Researchers from North Carolina State University and the University of Colorado, Boulder, have developed a statistical model that allows them to tell where a dust sample came from within the continental United States based on the DNA of fungi found in the sample.
Using the fungi observed in a dust sample collected in Raleigh, NC (blue point), the model predicts that the sample is most likely to have originated near Columbia, SC (red point). The error in this prediction, 138.7 miles, is close to the algorithm's median prediction error. Beyond a single "pin-in-a-map" prediction, the shaded regions capture areas of the U.S. for which the sample is likely to have originated with 90%, 75%, and 50% probability.
The primary goal of the research was to develop a new forensic biology tool for law enforcement or archeologists. “But it may also give us a greater understanding of the invisible ecosystems of microbial life that we know are all around us, but that we don’t fully comprehend,” says Neal Grantham, a Ph.D. student in statistics at NC State and lead author of a paper on the work.
The researchers developed the model using data from the Wild Life of Our Homes citizen science project conducted by the Your Wild Life lab based at NC State. The project collected dust samples from approximately 1,000 homes across the continental U.S., including samples from 47 of the 48 contiguous states.
The goal of that project was to test the dust samples for DNA to identify the microbial species present in and around our homes. One of the things the project found was that the types of fungus — or fungal taxa — varied widely from region to region.
“Based on that finding, we wanted to determine if you could predict where a dust sample came from based on the fungi present in the sample, and — most of the time — we can,” Grantham says.
The researchers developed a model that analyzed the fungal taxa present in a dust sample and predicted where the sample came from. About five percent of the time, the model’s predictions were within 35 miles of the correct sampling site. Those were the most accurate predictions. The worst five percent were off by at least 645 miles. The model’s median prediction error was 143 miles. However, the research team is already working to make the model more accurate by developing more advanced algorithms.
“The work we’ve done so far was to determine whether this concept was viable,” Grantham says. “Now that we know it is viable, we’re developing statistical methods that are better suited to the problem.
“Ultimately, we want to have an online tool for law enforcement to run the results of dust samples taken from a piece of clothing, a body, or a vehicle, and get information on where the clothing, body, or vehicle has been,” Grantham says.
Before DNA testing, prosecutors relied on less sophisticated forensic techniques, including microscopic hair analysis, to put criminals behind bars. 
Before DNA, when a hair was found at a crime scene, it was examined under a microscope and compared with hairs from a suspect. Though crime lab analysts knew that two hairs couldn’t be matched with perfect accuracy, hair comparisons proved to be powerful evidence linking suspects to crimes. In court, some examiners and prosecutors were certain that they had a “match.” But DNA exonerations are now forcing the criminal justice system to confront the limitations of hair analysis.
