A probabilistic approach to demixing odors

Abstract

The olfactory system faces a hard problem: on the basis of noisy information from olfactory receptor neurons (the neurons that transduce chemicals to neural activity), it must figure out which odors are present in the world. Odors almost never occur in isolation, and different odors excite overlapping populations of olfactory receptor neurons, so the central challenge of the olfactory system is to demix its input. Because of noise and the large number of possible odors, demixing is fundamentally a probabilistic inference task. We propose that the early olfactory system uses approximate Bayesian inference to solve it. The computations involve a dynamical loop between the olfactory bulb and the piriform cortex, with cortex explaining incoming activity from the olfactory receptor neurons in terms of a mixture of odors. The model is compatible with known anatomy and physiology, including pattern decorrelation, and it performs better than other models at demixing odors.

Context- and Output Layer-Dependent Long-Term Ensemble Plasticity in a Sensory Circuit

“…This is again likely ascribed to the circuit within and/or beyond the OB, since input patterns from OSNs evoked by closely related mixtures were reported to be highly correlated between S+ and S− even after odor discrimination training (Abraham et al., 2004). The feedback inputs from the piriform cortex might be an attractive candidate as demonstrated by our circuit model (Grabska-Barwińska et al., 2017). Indeed, it was recently shown that in naive mice the cortical feedback enhances the pattern separation of MC but not TC assemblies through yet-to-be identified inhibitory circuits (Otazu et al., 2015).…”

“…We thus examined the effect of cortical feedback on context-dependent improvement of pattern separation in MCs. For this purpose, we modified an existing computational model that successfully solves “demixing” problems (Grabska-Barwińska et al., 2017), i.e., identifies the presence of a particular odor in a mixture through approximate Bayesian inference (see STAR Methods). Upon odor stimulation, excitatory inputs are relayed from OSNs to MCs and from MCs to cortical cells.…”

“…The details of the model are described in our previous work (Grabska-Barwińska et al., 2017). Briefly, our model circuitry consists of four layers and five groups of variables: ORNs, MCs, cortical cells, and two compartments for each cell in the granule cell layer, with firing rates r i m i , c i, g i and g ik respectively.…”

Robust odor coding across states in piriform cortex requires recurrent circuitry: evidence for pattern completion in an associative network

“…Notably, theoretical models have shown that responses in a recurrently-connected network can track input changes nearly-instantaneously 38 , whereas feedforward networks respond at a temporal delay, suggesting that recurrent excitation within PCx could facilitate short-latency, robust responses. PCx also sends centrifugal inputs to OB, and could return an accurate representation of the current stimulus to the OB, allowing comparison between ongoing input and an internally-generated prediction based on the retrieved activation pattern 39 . This process may augment the selective representation of early OB inputs in PCx.…”

“…We propose the initial activation of a small subset of superficial PCx neurons can drive reactivation of a stable ensemble of deeper cells, enabling the recovery of robust PCx odor responses despite degraded OB input. Interestingly, PCx cells in deep layer II project back to OB 54,55,57 , suggesting that an accurate representation of the current stimulus is returned to the OB allowing comparison between ongoing input and the retrieved activation patterns, and potentially implementing a predictive loop between OB and PCx to refine odor representations in both regions 39 .…”