Figure 6
Model prediction of firing rates in wt retina. (a) Linear filters of three example cells from different clusters, derived from the STA fit (solid lines) and from the MLE fit (dashed lines); displayed at different scales in arbitrary units. To create the firing rate prediction, the filter response of the stimulus snippets with these filters was computed, and then the nonlinearity was applied. (b) Nonlinearity of the STA model; the nonlinearity of the MLE was the standard sigmoid function. (c) True firing rate, binned into 1 ms bins (black histogram) and the firing rate prediction of the STA fit of the model (colored traces) for three example cells; the highlighted region shows differential responses of cells with different types of filters, which are correctly predicted by the model. Decimal numbers in upper right corner indicate the performance of the model for that specific cell. (d) Same as (c) for the MLE fit of the model. (e) Prediction performance of the MLE fit plotted against the prediction performance of the STA fit; colors indicate cluster assignment. Small black dots represent RGCs that did not pass the reliability criterion of RI \( > \,0.15\), but are shown here for completeness. (f) Prediction performance of the STA fit plotted against reliability index. Colors indicate clusters assignment. Small black dots as in (e). (g) Prediction performance of the STA fit of the full LNP model plotted against the prediction performance of only the linear stage of the LNP model. Colors indicate cluster assignment. Small black dots as in (e).
