Design of fragrance molecules using computer-aided molecular design with machine learning

Abstract

The robust growth of the flavour and fragrances market necessitates the need for a model-based method for the identification of fragrance molecules with favourable product attributes. In this work, a computer-aided molecular design (CAMD) based approach was developed to design fragrance molecules. Hyperbox classifiers are implemented for predicting fragrance properties due to their ability to generate transparent results. The resulting models can be interpreted as disjunctive decision support rules that establish the quantitative relationship between the structural parameters of molecules to their odour characteristics. In addition, other relevant properties are modelled using group contribution (GC) models into the CAMD framework. In this work, molecular signature descriptors were used to formulate a CAMD model that connects different prediction models and machine learning algorithm in a common framework. A case study was conducted to illustrate the application of the proposed methodology in the design of fragrance additives used in body lotions.