AI analyzes entire gut microbiome to detect MS with high accuracy

Computer-based analyses of microorganisms in the digestive tract can distinguish people with multiple sclerosis (MS) from healthy individuals with high accuracy, according to a new study.

The analysis looked beyond gut bacteria to include other organisms like fungi and viruses. The findings suggest these comprehensive microbial profiles could eventually serve as a tool for noninvasive screening.

“This study suggests that multikingdom and functional gut microbiome markers can be utilized for non-invasive MS diagnosis, facilitating candidate biomarker panels for future clinical validation,” researchers wrote.

The study, “Multikingdom microbiome-based machine learning enables multiple sclerosis diagnosis,” was published in npj Biofilms and Microbiomes.

Recommended Reading

Aug. 6, 2025 by Bionews Staff

Gut Check: The Microbiome’s Role in MS

Looking beyond bacteria in the gut

The human digestive tract is home to billions of microscopic organisms, collectively known as the gut microbiome. Many of these tiny cohabitants of our bodies are bacteria — a specific category, or kingdom, of single-celled organisms. However, the gut microbiome also includes organisms from other kingdoms, including fungi, viruses, and archaea, which are unicellular organisms that resemble bacteria.

Previous studies have suggested that the gut microbiome is dysregulated in people with MS, but they have almost exclusively focused on bacteria, overlooking other organisms. Additionally, prior studies have mostly focused on the number of distinct bacteria rather than on their biological activity.

To address these gaps, researchers in China used machine learning to analyze hundreds of gut microbiome samples from people with or without MS. They examined bacteria and other kingdoms and also assessed measures of microbiome functional activity.

Machine learning is a type of artificial intelligence that involves feeding a computer a large dataset, and the computer then uses sophisticated mathematical rules to identify patterns in the data. These patterns can then be used to predict an outcome, such as whether someone has MS.

To evaluate the accuracy of their models, the researchers used a statistical measure called the area under the receiver operating characteristic curve (AUC). AUC values range from 0.5 to 1, with higher values indicating greater accuracy in distinguishing between people with and without MS.

Of note, because biological sex has been shown to influence microbiome composition, the researchers calculated separate AUCs for males and females.

Results from the main analysis showed high AUC values: 0.977 for males and 0.978 for females. These values were higher than those obtained when each kingdom or metabolic activity was examined individually.

Notably, the researchers honed in on the 30 most important features used to build their model. Using those features instead of all variables, the model achieved an accuracy of 0.99 in diagnosing MS in both men and women.

The researchers then used a separate dataset containing 154 samples to validate their model. In these analyses, the 30-feature model also performed well, with AUCs up to 0.849 for males and 0.763 in females.

“These results indicate that the models retained discriminatory ability in independent cohorts,” the researchers wrote.

In further analyses, the researchers’ model also identified relapsing-remitting MS with an accuracy on par with that seen in the main analysis (0.979 with the 30-marker panel), though it was notably poorer for progressive forms of MS (0.677).

Overall, the researchers said their results highlight the importance of considering the entire microbiome, not just bacteria, when studying MS.

“We demonstrated that diagnostic models based on multikingdom markers achieved high predictive values for MS diagnosis, highlighting the multifaceted role of the gut microbiome in MS [disease biology] and its potential for non-invasive diagnostics,” they concluded.