Neuronal encoding models of complex receptive fields: A comparison of nonparametric and parametric approaches

Parametric models have been widely used to estimate conditional intensity functions of neuronal spike train point processes and are efficient to construct from experimental data. Furthermore, parametric models are easy to interpret. However, neurons that have more complex receptive fields may not be sufficiently characterized through parametric modeling since it imposes strict structure on the encoding fields. In this paper, we consider a pyramidal neuron recorded from the rat hippocampus, called a “place” cell, that has a diverse apparently multimodal receptive field that encodes information about the spatial position while the rat freely-forages in a circular environment. We construct encoding models for this place cell using two nonparametric modeling approaches, our recently developed band-limited maximum likelihood (BLML) estimator and a kernel density estimator (KDE); and compare them to models constructed using two parametric approaches that have been previously applied to these neurons. We found that the BLML and KDE better capture the complex receptive field of the studied cell as measured by the KS-statistic and log-likelihood.