racking the Mystery of Smell: AI and Science Team Up to Decode Odors
PHILADELPHIA (August 31, 2023) – Have you ever wondered how we’re able to smell things? Scientists are diving into the intricate world of neuroscience to understand how our senses work, like how we turn light into sight and sound into hearing. But when it comes to smell, things get even more puzzling.
To untangle this mystery, a group of researchers from the Monell Chemical Senses Center and the startup Osmo are teaming up to explore how the chemicals floating in the air connect to the way we perceive odors in our brain. They found something really cool: a machine-learning model that can describe smells using words, almost as well as humans can. Their findings are published in the September 1 issue of Science.
Joel Mainland, one of the senior co-authors of the study, says that this model helps fill in gaps in our understanding of how smell works. This exciting collaboration takes us a step closer to recording and recreating scents using technology. It could also lead to discovering new scents for things like perfumes or even mosquito repellents.
How Our Brain and Nose Team Up
We have around 400 olfactory receptors in our noses. These are special proteins that connect with molecules in the air and send signals to our brain, making us smell things. This number is way more than the receptors we use for seeing colors (four) or tasting (around 40).
“In smell research, though, we’ve never fully understood why some molecules make us smell a certain way,” said Mainland. “But if a computer can figure out the link between the shape of molecules and the way we smell them, we could learn more about how our brains and noses work together.”
To crack this puzzle, Osmo’s CEO Alex Wiltschko and his team created a computer model. This model learned how to match the words people use to describe smells with the actual structure of the odor molecules. It’s like grouping similar smells together, such as “floral sweet” or “candy sweet.”
“Computers have conquered vision and hearing, but smell, our oldest sense, has remained a challenge,” said Wiltschko. “This study introduces a unique way of understanding human smell, connecting chemical structure to how we perceive odors.”
Mapping Smells with Computers
They trained the model using a bunch of data about 5,000 different smells and their molecular structures. The computer looked at how the molecule is shaped and then predicted which words would best describe its smell.
To check how good the model was, the researchers had a group of people describe new smells, and then they compared their descriptions with the computer’s guesses. The computer model performed really well, often predicting smells better than individual people. It could even predict how strong a smell might be, even though it wasn’t specifically trained for that.
This kind of research could help us understand more about how our sense of smell works and maybe even change the way we think about scents in the future. Scientists believe that the way smells are organized might be related to the nutrients they come from, rather than just their chemical structures. So, your brain might be thinking about food when you’re smelling something!
This study is just the beginning. The researchers want to explore how this map of smells could be organized based on the things our bodies need. It’s a whole new way of thinking about the scents around us. So, the next time you take a sniff of something, remember that there’s a lot more going on in your brain and nose than you might think!