Predictive model uses gut microbes to forecast human diseases, health outcomes

Predictive model uses gut microbes to forecast human diseases, health outcomes
Graphical abstract. Credit: Journal of Molecular Biology (2022). DOI: 10.1016/j.jmb.2022.167693

A new approach that uses artificial intelligence (AI) shows how to use microorganisms in the body and molecules in cells to predict human health outcomes, according to Penn State College of Medicine and University of Texas Southwestern Medical Center researchers. They say it could improve the accuracy of predicting the development of human diseases, such as inflammatory bowel disease and diabetes.

The is made up of trillions of microorganisms, such as fungi and bacteria that live in the body, usually in the gut, and impact overall health. These organisms, along with the metabolome—or the molecules found within cells and tissues—have an important impact on .

Published in the Journal of Molecular Biology, the present study proposes to learn useful features from datasets that measure both the microbiome and the metabolome and use them to substantially improve the risk prediction accuracy in datasets only measuring the microbiome. The results present a and AI-based, non-invasive approach using the that could identify individuals with an elevated risk for diseases.

Up until now, due to cost constraints, only a handful of studies measured both microbiome and metabolome data. Most studies only measured microbiome data without including data on metabolomes, which limited their usefulness for predicting disease risks. According to the researchers, combining the microbiome and metabolome together can help to more accurately predict disease outcomes and lead to a better understanding of the disease mechanisms.

"Deep-learning-based, non-invasive approaches have tremendous potential to improve the diagnosis and risk prediction for human diseases," said co-lead author Dajiang Liu, professor and vice chair for research of public health sciences and biochemistry and , and interim director of Penn State College of Medicine's AI initiative. "Combined with high-throughput technologies, such as DNA sequencing, it offers a cost-effective approach that helps identify at-risk patients and fast-forwards precision medicine."

The scientists proposed a novel integrative modeling framework called Microbiome-based Supervised Contrastive Learning Framework (MB-SupCon). Implementing the new method, they studied gut microbiome and metabolome data in stool samples from 720 patients to predict factors associated with Type 2 diabetes.

According to the researchers, MB-SupCon outperformed existing machine learning methods and proved highly accurate for predicting patients' insulin resistance status (84%), gender (78%) and race (80%).

When investigators used MB-SupCon in a large study, they observed similar advantages. According to the researchers, this non-invasive, cost-effective method could be broadly used to help predict in a variety of disease studies.

"The human microbiome is a major modifiable risk factor for human diseases," said co-lead author Xiaowei Zhan, a member of the University of Texas Southwestern Medical Center. "Our approach helps identify bacteria that influence disease risk. Modifying these bacteria can be a valuable new approach to treat human disorders that were not easily treatable before."

More information: Sen Yang et al, MB-SupCon: Microbiome-based Predictive Models via Supervised Contrastive Learning, Journal of Molecular Biology (2022). DOI: 10.1016/j.jmb.2022.167693

Journal information: Journal of Molecular Biology

Citation: Predictive model uses gut microbes to forecast human diseases, health outcomes (2022, July 20) retrieved 22 March 2023 from
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.

Explore further

Predicting cancer immunotherapy response from gut microbiomes using machine learning models


Feedback to editors