Presentation Details
| Biologically Informed Artificial Intelligence Models for Predicting Human Sensory Perception of Scents and Flavours Luana P.Queiroz1, 2, Ícaro S.C.Bernardes2, Ana M.Ribeiro1, Bernardo M.Aguilera-Mercado3, Idelfonso B.R.Nogueira2. 1LSRE-LCM, ALiCE, Faculty of Engineering, University of Porto, Porto, Portugal.2Chemical Engineering Department, Norwegian University of Science and Technology, Trondheim, Norway.3Corporate Fragrance R&D, The Procter & Gamble Company, Cincinnati, OH, USA |
Abstract
Human sensory perception is shaped by scent and flavor intensity, a key driver of consumer acceptance. Predicting olfactory intensity is a central challenge; traditional sensory panels are reliable but expensive and difficult to scale. Furthermore, the industry faces a loss of tacit expertise as experienced perfumers retire. While computational models offer scalability, they often lack biological plausibility and fail to capture the nonlinear, compound-specific dose–response behavior governing intensity. These gaps stem from scarce datasets and the difficulty of modeling complex psychophysical relationships. To overcome this, we introduce a biologically informed machine learning framework that integrates psychophysical principles into the learning process. The model embeds Hill’s law to constrain predictions to realistic dose–response behavior, preventing the unrealistic linear growth of conventional models. We set the minimum perceived intensity to 1.4 and estimate maximum concentration via log vapor pressure, sampling dose–response curves at ten points. These representations are coupled with Graph Neural Networks (GNNs) using molecular graphs to link chemical structure to perceptual outcomes. Validation in the perfume domain showed that embedding odor-character representations within the intensity prediction pipeline reduced mean squared error by approximately 20% compared to structure-only baselines. This study demonstrates that embedding principles like Hill’s law into AI enhances predictive power and interpretability, offering a tool for rational fragrance design that reduces reliance on costly sensory panels.
No part of this publication may be reproduced, distributed, or transmitted in any form or by any means, including photocopying, recording, or other electronic or mechanical methods, without the prior written permission of the author.
No part of this publication may be reproduced, distributed, or transmitted in any form or by any means, including photocopying, recording, or other electronic or mechanical methods, without the prior written permission of the author.
Content Locked. Log into a registered attendee account to access this presentation.