Figure 3 | Scientific Reports

Figure 3

From: Dynamic input-dependent encoding of individual basal ganglia neurons

Figure 3

GLMs fitted over different input statistics are not interchangeable. (A) Two GLMs (stimulus filter and post spike filter) of the same cell, generated using different mean DCs (100 and 66) added to the same frozen noise stimulus, which result in different firing rates: red – 24 Hz, blue – 15 Hz (B) Example predictions of the two GLMs in A to a novel stimulus with the higher DC. Top - the input stimulus. Middle – the responses of the experimental neuron (black) and the model neurons using the GLM calculated using the higher DC (red), and the lower DC (blue). Both model neurons’ mean firing rate is similar (24 Hz) to the rate of the real neuron, achieved by applying the proper constant DC. Bottom – the corresponding PSTHs. The correlation between the experimental PSTH and the model PSTH of the higher DC (red) is 0.89, and the lower DC (blue) is 0.6. (C) Generalization of GLMs calculated using different firing rates of the same cell leads to reduced accuracy. The GLM generated using a specific DC, which resulted in a specific firing rate, was then used as a model neuron to generate responses to a novel stimulus with different constant DCs. The model PSTHs were tested against the real responses of the cell using the correlation measure as in B. (D) Same as c, but for the whole population. GLMs were generated using a specific firing rate of a neuron, and their prediction accuracy was assessed (the mean of the whole population represented as a firing rate difference of zero, blue bar). The GLM was then used as a model neuron to predict responses of other firing rates of the neuron (represented as ± firing rate difference); the PSTHs correlations were measured, and their mean for each firing rate difference is presented. Significance is measured in relation to the mean PCCs of the original GLMs (firing rate difference of zero), and only significant mean correlations are presented. Black solid lines represent the exponential fitting functions for the positive (R2 = 0.48, p < 0.001) and negative (R2 = 0.32, p < 0.01) firing rate differences.

Back to article page