SPLTRAK Abstract Submission
Drawing the Borders of Olfactory Space
Emily J. Mayhew, Charles J. Arayata, Jonathan Magill, Lindsey L. Snyder, Chung Wen Yu, Joel D. Mainland
Monell Chemical Senses Center, Philadelphia, PA, United States

In order for researchers to rigorously sample it, the boundaries of olfactory space must be defined. In studies of vision and audition, stimuli can be chosen to span the visible or audible spectra; in olfaction, the axes and boundaries that define olfactory space are unknown. Without these defined boundaries, it is also difficult to quantify the number of possible odorous molecules. An estimate of 10,000 odorants is often cited, but the quality and source of this figure is unclear.  A model proposed by Boelens (1983) draws a border between odorous and odorless molecules using boiling point (bp) and hydrophobicity (Kow), but the accuracy of the model and the data used to construct the model are undefined. Here we selected 128 diverse compounds that span a wide range of chemical space. Human subjects evaluated these compounds in 3-alternative forced choice discrimination tests. We reserved 30 molecules to form a validation set, then employed several machine learning algorithms to iteratively train and test models on the remaining 98 compounds. Random Forest (RF) models achieved the highest accuracy (86 ± 2.2%) and showed improvement over logistic regression models constructed using only bp and Kow (77 ± 1.1%). We identified several physicochemical properties, in addition to bp and Kow, that help distinguish odorous from odorless molecules, including electrophilicity and structure linearity. Applying our current model to the GDB17, a database of all possible molecules with 17 or fewer heavy atoms (C, N, O, S, or halogens), we estimate that over 27 billion possible compounds are odorous. In drawing the borders of olfactory space more clearly, we provide necessary guidance for olfactory researchers to design experiments that representatively samples olfactory space.