Principal Odor Map

The Principal Odor Map (POM) is an innovative tool developed by Brian K. Lee and his colleagues in 2023 to simulate the connection between the chemical structure of odors and their olfactory perception properties.A principal odor map unifies diverse tasks in olfactory perception”, which has been published in the journal Science. POM uses graph neural network (GNN) technology to map the chemical structure of odor molecules into a high-dimensional space, thereby reflecting perceptual similarity rather than structural similarity. This method performs as well as some well-trained human "sniffers" in describing odor quality and can be used to predict odor intensity and perceptual similarity between odors.

The development of the POM not only helps researchers get closer to matching the molecular properties of odors with their perceptual properties, it also provides strong support for the development of new odorants. The researchers compiled a list of about 500,000 potential odors that have never been synthesized and mapped them in the POM to understand what they smell like. Exploring this space would require about 70 man-years of continuous sniffing time to collect data from a trained human sniffer.

The advantage of POM is that it can accurately represent known olfactory hierarchies and distance relationships, and can be generalized to a variety of odor prediction tasks. Compared with traditional cheminformatics models, POM performs better on several olfactory prediction tasks and successfully encodes a generalized mapping of structure-odor relationships. This provides a wide range of possibilities for odor prediction and paves the way for digital olfaction.