Osmo
Assembly Line
🧠 This Neural Net Maps Molecules to Aromas
Using a type of deep-learning algorithm called a graph neural network, researchers have built a model that maps chemical structure to odor descriptors. The model has successfully predicted how a panel of humans would describe new smells, and it could be an important step along a long path toward digitizing smells.
The model used a specific type of graph neural network called a message-passing neural network. It was trained on a combined fragrance industry dataset of over 5,000 molecules with their structures converted into graphs and tagged with professional odor notes. Part of the research group worked at Google when the work began, and a few have since formed an offshoot company, Osmo, in January 2023, supported by Google Ventures, Alphabet’s venture capital arm.
Building a Map of Odor
To solve the SOR problem and predict what a molecule smells like from its structure, we made a major breakthrough using a relatively new kind of machine learning called Graph Neural Networks. We spent years validating our work at Google Brain, some of which we just shared with the world in a blog post. We built new molecules no one had ever smelled before and predicted them with superhuman accuracy. We built molecules that smell bad to mosquitoes (e.g. insect repellents) and are more potent than DEET in human trials (this could save lives). And we discovered that species as diverse as humans, mice and insects may share the same odor map, just like our eyes all use RGB. Why is that? The light from the sun has been the same to living things for billions of years, so all of our eyes evolved to take advantage of that. Over evolutionary time, life all breathed the same air, so it’s reasonable to expect that we developed the same map. The scents in air we breathe are made by other living things, so perhaps our sense of smell evolved to smell the building blocks of life.