A Mathematical Representation of Precision Medicine
By Megan Mayerle, PhD
Diseases most often arise from a confluence of genetic and environmental factors. Precision medicine promises the ability to understand and treat disease in an individualized manner. To fully enable precision medicine, it is necessary to clearly define how each patient’s genome interacts with the unique combination of environmental factors that he or she has been exposed to and understand how these factors interact to cause disease at the individual level. In a recent Cell Leading Edge Commentary, Stanford researchers Jingjing Li, Xiao Li, Sai Zhang, and Michael Snyder propose a novel method to do just that.
For decades researchers have been using a very simple model, P = G + E, to describe how genetics (G) and environment (E) contribute to phenotypes (P) such as disease. However recent complex disease genome studies suggest this simple model is insufficient, suggesting “missing heritability”. At the same time, while it is well known that an individual’s lifestyle and environment can impact the development of disease, measurements of environmental contribution are often performed at a population level and are often limited to conventional epidemiological studies with little precision toward a personal level. Therefore, conventional approaches cannot be directly transferred to personal health management. Misinterpreting how individual genetics and personal behaviors and environment contribute to disease could significantly negatively impact how disease risk is managed, and how policies surrounding complex diseases are crafted.
The researchers attempted to find the origin of missing heritability computationally, building a deep neural network simulation that allowed them to define increasingly complex genetic scenarios and determine how much heritability is missed using standard methodologies. They showed that while standard methods are adequate for simple genetic scenarios, the more extensive the interactions between individual genetic contributions to disease became, the less accurate standard methods were, leading to over-simplification and underestimation of heritability.
They then approached the problem of individual environment and behavior, highlighting recent efforts using electronic health records in deep learning models to predict individual clinical outcomes, attempting to bridge population health and personalized medicine. While such efforts are highly promising, they suffer from a lack of data- physiological and behavioral measurements are often only been collected during clinical visits.
On the environmental front, wearable biosensor technologies, small electronic devices that can be comfortably worn or loosely attached to the body, are increasingly being used in the clinic. Biosensors that measure heart rate, heart rate variability, skin temperature, blood oxygen level, respiration rate, glucose level, and record individual exposure to microorganisms are already available, and are a means by which researchers and clinicians can gather individual physiological, behavioral, and environmental data in real time.
Li et al. synthesize these data and observations into a new model. Their model, called BADGE for Bayesian Aggregation of Disease Genomic and Environment, provides a novel model to study gene-environment interactions, and is able to hierarchically model genetic risk, environmental factors and their mutual interactions using probabilistic learning techniques. This model unifies major concepts in epidemiology, population health and personalized medicine, a major step forward for precision health.