The ability to predict the odor profile of a molecule solely based on its structure has long been a challenge for scientists. While vision research has wavelengths and hearing research has frequencies that can be measured and assessed by instruments, the sense of smell has lacked a similar method of measurement. Until now.
Existing knowledge of molecular structure can only take us so far in predicting the odor of a molecule. Oftentimes, there are exceptions where the odor and structure do not align. This discrepancy has perplexed previous models of olfaction. However, researchers have now made a major breakthrough by developing a machine learning (ML) generated model that accurately predicts the odor of these exceptions.
The research team applied machine learning techniques to create an innovative “odor map.” This map not only identifies molecules that have different structures but smell the same, but it also uncovers molecules that look very similar but have distinct odors. The implications of this breakthrough extend to the work of synthetic chemists in the food and fragrance industries, presenting new possibilities for the production of more sustainable flavors and fragrances.
Professor Jane Parker, a leading expert in the field, emphasizes the significance of this development for researchers in the food and fragrance industries. The odor map has the potential to uncover an untapped source of thousands, if not millions, of potential odorants. This opens up exciting opportunities for scientists to explore novel aromas and expand the range of scents available for various applications.
To achieve this breakthrough, Professor Parker collaborated with colleagues at prestigious institutions such as the Monell Chemical Senses Center at the University of Pennsylvania, Arizona State University, and Osmo, a company that emerged from Google’s machine learning lab. The University of Reading played a crucial role in the research by verifying the purity of the samples used to test the AI model’s predictions.
The use of gas chromatography enabled the separation of trace levels of impurities and the target molecule, allowing researchers to assess the individual molecules’ scents. This process identified instances where impurities overwhelmed or masked the desired odor, leading to more accurate interpretations of the AI model’s predictions.
Once the AI model was trained with data, its ability to predict the smell of novel compounds proved to be excellent. The model consistently matched the average scent scores determined by a panel of human testers. With this level of accuracy, the tool holds immense promise for synthetic chemists, enabling them to screen large numbers of molecules for aroma, similar to how the pharmaceutical industry searches for new medicines.
The development of a tool that can predict the odor profile of a molecule based on its structure marks a significant milestone in scientific research. By harnessing the power of machine learning, researchers have created an odor map that surpasses previous limitations. This breakthrough will undoubtedly revolutionize the work of synthetic chemists in the food and fragrance industries, leading to the discovery of new aromas and the production of more sustainable flavors and fragrances. The possibilities are vast, and the tool’s potential for innovation is boundless.