Computational Proxies for Dopamine-Mediated Learning in the Olfactory System of Mosquitoes

Abstract

Olfactory learning in mosquitoes plays an essential role in determining their host preference and has significant epidemiological consequences. Dopamine effects on olfactory learning in mosquito brain are still unclear, although it is known to actively modulate the neural activity of Projection Neurons (PNs) in the antennal lobe (AL). To investigate this, we study the firing patterns of the PNs before, during, and after the superfusion of dopamine over the brain of the mosquito. In particular, we compute the fixed points in classifying odor space corresponding to each firing pattern. Comparison of the locations of fixed points in each phase of the experiment indicates that the superfusion of dopamine causes fixed points to dislocate. To investigate how neurons facilitate such function in the AL, we extend a lateral inhibition firing-rate model of the AL to include dopamine neurons. We then implement various time-dependent learning rules (gradient based, Hebbian learning, reward-based) that minimize the norm difference between expected firing rates and observed ones to quantify the connectome weights of the Dopaminergic (DA) neurons. The comparison of the predicted trajectories and updated connection matrices reveal that these update methods have different element-to-element modulation yet the similar pattern of modulation during and after the application of dopamine. Furthermore, the successful generation of the firing patterns of the PNs and the relative weights of the lateral inhibition neural connectomes suggest that the DA neurons selectively modulate the neural responses in the AL and follow a selective recovery pattern.