Scientists from the Boston University Schools of Medicine and Public Health have developed machine learning methods that are capable of identifying and characterizing metabolism-disrupting chemicals. Their research is published in Environmental Health Perspectives.
Sitting on your sofa too much might make you gain weight. Sounds like a statement of common sense, right? However, weight gain might not solely be a result of adopting an overly sedentary lifestyle. Rather, it could be caused by exposure to certain chemicals that might be present in your furniture.
These chemicals are known as metabolism-disrupting chemicals (MDCs) or “obesogens” and can be found in various household items and throughout the environment. As the name suggests, MDCs may trigger changes to an individual’s metabolic processes, creating a predisposition for weight gain through the stimulation of fat cell (adipocyte) formation.
Scientific research has only recently started to explore exactly which kind of fat cells are formed – there are several different types – as a result of being exposed to such chemicals. “This is an important question, since not all fat cells are ‘created equal’,” says Dr. Stefano Montifrom the Boston University Department of Medicine. “White fat cells store energy, contributing to obesity. Brown and brite (brown-on-white) fat cells burn energy, reducing obesity. Our previous work suggests that environmental chemicals are more likely to stimulate white fat cell formation.”
Monti explains that a correlation exists between the increased production of (and exposure to) environmental MDCs, and the rapid increase in obesity and metabolic diseases observed in humans. “Recent studies have demonstrated that the increase in BMI seen in recent years cannot simply be attributed to excessive calorie intake and/or insufficient energy expenditure,” he adds.
In order to limit our exposure to and use of these potentially harmful chemicals, we need to be able to know what and where they are, which has proven challenging. However, Monti and colleagues, including Dr. Jennifer Schlezinger, have published a new study that utilized machine learning approaches to successfully identify and characterize MDCs in a set of unclassified chemicals.
What is machine learning?
A branch of artificial intelligence (AI), machine learning utilizes data and algorithms to replicate the way that human beings learn. To learn a task, for example, human beings repeat it and perform the task until it is optimized. The same occurs in machine learning; with accuracy improving each time.
Why machine learning?
Why use machine learning in this context? The decision was based on Monti and colleagues’ desire to create an approach that would be both unbiased and data-driven. Using machine learning, the team were able to effectively “learn” from past research studies. “We ‘profiled’ a set of more than 60 chemicals with known effects (i.e., known to be either obesogens, or non-obesogens) and used them to ‘train’ a computer model to predict their metabolism-disrupting potential,” Monti describes.
The profiling stage of the experiment involved treating pre-adipocyte cells – derived from mice – with each of the chemicals and extracting mRNA from them. Next, the mRNA was sequenced using RNA-sequencing (RNA-seq) methods for transcriptional analysis. This process provided the researchers with information on how the cells’ genes had responded to the chemical exposure. “These RNA-sequencing profiles, together with the known chemical labels, were fed to a computer model that was trained to distinguish between the two classes, and then applied to the classification of unlabeled chemicals,” says Monti.
The RNA-seq profiles provided information on the effects of short-term exposure to the chemicals, whereas the labels (e.g., obesogen or non-obesogen) were used to provide longer-term exposure effects. Therefore, the machine learning model was trained to use the short-term expression profiles to predict the possible long-term exposure effects of the unlabeled chemicals. Monti emphasizes that this is a subtle yet important point.
The design of the experiment builds on previous work, the Carcinogenome Project, that aimed to identify potential carcinogens. “Together, the two studies provide a conceptual, experimental, and computational framework (i.e., a comprehensive ‘recipe’) of general applicability to the screening of large sets of chemicals for their potential long-term adverse effects including, but not limited to, metabolic disruption and carcinogenicity,” states Monti.
The full effect of MDC exposure
The research group want to emphasize that the applications of their latest study extend beyond the specifics of the method used and its predictive capabilities. The profiled chemicals in the study also included drugs that are used for the treatment of metabolic diseases. Thus, their methodology enabled the scientists to take a closer look at how these drugs impact a cell’s metabolism. “This understanding will in turn be instrumental to the design of more effective and targeted drugs with minimal side-effects,” Monti says.
Identifying a chemical as an MDC is merely the first step, Monti explains: “We selected two of the highly ranked predictions (tonalide and quinoxyfen, two commonly used pesticides), and performed extensive functional validation that conclusively confirmed their adverse effects in human fat forming cells. However, further testing would be needed to warrant any [regulatory] action,” he concludes.