Automated optimization of a reduced layer 5 pyramidal cell model based on experimental data

Abstract

The construction of compartmental models of neurons involves tuning a set of parameters to make the model neuron behave as realistically as possible. While the parameter space of single-compartment models or other simple models can be exhaustively searched, the introduction of dendritic geometry causes the number of parameters to balloon. As parameter tuning is a daunting and time-consuming task when performed manually, reliable methods for automatically optimizing compartmental models are desperately needed, as only optimized models can capture the behavior of real neurons. Here we present a three-step strategy to automatically build reduced models of layer 5 pyramidal neurons that closely reproduce experimental data. First, we reduce the pattern of dendritic branches of a detailed model to a set of equivalent primary dendrites. Second, the ion channel densities are estimated using a multi-objective optimization strategy to fit the voltage trace recorded under two conditions - with and without the apical dendrite occluded by pinching. Finally, we tune dendritic calcium channel parameters to model the initiation of dendritic calcium spikes and the coupling between soma and dendrite. More generally, this new method can be applied to construct families of models of different neuron types, with applications ranging from the study of information processing in single neurons to realistic simulations of large-scale network dynamics.