Monitoring the Exposome with AI Tools
There are many environmental factors that influence our risk of disease, and so many interactions among these factors. The right AI tools can help untangle the knots.
By John Halamka, M.D., Diercks President, Mayo Clinic Platform and Paul Cerrato, MA, senior research analyst and communications specialist, Mayo Clinic Platform
Medical scientists have known for decades that human disease has two broad causes: genetics and environment. The genome describes the numerous genetic markers that influence our risks while the exposome refers to the many environmental influences. In recent years, we have gathered evidence to pinpoint the specific causative agents in both categories. And since the completion of the human Genome Project in 2003, a growing number of researchers have turned their attention to the exposome—and for good reason: research suggests that only 10% of disease risk is genetic in origin, while others estimate that 70-90% of chronic diseases are environmentally related.
One German investigator describes the human exposome as “every chemical, biological, physical, and social exposure an individual accumulates from conception onward.” That includes exposure to toxic metals, air pollution, psychosocial stressors, dietary choices, sleep patterns, household income, and much more. All these causative agents influence how well we resist infection, respond to inflammation, and succumb to generative diseases like diabetes mellitus and cancer, to name a few. The problem, however, is determining how to accurately measure all these factors, and then how do we mitigate the risk of the related diseases. Several initiatives are now underway to address the problems, including the Exposome Moonshot, The European Human Exposome Network, and The Network for Exposomics in the U.S. (NEXUS).
Thomas Hartung, with the Exposome Moonshot, says: “Deep learning can harmonize disparate data, extract patterns from millions of features, and predict the impact of complex mixtures of chemicals with speed. AI tools can transform the exposome concept into a practical framework for risk prediction, mechanistic insight, and preventive action.” More specifically, machine learning can help generate an integrated picture of the exposome by combining data from mass spectrometry, transcriptomics, geolocation data, and social media. In addition, causal inference models can be used to analyze longitudinal data and health outcome records to move “beyond simple correlations to untangle true cause-and-effect relationships and highlight the mechanistic pathways most ripe for intervention.” Hartung also advocates for real time hazard exposure using wearable-sensor networks, including air quality monitors to detect pollutants, a measure that would surely help the millions of patients suffering from respiratory disorders.
But measuring pollutants is only one piece of the puzzle. Examining the subtle physiological effects of food is equally important. Those effects go beyond the usual list of nutrients most often cited, namely vitamins, minerals, fats, carbohydrates, and protein. The Nutrition Dark Matter Library includes a long list of other food chemicals with known or suspected effects. The list includes polyphenols, alkaloids, and terpenoids, many of which have drug-like characteristics. Garlic alone contains 70 or more nutrients, including allicin, which has antimicrobial and cardioprotective properties.
Investigators interested in exploring the role of environmental factors that impact health and disease have taken a page from the genome’s “playbook.” Genome-wide association studies have tried to identify genetic variants that contribute to diseases in various populations. Exposome-wide association studies take a similar approach, searching for key environmental factors that may protect or harm the public. One research group, for instance, has found an association between lifestyle and socioeconomic status and several disorders, including allergic rhinitis, bone loss, and fibroids.
While we are only at the beginning of the AI/exposome journey, the result of the projects discussed above will no doubt have an impact on public health for years to come.
Recent Posts
By John Halamka and Paul Cerrato — Generative AI has the potential to transform patient care, but it can also wreak havoc if left unattended.
By John Halamka and Paul Cerrato — Since ChatGPT was launched in 2022, the entire world has suddenly taken notice of AI’s capabilities. AI agents use such large language models to actively reshape business, finance, and the healthcare ecosystem. Are we ready for this shift?
By John Halamka and Paul Cerrato—AI permeates our everyday lives in countless ways. It only makes sense for it to also improve the way medical school students are trained, and how clinicians in practice carry out their everyday responsibilities.