Predicting pleasantness from joint perceptual and structural embedding space
Nicholas McCarthy1, Qurrat Ul Ain1, Jeremiah Hayes1
1Accenture Dock
Olfactory pleasantness of a single compound or a compound mixture refers to the quality of being perceived as having a pleasing smell. Pleasantness perception can generally be influenced by the intensity of a stimuli and individual response time, as demonstrated by several psychophysical techniques. Oftentimes, when combined in equal intensities, compound mixtures may yield similar smell (known as ‘olfactory white’), making it even harder to predict the pleasantness. It is also very difficult to correctly predict the smell of an odorant solely from its physiochemical properties or structural features. Molecules with a similar structure can produce very different odours. Conversely, structurally diverse molecules can sometimes produce identical smells.
In this work we address the challenge of mixture pleasantness prediction by creating a joint embedding space of molecular properties, chemical structures, and stimuli percepts (odour, taste and odour strength) using knowledge graphs. These embeddings are then used to train an LSTM recurrent neural network to predict the pleasantness of n-ary compound mixtures with improved accuracy. The novelty of our approach is the utilisation of descriptors (physiochemical, structural, perceptual) and creation of an embedding space that enables the accurate approximation of previously unseen molecules; further facilitating the predictive power of model trained on embeddings.
Keywords: Olfactory perception, Odour, knowledge graph, embedding
References:
[1] Sell CS. On the unpredictability of odor, 2006. Angew. Chem. Int. Ed Engl.
[2] Weiss T. et al., Perceptual convergence of multi-component mixtures
in olfaction implies an olfactory white, 2012. PNAS
[3] Keller A. et al., Predicting human olfactory perception from chemical features of odor molecules. 2017. Science
[4] Lapid H. et al., Prediction models for the pleasantness of binary mixtures in olfaction. 2008. Chem. Senses
[5] Khan RM. et al., Predicting Odor Pleasantness from Odorant Structure: Pleasantness as a Reflection of the Physical World. 2007. JNeurosci.
[6] Haddad R. et al., Predicting Odor pleasantness with an electric nose. 2010. PLOS Computational Biology
