Researchers Use AI to Predict Smell of New Molecules
Scientists at the University of Reading have used machine learning to predict the odour profile of molecules solely based on their structures. This development has the potential to significantly transform the fragrance and flavour industries. This can lead to the development of tools that identifies molecules with similar structures that smell different and those with different structures that smell alike. The research was published in Science in August 2023.
Cracking the Scent Code
Professor Jane Parker from the University of Reading led the research and explained the significance of this breakthrough. She pointed out that while vision research has wavelengths and hearing research has frequencies that can be measured and assessed, smell has been a challenging frontier due to the lack of a precise measurement method. “We don’t currently have a way to measure or accurately predict the odour of a molecule based on its molecular structure,” she said.
Developing Odour Predicting AI Models
To bridge this gap, the research team turned to machine learning. They developed an “odour map” using AI. This could prove invaluable to synthetic chemists in the food and fragrance industries. Unlike previous models of olfaction, this AI-generated model correctly predicts the odour of molecules that deviate from the norm. Professor Parker noted the map’s versatility, emphasizing that it doesn’t limit itself to known odorants or structurally similar compounds. Instead, it can describe a wide range of unrelated molecules with distinct molecular characteristics.
The Potential Impact of Assisted Smell Detection
The Principal Odour Map (POM), as it’s called, preserves the structure of odour perceptual space. This innovative tool opens doors to a wealth of potential odorants. Colaborating institutions included the Monell Chemical Senses Center at the University of Pennsylvania, Arizona State University, and Osmo, a machine learning lab spun out of Google.
The University of Reading played a vital role in ensuring the accuracy of the AI model. They assessed the purity of the samples used for testing. Using gas chromatography, they separated trace levels of impurities from the target molecule and analysed their individual scents. This approach helped confirm the AI’s predictions and uncover any impurity-related discrepancies. Once the model had been trained with data, its ability to predict the smell of a new compound proved to be highly accurate, matching human scent scores on average.
A New Frontier for Synthetic Chemistry
This AI tool promises to be a game changer to synthetic chemists, allowing them to explore new aromas and screen vast numbers of molecules for their scent properties, much like the pharmaceutical industry screens compounds for new medicines. It can reshape industries and deepen our understanding of the fascinating world of scents and flavours.
References:
- Phys.org: AI with a nose for molecular odors
- Science: Predicting odor percepts from molecular structure
