Learning for Statistical Inference helps researchers make sense of complex microbiome data and find the microbes behind important biological changes.
Key points
Researchers at the University of Missouri developed an AI tool called MeLSI to study data about the microbes living in the gut.
MeLSI finds specific microbes that cause biological changes and performs better than standard methods of analysis.
The tool could help scientists detect diseases earlier and develop medical treatments that target specific problems.
Andrew Corselli
Researchers at the University of Missouri are using artificial intelligence to find hidden clues among the trillions of bacteria living in the human gut. These clues include changes that could one day help scientists detect diseases earlier.
The Mizzou team combines knowledge of medicine, data science, and engineering. Together, its members developed a new machine-learning tool that can find important biological signals that traditional methods miss.
The tool is called Metric Learning for Statistical Inference, or MeLSI. It makes complex microbiome data easier to understand by finding the microbes behind important biological changes. This matters because researchers increasingly believe that changes in the microbiome may begin months or even years before symptoms appear.
“We’re starting to see that small changes in the gut microbiome may happen long before people show symptoms of brain and nervous system diseases such as Alzheimer’s,” said Ai-Ling Lin. She is a professor of biological sciences in Mizzou’s College of Arts and Science, a professor of radiology in the School of Medicine, and a NextGen Precision Health investigator. “If we can find those changes earlier, we could discover ways to reduce their effects or even reverse them.”
Traditional methods of analysis can show broad differences between the microbes found in healthy people and those found in people with diseases. MeLSI goes further by identifying the specific microbes responsible for those differences. This helps researchers better understand what is changing in the microbiome.
“Standard methods are like using the same simple ruler to measure everything,” said Nathan Bresette, the lead author and a doctoral student at Mizzou’s Institute for Data Science and Informatics. “Instead, MeLSI works more like a smart scanner. It learns to ignore background clutter and focus on what matters most.”
When the Mizzou researchers tested MeLSI using data from several studies published earlier, the tool repeatedly found meaningful patterns that traditional methods had missed.
In one study involving more than 1,100 adults, MeLSI did more than confirm that men’s and women’s gut microbiomes are different. It identified the specific bacteria that played the biggest role in those differences, giving researchers a clear starting point for further study. In another study, which compared people who ate a Western-style diet with people who ate a high-fiber diet, MeLSI detected a real effect of diet on the microbiome that standard methods had missed.
Together, these findings show that the tool developed at Mizzou could give researchers a powerful new way to study how the microbiome affects health and disease. This could eventually lead to earlier disease detection and treatments that target specific problems.
Here is an exclusive interview with Bresette for Tech Briefs, edited to make it shorter and clearer.
Tech Briefs: What was the biggest technical challenge you faced while developing MeLSI?
Bresette: The biggest challenge was bringing mathematics, statistics, and machine learning together in one method that could adapt to the microbiome. Each field has its own way of thinking. It took a lot of work to make them work together so the method could learn from the data while still producing results people could trust.
Tech Briefs: What was the biggest challenge you faced while testing MeLSI?
Bresette: We had to make sure that the patterns MeLSI found were real. Before using it on real studies, we tested it with computer-generated data where we already knew the correct answers. We also included cases where there were no real differences, to make sure the tool did not report patterns that were not actually there. Then we tested it on data from published studies to see whether it worked well with real microbiome data.
Tech Briefs: Do you have any plans for further research or next steps?
Bresette: Yes. We have already started using MeLSI in our own lab at Mizzou to find differences in the microbiomes of groups in our experiments. The tool is also free and available to other researchers through Bioconductor. We want to see how other labs use it and continue improving it based on what they discover.
Tech Briefs: What advice would you give researchers who want to turn their ideas into reality?
Bresette: Focus on a problem worth solving, develop a useful solution, and test it carefully before making it widely available. Also, share your work so other people can use it. They may discover things you would never have thought of yourself.
Tech Briefs: Is there anything else you would like to add?
Bresette: I would especially emphasize the importance of making results understandable. It is not enough to know that two groups of people have different microbiomes. Researchers need to know which bacteria are causing those differences. Understanding what is really happening in the microbiome will lead to new discoveries, and MeLSI was designed to help with that. I am also very grateful to my co-authors and to Mizzou for their support.