Optimal fragrances formulation using a deep learning neural network architecture: A novel systematic approach

Abstract

Human civilization has been economically exploring the enjoyable smell of substances for centuries, giving rise to multi-billion-dollar business. Few works have addressed the formulation of perfumes using a systematic approach based on computational techniques. Thus, the objective of the present work is to develop a novel systematic strategy for optimal perfume design. The strategy comprises a deep learning model trained from high-fidelity simulations, an objective function that reflects the desirable spectrum of the perfume, and a meta-heuristic optimization method. It was applied to define the perfume composition that produces an odor spectrum of pine forest and floral while minimizing non-desirable odors. Hence, we propose an objective function to encode the peculiarities of a fragrance design comprising the question: Which perfume composition attains the desirable odor spectrum across time and space? The results demonstrated the methodology value in fragranced product design by offering a framework to handle the formulation problem.