Odor prediction and aroma mixture design using machine learning model and molecular surface charge density profiles

Abstract

Structure-Odor Relationship (SOR) is the key to the understanding of the mechanism of odors, which further facilitates the odor prediction and aroma design. However, the modeling of SOR is still a challenge since the understanding of odor mechanism is unresolved. In this paper, an Artificial Neural Network (ANN) model is developed to generate the SOR model for aroma mixtures, in which the molecular surface charge density profiles (σ-profiles) are used as the descriptors. In this ANN model, the aroma mixture properties are the input, and the σ-profiles of the mixtures are the output, which can directly predict the σ-profiles of the aroma mixtures under certain odor requirements. Then, Euclidean distance-based ingredient screening method is applied to search the aroma mixture that fits the predicted σ-profiles from the ANN model. Finally, a Computer-Aided Aroma Design (CAAD) framework is formulated for the design of aroma mixtures. Two case studies of aroma mixture design are presented. The results are verified by the electronic nose, which highlight the effect of the developed SOR model as well as the CAAD framework.