HE PULLS UP a video to show me what he means. In it, a black dog named Lucy approaches a series of six stations, each separated by a small barrier. At every one, a glass cup of human urine with a screened lid sits at the level of the animal’s nose. Lucy takes a brief sniff of each sample, sometimes digging her snout in to get a better whiff. She is performing a kind of diagnostic test: searching for the telltale scent of prostate cancer, which, it turns out, leaves a volatile, discernible signature in a man’s pee. Discernible if you’re a dog, anyway. When Lucy finds what she’s looking for, she sits down and receives a treat.
Among humans—whose toolmaking prowess has given the world self-driving suitcases and reusable rocket boosters—prostate cancer is notoriously difficult to detect. The prevailing method is to check a patient’s blood for elevated levels of a protein called prostate-specific antigen. But the test has a miserable track record. The scientist who first discovered PSA has described the test as “hardly more effective than a coin toss.” A false positive can lead to a prostate biopsy, a harrowing procedure that involves inserting a large, hollow needle through the wall of the rectum to retrieve a tissue sample from the prostate itself.
Properly trained dogs, on the other hand, can detect prostate cancer with better than 90 percent accuracy, and with sleek, tail-wagging efficiency. In the video, Lucy works her way through six samples in just a couple of minutes. This drives Mershin up the wall. “We have $100 million worth of equipment downstairs. And the dog can beat me?” he says. “That is pissing me off.”
Mershin is not a doctor. He’s a physicist by training. He runs a lab called the Label Free Research Group, which exists to spite the boundaries between physics, biology, materials science, and information science. In his office, Mershin keeps a pair of sunglasses that can measure brain waves, along with magazines on aviation and books on urology, the physics of consciousness, and coding in Python. He speaks rapidly in an accent that sits somewhere between his two native languages, and he changes subjects at the slightest provocation. He refuses to wear matching socks, because why should socks match? He is short and round, with a mane of strawberry blond curls that bounce when he gets excited.
Mershin’s lab, where he keeps that $100 million worth of equipment, sits a few floors down from his office at MIT. In one room, researchers are trying to invent new colors; in another, to create the lightest, strongest materials on earth. But I’m here because this facility is doing some of the most important research in the world toward developing AO—artificial olfaction.
Plenty of robots these days can see, hear, speak, and (crudely) think. But good luck finding one that can smell. In part, that’s simply because olfaction has always been deeply underrated by humans—a species of cerebral, hypervisual snobs. Kant dismissed smell as the “most dispensable” of our five senses. One 2011 poll found that 53 percent of people ages 16 to 22 would rather give up their sense of smell than give up their smartphones and computers.
But in the past several years, it has become increasingly clear that smell, in the right snout, can be a kind of superpower. For millennia, humans have prized dogs for their tracking abilities; police and armed forces have long used them to sniff out bombs, drugs, and bodies. But since about the early 2000s, an avalanche of findings has dramatically expanded our sense of what dogs can do with their noses. It started when researchers realized that canines can smell the early onset of melanoma. Then it turned out they can do the same for breast cancer, lung cancer, colorectal cancer, and ovarian cancer. They can smell the time of day in the movement of air around a room; sense diabetic episodes hours in advance; and detect human emotional states in the absence of visual cues. And it’s not just dogs. Tipped off by a Scottish nurse with a highly attuned nose, scientists have recently learned that people with Parkinson’s disease begin emitting a distinct “woody, musky odor” years before they show symptoms.
Andreas Mershin believes we don’t have to understand how mammals smell to build an artificial nose. He’s betting that things will work the other way around.
All this adds up to a revelation not just about dogs but about the physical world itself. Events and diseases and mental states leave reports in the air—ones that are intelligible to highly attuned olfactory systems but otherwise illegible to science. Smell, it appears, is sometimes the best way of detecting and discriminating between otherwise hidden things out in the world. And often, the next-best method of detecting that same thing is expensive (gas chromatography/mass spectrometry) or excruciating (tissue biopsies) or impossible (mind reading).
Unfortunately, the other reason we don’t have robots that can smell is that olfaction remains a stubborn biological enigma. Scientists are still piecing together the basics of how we sense all those volatile compounds and how our brains classify that information. “There are more unknowns than knowns,” says Hiroaki Matsunami, a researcher at Duke University.
Mershin, however, believes that we don’t really have to understand how mammals smell to build an artificial nose. He’s betting that things will work the other way around: To understand the nose, we have to build one first. In his efforts with a brilliant mentor named Shuguang Zhang, Mershin has built a device that can just begin to give dogs—his panting adversaries—a run for their money.
As for the precise nature of those interactions, Buck and Axel could only theorize. They posited a sort of lock-and-key relationship between the olfactory receptors in our noses and the molecules in the air. But the number of receptors they discovered instantly posed a mathematical problem. Humans have about 400 kinds of olfactory receptors (far fewer than mice), but we can smell about 10,000 distinct odors. So Buck and Axel theorized that smell was combinatorial. Each receptor, their research showed, is uniquely primed to react to a few different molecules, and our noses sense distinct odors when many receptors fire at the same time. John Kauer, then a researcher at Tufts University, relates the idea to playing chords on a piano. “The piano only has 88 notes,” he says. “If you were only able to use one note per odor, you could only detect 88 different odors.” If odors are more like chords, then the math suddenly works out.
Inspired by Buck and Axel, who won the Nobel Prize in 2004 for their work, Mershin and other scientists conceived of odors as simply lists of molecules. If you want to understand the smell of a clove of garlic, the thinking went, the answer lies in its chemical components. “Somewhere in these molecules,” Mershin believed through the mid-2000s, “the smell of garlic is written.”
After Buck and Axel released their major findings, it didn’t take long before the first major efforts to build an artificial nose got underway. Darpa wanted to replace dogs as a tool for finding land mines, so beginning in 1997, it poured $25 million into a program called Dog’s Nose. The agency funded scientists across the country to build a bunch of would-be sniffing machines and then brought them to a field in Missouri for testing. The ground was sown with every manner of land mine, from small antipersonnel devices the size of tuna fish tins to hefty antitank munitions. Although stepping on the mines could no longer set them off—the fuses had been removed—the buried explosive ordnance could still be set off by, say, a lightning strike. “As soon as there was any hint of a thunderstorm,” says Kauer, who participated in the program, “we evacuated.”
Kauer had built a gray, shoebox-sized device that he eventually christened the ScenTrak. His gadget wasn’t equipped with actual olfactory receptors. Instead, it was packed with long strands of molecules called polymers that Kauer knew would react to DNT, a molecule common in most land mines. When the ScenTrak came across an explosive, the DNT bound to the polymers, causing the ScenTrak to set off an alert. “Land mine!” the box cried.
At least, that’s how it worked in ideal conditions. ScenTrak was able to pick out nearby traces of DNT in the air of an otherwise odorless lab. Out in the field, though, when Kauer scanned the ScenTrak back and forth over a patch of ground, it became confused. The polymers would react to DNT, but also to the weather, to plants, or to certain kinds of soil.
Other devices in the competition, including one called Fido and another called Cyranose, were based on roughly the same theory. They all used polymers sensitive to specific compounds. And they all proved somewhat narrowly functional. (Fido is now used at military checkpoints to scan for explosives at close range.) But these devices don’t really smell, any more than, say, a carbon monoxide sensor can smell. They often misfire in scent-rich environments where odors—apparently made of some of the same compounds—may waft in from various nonexplosive sources.
In part, that’s because the theory these devices were built on was too reductive. Today, most scientists believe that the lock-and-key theory of olfactory bonding is far too simple. In some cases, it turns out, molecules with very similar shapes have completely different odors; in others, very differently shaped compounds smell alike. A molecule’s shape, in other words, is not synonymous with its smell. Instead, many receptors bind to many different molecules and vice versa. But each receptor has what some scientists call a distinct “affinity” for each molecule. It’s that special affinity, the theory now goes, along with the combinatorial nature of olfactory reactions, that accounts for unique scents. The piano doesn’t just have 88 keys that can form chords; it also has pedals and dynamics. “You hit piano keys at different strengths, heavy and light and so on,” Zhang says. “Heavy, you get one sound; light, you get another sound.” Or to put it another way: The theory of smell just gets more complicated.
About half of any olfactory receptor sits outside the cell, ready to interact with molecules. Then a middle section sits inside the cell membrane, and the rest resides inside the cell. When the exterior part of the receptor binds to a molecule, it changes shape, and the cell sends a message to your brain. While the heads and tails of an olfactory receptor—the parts that sit inside and outside the cell—love water, its middle section is hydrophobic, like the cell membrane that encases it. That means that when you take the receptors out of a cell and put them in water, they tend to clump together instead of dissolving, which makes them nearly impossible to isolate and work with.
Zhang has been toiling away at his goal since 2003. At one point, he spent eight years simply trying to create water-soluble receptors. (“And it’s solved,” he says. “It’s done.”) But even still, he has never succeeded at seeing a receptor. Nor has anyone else. They are simply too small. Zhang describes that basic interaction between the odor molecule and the receptor as a “total black box.”
Still, Zhang’s work did prove to be very useful when, in 2007, Darpa launched a second smell project, called RealNose. Spurred by the wars in Iraq and Afghanistan, RealNose had a new mission and a new sense of urgency. Instead of searching for land mines, the mechanical noses needed to be able to identify IEDs, which were laying waste to American troops. And this time, scientists couldn’t use polymers or other synthetic devices to mimic what the receptors did. They had to use mammalian olfactory receptors as their sensors.
Zhang had a big advantage over other scientists competing for those Darpa grants. Thanks to his work, he had one of the only labs in the world that had experience growing olfactory receptors in embryonic cells and then working with them in the lab. But Mershin wasn’t thrilled about Darpa’s requirements. “For many, many, many months I rebelled,” he says. He didn’t want to bother with those finicky olfactory receptors, and he tried to convince Darpa that its requirement was a bad idea. Why did they need to use the actual, biological structure when it would be easier to use something synthetic? Something that wouldn’t stop working just because it was tilted or upside down? “Sure, we want to fly like birds, but we don’t build jet engines out of feathers,” he thought. “We want something better than birds!” Mershin just wanted a sensor that could tell you what molecules were present in a room. But he didn’t want to miss out on the funding, so he conceded.
Mershin and Zhang decided they would grow a bunch of olfactory receptors in their lab and then essentially smear them onto a circuit board. They figured, statistically speaking, that if they slathered on enough receptors, they’d wind up with enough of them oriented in the right direction. Then they would connect the circuit boards to an electrical current. When the receptors interacted with volatile compounds, they would change their shape, just like they do in a regular nose. But instead of sending a message to a brain, the interaction would be recorded as a simple blip in electrical current.
On a clear day in early spring, Mershin leads me into his lab and rifles through some cardboard boxes and equipment until he unearths a container of old artificial nose prototypes. In one hand, he pulls out a plastic bottle with two metal nozzles haphazardly held in place with epoxy. From his other hand, a thin plastic chip dangles from some electrical wires. “This here is the first nose,” he says.
It was a failure. The receptors seemed to work, but the bottle was too big; smells would linger too long for the scientists to get a clear reading. So they followed it up with more prototypes, experimenting with different ways to deliver the right odiferous blast of air to the chip, and to different numbers of chips.
From his hopper of prototypes, Mershin eventually pulls out the device, called the Nano-Nose, that he and Zhang ultimately submitted to Darpa. The whole contraption is about the size of an extra-large roasting pan and is emblazoned with the words “Property of the US Federal Government.” “Because it was for Darpa,” Mershin says, “we had to make it look bulletproof.”
After all their prototypes, they had eventually homed in on a design that used an array of eight circuit boards, each about the size of a credit card. Inside that bulletproof metal housing, each board sat in a separate airtight bay, capable of receiving its own puff of odor and responding with its own electrical pattern. Smells could be sent into the box, and directed to each board, by an air pump that mimics taking a deep sniff.
Zhang and Mershin built the device in a 15-month sprint, and it still wasn’t finished when Darpa’s deadline arrived. When it came time to show their work, Mershin loaded up a large van with the contents of nearly an entire lab—hoses, tubes, pipes, syringes, a 300-pound optical table, and a frequency generator worth $70,000—and drove it from Boston to Baltimore. He even brought their own odor delivery system: a modified inkjet printer called the StinkJet.
Mershin had originally envisioned putting a supercomputer underneath the Nano-Nose that would dig through databases listing thousands of compounds and print out those the nose registered. But they’d never gotten around to that part. Instead, they resorted to what Mershin thought of as a hack.
During lunch breaks, the team would rush the nose back to their hotel room, soldering pieces onto it to keep improving it while ordering room service.
Darpa had given them a list of odorants that their machine would be asked to recognize. So first off, Mershin and Zhang sent those odorants through the Nano-Nose and recorded its responses; the idea was to train the nose, with the help of a laptop and a pattern-matching algorithm, on what it was supposed to be smelling for. Then, in the actual test, they would sample each mystery odorant eight times—once through each of the eight bays—and run it through a gauntlet of varying electrical conditions. This amounted to a process of elimination, meant to help the pattern-matching algorithm filter out false positives. It wasn’t as sophisticated as a data-mining supercomputer, but they thought it might work.
The Darpa tests were highly controlled. Mershin and his team were not allowed to be in the room with their machine while the trial was running, and Mershin wasn’t even allowed to go to the bathroom without a security escort. During lunch breaks, the team would rush the nose back to their hotel room, soldering pieces onto it to keep improving it while ordering room service.
In the end, the mad dash paid off. The Nano-Nose passed the sniff-off and was able to sense isolated odors in the lab. It even beat dogs in a controlled environment, sniffing out odors in lower concentrations than canines could detect. And it didn’t need a supercomputer. In fact, Mershin says, the Nano-Nose was better without it. To him, the project revealed a fundamentally important aspect of olfaction: Our noses are not analytical tools. They don’t analyze the components of a scent. “The molecule is what carries the message,” Mershin says, but you can’t understand what our perception will be just from knowing the molecule. “We thought that when you sniff something, a list of molecules and concentrations comes up,” he says. “Not the case.”
As it turned out, Mershin’s hack actually mirrors how mammals process smells. Instead of giving equal computational attention to all the compounds we inhale, our brains hierarchically sort information based on what’s important to us. We can tune out smells in a room if we’re not interested. Our receptors are still sensing compounds, but our brains aren’t paying attention. Conversely, if we narrow our attention on the signals our receptors are sending, we can pick out the subtle scent of shallots or fennel in a pasta sauce brimming with the competing scents of tomato, peppers, and garlic.
Mershin realized that to understand smell and to use it as a tool, he didn’t need a list of molecules. He needed to know what something smelled like, not what it was made of, and those are fundamentally different things. “It was the biggest lesson I’ve ever had in my entire scientific career,” he says. “We thought we understood how noses worked. We didn’t know anything about how noses worked.”
As impressive as the Nano-Nose is, it will take more than a boxful of blipping circuit boards to replicate everything Kato does when he’s tracking a scent. Paul Waggoner, a scientist who studies canine olfaction at Auburn University, estimates we are “decades away” from creating machines that could successfully compete with natural olfactory abilities. Waggoner, who also has his own patented training program for detection dogs, argues that machines break down early in the smelling process. “It all starts with the sampling,” he says. Essentially, machines don’t sniff very well. Dogs inhale and exhale about five times every second, through nostrils that route the intake and outward flow of breath through different channels. All that snorting creates a pressure differential—a kind of smell vortex—that helps them pull a rich, new sample into their nose with each sniff. And while the Nano-Nose might be able to narrow its focus on a target scent, a dog’s ability to do so over great distances is stunning.
What happens in Kato’s brain when he finally catches that scent? Well, no one knows. The higher up the chain we go, from olfactory receptors to how the brain processes and understands that information, “the darker and darker it gets,” Waggoner says.
Still, dogs are not perfect sniffers themselves. On a second visit with Thompson, I watched another dog, a 3-year-old Malinois named Annie, completely lose focus on tracking down a bone when she encountered several pigs in a nearby field. “When dogs aren’t used to stuff, it’s very difficult,” Thompson explains. Dogs get frustrated and tired. They feed off their owners’ emotions. And of course, dogs don’t scale. Highly trained bomb- and disease-sniffing dogs are in short supply and expensive, as much as $25,000 per pooch. Already, the US security sector doesn’t have enough dogs to cycle through all the different agencies—from the TSA and local law enforcement to the military—that need them. Medical detection dogs are even trickier: Not only are there very few of them, they don’t exactly plug easily into a medical setting. Despite all of the incredible findings in the past several years—the 90 to 100 percent accuracy rates at detecting early cancer—medical detection dogs have not been widely adopted as diagnostic helpers.
Over the past few years, Zhang has continued to tinker with the olfactory receptors he and Mershin use in their Nano-Nose. Most importantly, he’s stopped growing them in embryonic cells, having devised a way to cultivate them in a biologically inert form. It all happens in a test tube now. The receptors are still tricky to handle—Mershin says they are by far the most difficult aspect of the device—but these are more stable and malleable than their organic counterparts. Mershin and Zhang have also progressively shrunk the Nano-Nose’s circuit boards. That means the entire apparatus could now be attached to a port on a bioreactor to sniff what’s happening inside. It could go inside a factory and smell products for quality control or be put inside a grain silo to smell for food spoilage. But Mershin and Zhang say they have no interest in turning their research into a business at the moment.
So far, the only company daring enough to design a commercial technology that uses olfactory receptors—with a design very similar to the Nano-Nose—is a small Silicon Valley startup called Aromyx. In some ways, it is even more ambitious than Mershin and Zhang. The Nano-Nose uses only about 20 kinds of receptors and customizes each nose depending on its purpose. But Aromyx wants to pack all 400 human olfactory receptors onto its EssenceChip, a 3- by 5-inch plastic plate dotted with small wells to hold the receptors. When the EssenceChip is exposed to an odor, the receptors fire and the chip records that activation pattern. What’s the smell of Coca-Cola? Or Chanel No. 5? The answer, again, isn’t a list of molecules. “It’s a pattern of receptor responses,” says Aromyx founder Chris Hanson. Thus far, Aromyx has stabilized only a few of those 400 receptors. As they add receptors, the thinking goes, their digital olfactory rendering will become finer and more detailed.
“This is a window into human sensory experience,” Hanson says. If so, it’s a fragile one. Aromyx still grows its receptors in yeast cells, and the company has struggled to put together a basic product for a demo. When Aromyx recently changed offices and moved seven miles from Palo Alto to Mountain View, some of its cell lines were destroyed in the shuffle.
In his office, Mershin gives me the place of honor: a black velour chair where Florin, another prostate-cancer-detection dog, sat when she came to visit. Florin and Lucy belong to a group in the UK called Medical Detection Dogs, which has trained many of the animals that have been able to sniff out cancers.
Right now, Mershin and Zhang are training an AI system on a bunch of data, some of it collected by Medical Detection Dogs on how their animals responded to specific urine samples—whether they alerted to cancer, how long they lingered, and the like—and some of it collected by Mershin and Zhang when they ran the same urine samples through a gas chromatographer/mass spectrometer. Mershin says these streams of data will help them select which receptors they need to put into the Nano-Nose. But the main event will come when he runs those same urine samples through the Nano-Nose and begins collecting data on its responses. Then he’ll mine all three data sets for correlations. Mershin already has all the urine frozen in his lab, ready to go.
The idea is to ultimately run a kind of Turing test, but for smell—to imitate the dogs’ results until no one can differentiate between the Nano-Nose’s reactions and a canine’s. If all goes well, the Nano-Nose will become more than just a sensing device; it will be a true diagnostic tool. The richer the database, the better the nose will be.
Ultimately, Mershin wants to see the Nano-Nose incorporated into your cell phone. He imagines using this intimate version of his device—one that rests at all hours against its owners’ body—to collect longitudinal data about its wearer’s health. Eventually, the nose would be able to alert you to get that mole on your thigh checked out, or warn you that your blood sugar is dropping dangerously low, or perhaps that you’ve started emitting the woody, musky odor of Parkinson’s disease. The Nano-Nose could accompany you everywhere and keep tabs on you in ways that doctors never could. Everything that a dog can detect via smell, it would detect.
That’s a powerful idea, but it’s also an unsettling one. How much control over your odor profile data would you retain? And if your phone is capable of sniffing you, what other devices would do the same? In a world where digital olfactory sensors have become small enough to fit into your pocket, presumably they’ll end up elsewhere—much the way video cameras did before them. If your diseases and mental states leave suddenly legible reports in the air, no doubt people besides you and your doctor will be curious to read them. (Your insurance company, for instance.)
Poppy Crum, chief scientist at Dolby Labs, is rooting for technologies like the Nano-Nose, which she believes could democratize the early diagnosis of disease. But she also sees artificial olfaction as one of a host of rising technologies—some much farther along than others—that use sensors and data to suss out otherwise hidden inner states. Those technologies all require new standards for transparency and user control of data—standards that aren’t going to come from companies or researchers. “I think that’s something that has to be legislated,” Crum says.
Mershin, for his part, isn’t so worried about the dawn of an olfactory surveillance state. Instead, as a consummately overstimulated person, he dreads a world where devices start sending you odors. “I would be very supportive of all the technologies that smell you. I would be very leery of technologies that want you to smell them,” he says. “Don’t let the phones start putting scents in your head. Bad idea.” In other words, let your phone be the dog; you be the handler.
Source: The Quest to Make a Bot That Can Smell as Well as a Dog | WIRED