The paper presents a comparative analysis of color vision and olfaction, highlighting the unique challenges of olfaction due to the complexity of stimuli, high dimensionality of the sensory apparatus, and the lack of a clear "ground truth". It argues for the centrality of odorant-receptor interactions in developing a theory of olfaction, which could have widespread industrial applications and enhance our understanding of smell.
As an initial use case, the authors demonstrate the application of machine learning-based classification of neural responses to odors recorded in the mouse olfactory bulb using calcium imaging. The goal is to create an "MNIST database for olfaction", called 'oMNIST', to catalyze olfaction research in Neuroscience and Data Science, similar to how the availability of standard datasets drove progress in computer vision.
The authors describe a typical neural data pipeline, from acquisition to analysis, and present results showing that even a simple model can effectively separate the odor space based on the aggregated spatial representations of glomerular activation patterns. The error patterns reveal potential sources of noise, including inter-subject variability, similarity of activation patterns for certain odorants, and spurious activation outside the olfactory bulb. The authors discuss the next steps in building a generative model of olfaction that can learn the joint distribution of odorant-receptor interactions, leveraging the latent biological variables involved.
Til et andet sprog
fra kildeindhold
arxiv.org
Vigtigste indsigter udtrukket fra
by Vivek Agarwa... kl. arxiv.org 04-09-2024
https://arxiv.org/pdf/2404.05501.pdfDybere Forespørgsler