Figure 3
Hierarchical clustering of RGC electrical temporal filters estimated by STA. (a–c) Electrical temporal filters of all reliably responding RGCs from wt retina recovered from the STA fit of the LNP model, displayed separately for the three clusters identified by the hierarchical clustering algorithm. Thin lines are individual cell filters, thick lines indicate the average filter for one cluster. (d,e,f) 1st, 2nd and 3rd principal component (PC) recovered from principal component analysis of the ensemble of temporal filters from all cells. (g) Dendrogram showing the separation of consecutively joined clusters along the clustering metric (distance in euclidean space). (h) Scatter plot of the projections of the temporal filters onto the 1st and 2nd PCs (shown in (d,e)). Colors indicate cluster assignment. Grey dots indicate the projections of filters of RGCs from rd10 retina, projected onto the same PCs. The lower inset shows the filters of all rd10 RGCs whose filter projections are negative in PC2; the upper inset shows the filters of all rd10 RGCs whose filter projections are positive in PC2. The black arrow marks the cell for which the assignment to clusters did not agree between STA and MLE estimate of the filters. (i) STA estimates of the filters of example RGCs from wt retina shown in Fig. 2b (rows 4-12, same order), with cluster assignment indicated by color. (j) Same as (i), but for MLE estimates of the filters. (k) Stacked histogram of the distribution of reliability indices (RIs) for all cells with \(RI > 0.15\), color-coded according to cluster assignment. (l) Distribution of distance to the edge of the closest active stimulation area of all responsive cells (\(RI > 0.15\)), color-coded according to cluster assignment. (m) STA estimates of the filters of example RGCs from rd10 retina shown in Fig. 2e (same order). (n) Same as (m), but for MLE estimates of the filters.
