Presentation Details
| A Low-Dimensional Code for Perceptual Similarity in Olfaction Walter Bast1, Cina Aghamohammadi2, Priyanka Gupta1, Tatiana Engel2, Florin Albeanu1. 1Cold Spring Harbor Laboratory, Cold Spring Harbor, NY, USA.2Princeton Neuroscience Institute, Princeton, NJ, USA |
Abstract
Despite recent advances, the relationship between odorant receptor (OR) activity patterns and olfactory percepts remains unclear. To investigate how OR response spectra relate to perceived stimulus similarity, we developed a novel approach that exploits the anatomical convergence of ORs onto distinct glomeruli to control sensory input at the level of individual OR types. Using two-photon and widefield imaging, we identified numerous glomeruli and determined their responses to 121 odorants. We then created synthetic olfactory stimuli by optogenetically activating selected glomerular combinations. To determine perceptual distances between these glomerular sets, we trained mice to report perceived differences between test stimuli and a reference glomerular pattern. We find that individual glomeruli within the reference set differ in perceptual relevance, as activation of some glomeruli elicits larger perceptual changes than others. To investigate whether the perceptual similarity among patterns is primarily determined by their glomerular odorant response profiles, we developed an autoencoder-based method to extract latent factors from odorant responses. This approach captured the responses of ~40 glomeruli per animal with ~90% accuracy using only ~12 dimensions. A behavioral model trained on these latent factors successfully predicted individual animals’ responses to novel optogenetic stimuli. We are currently using this model to search for glomerular patterns with varying degrees of perceptual similarity to the reference percept, while confirming these predictions experimentally. Our results contribute to understanding how OR activations translate into olfactory percepts, and suggest that olfactory perception is low-dimensional and inherently structured for efficient odor representation.
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.