Prediction of Odor Descriptor Group of Essential Oils from Mass Spectra using Machine Learning

Abstract

Olfactory perception is still a difficult task from its chemical properties to perceive the odor. In this paper, we report a computational method to predict the odor descriptor group from its mass spectrum. When the database only indicates the existence of each odor descriptor, only binary data are available. However, the prediction accuracy is very low because we cannot consider the similarities among descriptors. Thus, clustering of odor descriptors are necessary to make groups of similar odor descriptors. Although it is not easy to map from one to another as mass spectra dataset are highly dimensional and their structure are nonlinear, we use nonlinear dimensionality reduction on mass spectra and performs the hierarchical clustering to make odor descriptor groups. This model helps to predict a group of descriptors successfully through computer simulations.