Fig. 5
From: Generalized leaky integrate-and-fire models classify multiple neuron types
Different mechanisms improve model performance for inhibitory and excitatory neurons. The traditional leaky integrate and fire model (GLIF1) yielded surprisingly high model performance. Overall, inhibitory models were more successful at reproducing spike times than excitatory models. Reset rules implemented on their own (GLIF2) decreased model performance. After-spike currents (GLIF3) improved inhibitory model performance, whereas a combination of both after spike currents and reset rules (GLIF4) were required to gain performance of excitatory models. The voltage-dependent adapting threshold (GLIF5) improved performance of excitatory models even more, but had only a slight effect on inhibitory models. The thick blue line denotes all excitatory neurons, the thick red line denotes all inhibitory neurons, and the thick black line is for all neurons. Thin lines are different transgenic lines. Accompanying data is available in Table 3. p-values for significant differences between GLIF model levels can be found in Supplementary Figures 8, 9, and 10. Briefly, for the “all”, “excitatory”, and “inhibitory” groupings the p-values are smaller than 0.01 (and often much smaller) for all but between the excitatory GLIF2 and GLIF3 (Supplementary Fig. 8). Differences between GLIF levels of different transgenic lines are sometimes statistically significant and sometimes not (Supplementary Figures 9 and 10)
