Building a Better Nose
Dogs, ants, and other organisms that sniff out cancer could help train an AI to do the same.
For dog lovers, the idea of friendly canines as living, breathing, tail-wagging cancer detectors is a hopeful one. Not only do dogs conjure smiles, but their known olfactory abilities would offer a strange contrast to the sterile medical exam rooms many find dreadful: brushed steel countertops, white lab coats, buttercup walls, and the penetrating smell of disinfectants.
But if dogs have already shown the ability to detect cancer on human breath or urine, researchers have now found one better: Ants could be a more cost-effective means of harnessing the same super-sniffing abilities of their distant cousin canines to help detect cancer and other illnesses in humans. We may eventually be able to use both dogs and ants to train artificial intelligence-powered devices to do the same thing.
“Insects have a life that is much shorter than that of mammals. They have to learn fast,” says Patrizia d’Ettorre, an expert in ant behavior at University Paris 13 in France. D’Ettorre co-authored a recent study in which ants learned to detect a unique scent in just 30 minutes. “Ants are very efficient in learning odors because their main communication strategy is chemical communication,” she says. This is true of a number of insect species, though d’Ettorre points out that non-flying, non-stinging insects are the most practical for medical use—for obvious reasons.
Using ants or dogs to detect hidden human diseases takes advantage of evolutionary design, which has endowed these modern animals with the sensitive ability to detect trace amounts of chemicals known as volatile organic compounds (VOCs). In the same way that a golden retriever can sniff out a person buried deep in an avalanche, the hope is that animals trained to detect cancer could save lives by detecting tiny traces of those chemicals early in the disease. Ants are master sniffers the same as dogs, but before the French team did this work, nobody had trained them to detect cancer.
D’Ettorre and her colleagues showed they could do just that by rewarding ants with a sugar solution when they correctly identified the odor of ovarian cancer. Later, they trained other ants to detect a line of cells derived from adenocarcinoma breast cancer, even compared to a different line of healthy breast cells. Finally, they trained a third group of ants to distinguish between two different types of breast cancer—though additional studies are needed to prove ants can detect cancer even in less pure samples from living humans, smelling cancerous cells through the fog of our other bodily odors.
“We tested VOCs from cancer cell cultures,” d’Ettorre says. “These might be less complex than body odors modified by the presence of a tumor.” Such complexity is why dog trainers like Heather Junqueira do similar, reinforcement-based training for pups used to detect cancer, then follow it with dozens of unique samples. Junqueira, founder of BioScent, prefers to use about 40 different samples containing different volatile organic compounds before she considers a dog fully trained.
Still, while dogs’ sniffing skills are likely part of future cancer testing, dogs themselves aren’t the best long-term candidates, according to Junqueira, whose company started using beagles to sniff out cancer but shifted to coronavirus detection at the start of the 2020 pandemic.
“I don’t see using the dogs as a screening tool for cancer long term,” Junqueira says. It’s hard for dogs to stay specific to one certain smell. Over time, dogs trained in identifying one type of cancer may begin to recognize volatile organic compounds associated with other cancers. If they’re rewarded for identifying a different cancer smell, it then becomes difficult, if not impossible, for them to return to solely tracking one specific scent. There have even been cases, Junqueira says, when dogs thought to be identifying cancer had been unwittingly trained to identify the scent of a particular hospital. This is another difficulty: ensuring that the smell a dog has signaled for is, in fact, cancer itself, particularly when humans don’t know the precise volatile organic compound signature of every ailment.
One possibility is that dogs can help us identify the volatile organic compounds related to a particular cancer and then artificial intelligence can be used for diagnostic purposes. In this way, AI could mimic a dog’s ability. The next challenge for the ants, however, will simply be identifying cancer in a more complex and realistic sample: urine from tumor-laden mice.
“Having an at-home test, as simple as a breathalyzer or a pee stick, could really help in terms of monitoring the condition in real-time.”
Engineers like Mangilal Agarwal of Indiana University–Purdue University Indianapolis are trying to solve a bigger puzzle: How to develop nano-sensors that are as accurate and sensitive as a dog or an ant. Such a sensor could not only diagnose but offer daily, non-invasive surveillance for people at risk of, say, breast or prostate cancer. It could also help monitor a person’s treatment progress, as the sensor could measure a decline in cancer-related volatile organic compounds.
“Having an at-home test, as simple as a breathalyzer or a pee stick, could really help in terms of monitoring the condition in real-time,” Agarwal says. There are lots of challenges, however. Currently, scientists can identify the specific volatile organic compounds they need to detect, and they have done so with breast and prostate cancers, but it’s an expensive and time-consuming process that relies on huge mass spectrometer instruments in the lab. Agarwal is working on developing portable equipment. There are also issues with confounding smells, such as medications that might mask a specific volatile organic compound. Imagine searching for one specific needle in a haystack made of other needles.
What remains unclear is whether science can match or even exceed animal senses that have evolved over millions of years. “Can biotechnology mimic canines’ ability to detect medical conditions? The simple answer is yes,” Agarwal says. “Can it be as accurate as dogs? Yes. Can it beat the dog? Maybe not.”