Years or even decades before the first symptoms of life-altering conditions such as cancer, Alzheimer’s disease, and autoimmune disorders emerge, immune and other changes could be occurring behind the scenes that gradually push seemingly healthy individuals toward these outcomes. Now, researchers have developed a new way to assess immune health that may one day allow clinicians to detect and even stop or reverse disease progression before it becomes too late.
In their recent study, a team of Yale and NIH researchers set out to see, by combining multimodal immune profiling and artificial intelligence, whether they could predict the probability—based on a personalized immune profile—of whether a person is healthy. Their analyses revealed a quantitative immune health metric (IHM) that could accurately detect deviations from health across diverse datasets spanning a range of diseases. Their findings could lead to the development of precision medicine tools that both assess an individual’s current health based on their immune profile and predict future health outcomes. The researchers published their findings in Nature Medicine on July 3.
“Our research points to a way forward on how to assess someone’s immune system, with implications for detecting hidden pathology to see sub-clinically how they are doing,” says John Tsang, PhD, professor of immunobiology and biomedical engineering and the study’s senior author. “If someone is clinically healthy but their immune health suggests that they are slowly inching toward pathology, instead of trying to fix a disease much later at symptom onset, we hope to focus on prevention.”
Researchers use artificial intelligence to analyze immune variation
Tsang’s laboratory investigates variations in the human immune system. In its latest study, the team first utilized a cohort of 228 patients with 22 monogenic conditions and 42 healthy controls. Monogenic conditions occur when a single genetic mutation causes disease, often severe disease.
If someone is clinically healthy but their immune health suggests that they are slowly inching toward pathology, instead of trying to fix a disease much later at symptom onset, we hope to focus on prevention.
John Tsang, PhD, MMath
The researchers measured a variety of variables including gene expression in the blood, circulating proteins in blood serum, and immune cell frequencies. “We found, surprisingly, that despite different genetic perturbations, there are some common features that are shared among these patients compared to healthy controls,” says Tsang. “This was our first hint that, despite different causes, there may be a common signal for deviation from health being sensed by the immune system.”
Next, the team utilized two machine-learning approaches—unsupervised and supervised—for assessing variations in and deviations from immune health. An unsupervised analysis means that the machine analyzed the data without knowing which diseases the patients had, if any. Both approaches generated predictive models that, given a particular person’s immune profile, could compute the probability of whether that individual is healthy. Interestingly, these scores of immune health, or IHM, were consistent across both types of machine learning.
After developing the IHM, the researchers tested it on 10 independent datasets. In each of these new cohorts, the model was successful in measuring health deviations based on the immune profiles of patients experiencing a range of conditions, including autoimmune diseases and heart failure. It also accurately detected natural declines in health associated with age in otherwise healthy individuals in multiple cohorts, including the team’s original cohort and individuals from the Baltimore Aging Study. Finally, it predicted treatment responses for various diseases and immune responses following various vaccinations.
Future precision medicine tools could monitor current and future health via the immune system
Tsang hopes his research will lead to precision medicine tools for detecting current and future disease based on an individual’s immune profile beyond genetics, given that the immune system can sense and remember diverse environmental exposures. Because debilitating conditions, such as those associated with aging like Alzheimer’s disease, often develop for years before showing signs, “we actually have a large window to intervene if we can detect them early,” he says.
Beyond the prospect of disease prevention, the team’s approach also shows promise in predicting future health outcomes. For instance, the findings could lead to the development of tools that predict how strongly an individual will react to various vaccines. Older adults, for example, typically produce fewer antibodies post-vaccine, but we are not yet able to predict which individuals in the older population have the greatest risk of having poor responses.
In further research, Tsang hopes to build on this work to create models that can inform clinicians of the trajectory of a patient’s health. He envisions a future in which “a patient comes in and their clinician can not only evaluate any hidden health changes of that patient now, but also look ahead to what may lie in the future,” he says.