Chemical biomarker profiles extraction for honeybee pathogens using machine learning

Official URL: https://risc01.sabanciuniv.edu/record=b2708492_(Table of contents)

In recent years, we have increased on our reliance upon honey bee pollination services yet bee health has been declining on a global scale. The decline in bee health is a complex multifactorial problem and it is caused by a number of interacting stressors. The stressors are mainly stemming from pesticide exposure, parasitic infections, poor nutrition, and loss of foraging habitat. However, how these stressors exactly interact to produce a synergistic decline in bee health remains elusive because previous studies have mainly focused on one or two stressors at a time using traditional experimental testing in the laboratory. Here we utilize a systems biology approach that is a non-hypothesis data driven analysis. We integrate the exposome profile of 87 honey bee hives, sampled from rural to urban areas, with the abundance datasets of the 20 most common bee diseases to determine the specific interactions responsible for a decline in bee health. From this analysis, we have developed chemical biomarker libraries for 13 of the bee diseases that are able to predict whether a hive is infected or not. The biomarker libraries were validated using five different machine learning techniques that consistently demonstrated our chemical biomarker libraries can predict whether a hive is infected with a particular disease or not with roughly 85% accuracy, precision, sensitivity, selectivity, and recall. In addition, using a network analysis across the integrated datasets, we found that across the bee diseases there are five metabolite hubs that are suspected to be potential targets that are responsible for an increase in susceptibility of the honey bee to multiple infections or can explain how multiple infections lead to a synergistic decline in bee health mechanistically. Moreover, we identified a number of environmental pollutants that are highly toxic to humans, which are also associated with bee diseases and are linked to detoxification and oxidative stress response genes. Our findings suggest that not only can bees be used a bioindicators or sentinels for monitoring environmental quality for human health, but the exposures themselves to the honey bees are likely to be a detriment to their health as well. These environmental exposures from polluted environments are likely another stressor that is negatively impacting bee health and their implications have yet to be fully recognized in the most recent decline in bee health. From the systems biology analysis we provide chemical biomarkers that can be used as a possible rapid diagnostic tool such that beekeepers can change management practices to improve honey bee health before the colony collapses from parasitic infections. Novel stressors have been identified that are likely negatively impacting bee health and these are interacting with a multitude of exposures that are linked to an increase in disease prevalence in the honey bee hive. Collectively, our findings support the notion that the One Health paradigm is likely to be the most effective strategy for addressing the complexity of declining bee health and for improving it moving forward.