A mouse bio-electronic nose for sensitive and versatile chemical detection

Abstract

When it comes to simultaneous versatility, speed, and specificity in detecting volatile chemicals, biological olfactory systems far outperform all artificial chemical detection devices. Consequently, the use of trained animals for chemical detection in security, defense, healthcare, agriculture, and other applications has grown astronomically. However, the use of animals in this capacity requires extensive training and behaviorbased communication. Here we propose an alternative strategy, a bio-electronic nose, that capitalizes on the superior capability of the mammalian olfactory system, but bypasses behavioral output by reading olfactory information directly from the brain. We engineered a brain-machine interface that captures neuronal signals from an early stage of olfactory processing in awake mice, and used machine learning techniques to form a sensitive and selective chemical detector. We chronically implanted a grid electrode array on the surface of the mouse olfactory bulb and systematically recorded responses to a large battery of odorants and odorant mixtures across a wide range of concentrations. The bio-electronic nose has a comparable sensitivity to the trained animal and can detect odors on a variable background. We also introduce a novel genetic engineering approach designed to improve the sensitivity of our bio-electronic nose for specific chemical targets. Our bio-electronic nose outperforms current detection methods and unlocks a wide spectrum of civil, medical and environmental applications. where x Iog 10 of the concentration, R min and R max mark minimal and maximal responses, respectively, EC 50 is the concentration at half maximal response, and n is the Hill slope. The parameters were estimated using nonlinear regression. The coefficients were estimated using iterative least square estimation. (Matlab nlinfit). P-values in fig. 4 were calculated using two-tailed t-test (Matlab)

Across-animal odor decoding by probabilistic manifold alignment

“…We used a linear SVM with L2 regularization to decode stimulus identity from the latent trajectories Frequency of the MAP estimate detecting odor A, as a function of the mixture ratio (middle) and three example posteriors (right); dashed lines show individual simulations, with average in solid line. [12,24]. All controls were constructed using the library scikit-learn [25].…”

“…We tested our model on neural recordings from a 64-site grid-electrode stereotaxically implanted over the dorsal part of the olfactory bulb in five mice [12]. We simultaneously recorded a pressure readout of the animal's sniff cycle.…”

“…CS supervised the project. PHV collected the neural data in DR lab, previously published in Shor et al 2020 [12].…”

“…The alignment of low-dimensional neural manifolds is equally important in brain-computer interface (BCI) applications, where the assumption of a common low dimensional structure in neural responses is used to compensate for instability in neural recordings and ensure robust performance over time [8,9]. This is not only restricted to motor control; in the sensory domain, a new generation of chemical detectors exploit the unsurpassed sensitivity of rodent olfactory receptors, but bypass the limitations of training animals to report odor identity by directly decoding it from neural responses [10,11,12]. Unfortunately, the applicability of this idea is limited by the need to learn the mapping between neural activity and chemical identity on an animal-by-animal basis, which involves costly data collection [13].…”

“…The basic structure of the rodent olfactory bulb (OB) is well-understood [14]: a vast number of different olfactory receptors project to distinct OB subpopulations (glomeruli) whose spatial organization is largely preserved across animals [15,16]. Odor identity is encoded in low-dimensional transients of their population dynamics [1,17,12], though how the glomerular topography translates into stimulus-dependent network dynamics is less clear. Nonetheless, given the anatomical stereotypy Different odors elicit distinct trajectories in a shared neural latent dynamical space, with animal-specific measurements resulting in distinct observations.…”