Modern technology makes it possible to capture a visual scene as a photograph, alter it, send it to another country nearly instantaneously, and store it without concern for degradation. None of this is currently possible in olfaction. Although perfumers and flavorists are adept at mixing odorous molecules to produce a desired perceptual effect, the rules underlying this process are poorly understood at a quantitative level. Current methods for displaying odors to a subject are akin to requiring a Polaroid of every visual stimulus of interest. A more efficient method for probing the olfactory system would be to use a set of 'primary odors'?some limited number of odors from which all other complex odors could be reproduced by appropriate mixtures. Both auditory and visual stimuli have been digitized, and this will eventually be possible in olfaction as well. Predicting odor from chemical structure has been a problem in the field since its inception, but recent advances in machine learning algorithms have made great progress in analogous problems, such as facial recognition. The research proposed here will combine these machine learning techniques with high quality human psychophysics to understand how to predict the smell of a molecule or mixture of odorants, which will ultimately help improve our understanding of disease diagnosis using odors as well as eating-related health and illness.
HEALTH RELEVANCE The sense of smell plays a critical role in preferences and aversions for specific foods. The proposed research will combine machine learning techniques with high quality human psychophysics to create a model that can predict the smell of odorous molecules. This model will allow us to describe and control odors, which will increase our understanding of food preference and eating-related health and wellness.
