Predicting Odor Perception of Mixed Scent from Mass Spectrometry
Tanoy Debnath, Dani Prasetyawan, Takamichi Nakamoto
Abstract
The ability to perceive odors from chemical mixtures with subtle differences is difficult for humans. We cannot expect impression of the mixture of two original samples through sensory test. The essential oils themselves are complex mixtures. Moreover, if one mixes essential oils of two different groups (citrus & spicy), it will be difficult to predict the smell impression from this new blend of chemicals. However, the machine can predict the unexpected impression of the smell of new mixture. In this study, we synthesized binary-mixture mass spectra of essential oils numerically and then predicted their impression scores using neural networks. Next, we performed a human sensory test (2-Alternative forced choice) to distinguish the mixed scent from the original samples to validate the results of the neural network model. The experimental results suggest that the machine learning model can predict the odor of the mixture which has been validated by the human sensory test. Our main contribution here was to show the fundamental method to predict odor perception of the mixed scent from mass spectrometry data.