A framework for predicting odor threshold values of perfumes by scientific machine learning and transfer learning

Aug 5, 2026

Knowledge of odor thresholds is very important for the perfume industry, yet measuring them is difficult because it involves complex experimental procedures with human subjects. To address this challenge, a modeling framework is proposed using scientific machine learning. A transfer learning strategy leverages a graph convolutional network that predicts semantic odor descriptors; its outputs feed a feedforward neural network that estimates odor threshold values from molecular structures. Performance is compared against a non-transfer benchmark, literature correlations, and a dummy regressor, showing improved predictive accuracy.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages12

Reading Time30 min

AccessFree to download

Introduction

Odor properties and relevance

Types of odor thresholds

Measuring odor thresholds

Dilution-to-threshold methods

Olfactometry and scentometry

Heliyon 9 (2023) e20813 Contents lists available at ScienceDirect Heliyon journal [homepage:](homepage: www.cell.com/heliyon)[ www.cell.com/heliyon](homepage: www.cell.com/heliyon) | A framework for predicting odor threshold values of perfumes by scientific machine learning and transfer learning
Luis M.C. Oliveira a, b, **, Vinícius V. Santana a, b, Alírio E. Rodrigues a, b, Ana M. Ribeiro a, b, Idelfonso B. R. Nogueira c, *
a LSRE-LCM-Laboratory of Separation and Reaction Engineering – Laboratory of Catalysis and Materials, Faculty of Engineering, University of Porto, Rua Dr. Roberto Frias, 4200-465, Porto, Portugal
b ALiCE-Associate Laboratory in Chemical Engineering, Faculty of Engineering, University of Porto, Rua Dr. Roberto Frias, 4200-465, Porto, Portugal
c Department of Chemical Engineering, Norwegian University of Science and Technology, Sem Sælandsvei 4, Kjemiblokk 5, Trondheim, Norway | | | | --- | --- | --- | | A R T I C L E I N F O | | A B S T R A C T | | Keywords: Transfer learning Odor thresholds Machine learning
Artificial neural networks Graph convolutional networks Feedforward neural networks | | Knowledge of odor thresholds is very important for the perfume industry. Due to the difficulty associated with measuring odor thresholds, empirical models capable of estimating these values can be an invaluable contribution to the field. This work developed a framework based on scientific machine learning strategies. A transfer learning-based strategy was devised, where information from a graph convolutional network predicting semantic odor descriptors was used as input data for the feedforward neural network responsible for estimating odor thresholds for chemical substances based on their molecular structures. The predictive performance of this model was compared to a benchmark odor threshold prediction model based on molecular structures that did not utilize transfer learning. Furthermore, the prediction was compared to a correlation previously proposed in the literature and a dummy regressor. Results demonstrated that the transfer learning-based strategy displayed a better predictive performance, suggesting this technique can be useful for predicting odor thresholds. | 1. Introduction An odor, as sensed by a human individual, has two relevant properties: intensity (the strength of the smell) and character (what the smell is like). The way these properties are sensed is influenced by an individual’s prior experiences and expectations. They determine the extent to which they can be affected by odorous air pollution [1]. Odor thresholds can be divided into two types: detection threshold (ODT), the level at which the human olfactory sense can detect a certain smell but not recognize it, and recognition threshold, the level at which that smell can be recognized [2]. Knowledge of thresholds is very important for the industry, not only in terms of process functioning (as in the perfume industry) but also in terms of safety, as many substances can have severe negative health effects at certain concentrations. There are several methods for measuring odor thresholds. The most widely used type of methodology is the dilution-to-threshold * Corresponding author. ** Corresponding author. LSRE-LCM - Laboratory of Separation and Reaction Engineering – Laboratory of Catalysis and Materials, Faculty of Engineering, University of Porto, Rua Dr. Roberto Frias, 4200-465, Porto, Portugal. E-mail addresses: [up201403344@edu.fe.up.pt](up201403344@edu.fe.up.pt) (L.M.C. Oliveira), idelfonso.b.d.r.nogueira@ntnu.no (I. B. R. Nogueira). [https://doi.org/10.1016/j.heliyon.2023.e20813](https://doi.org/10.1016/j.heliyon.2023.e20813) Received 22 March 2023; Received in revised form 27 August 2023; Accepted 7 October 2023 Available online 10 October 2023 2405-8440/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license ([http://creativecommons.org/license

FAQs

Why are odor threshold values important in the perfume industry?

Odor thresholds guide both process functioning and safety considerations, since many substances can pose health risks at certain concentrations.

How does the proposed transfer learning framework work?

A graph convolutional network predicts semantic odor descriptors from molecular structures, and those predictions are used as input for a feedforward neural network that estimates odor threshold values.

What benchmarks were used to evaluate the model?

The approach was compared with a benchmark prediction model that does not use transfer learning, as well as a previously proposed literature correlation and a dummy regressor.

Aug 5, 2026

Upload by Theodore