Computational modeling of neuronal circuits: heterogeneous connectivity and nonlinear transformation in olfactory processing

The first processing circuit in the olfactory system (antennal lobe or olfactory bulb) exhibits several astonishing nonlinear features, but how these characteristics emerge collectively in olfaction remains an open question. In this thesis, we tackle three mysteries of olfactory processing. First, the olfactory circuit can simultaneously be robust against concentration fluctuation, independent of smaller concentration variation, yet sensitive to larger intermediate concentration change. Second, studies on olfactory bulb have shown that morphing between two odors across a series of mixture ratios generates representations in the output neurons that split the gradual gradient into two or three discrete clusters. Third, the antennal lobe of insects is capable of separating sensory representations from two different input odor signals. Surprisingly, our recent experimental work suggests that the same circuit generalizes the representations of other odor pairs. Here, we demonstrate how heterogeneous lateral inhibition and nonlinear intraglomerular transformation simultaneously contribute to these three distinct functions in the olfactory processing. In concert with two recent discoveries, the heterogeneity of recurrent network connectivity and the nonlinear signal transformation between olfactory neurons at different layers, our theoretical model enables putting a collective interpretation of mechanisms in concentration-invariant odor representations, discrete representations for odor mixtures, and olfactory coding in modulating separation between odor pairs. Our results indicate that inhibitory local neurons interconnecting glomeruli may play a central role to reshape odor representations by inhibiting specific output neuron ensembles. These outcomes of olfactory processing rely on the interactions between odor-evoked input patterns and the wiring of olfactory circuits