Figure 4.

Inferring the spatial resolution of optogenetically and electrically stimulated RGCs from a support vector machines classifier. (A) Classification results are shown as f₁ score values for data obtained from optogenetic experiments. The different colors denote different experimental days. On the upper side of the plot are data from the 2.5 Hz switching frequency and on the lower side are data from the 10 Hz switching frequency. A classifier’s performance is considered to be significant when the f₁ score is above the “chance level” which, in this case, we considered to be at 0.5. With the exception of one recording day, the metrics show a notably high value for all of the spatial frequencies considered, reaching saturation at around 30 μm for 2.5 Hz and 50 μm for 10 Hz. (C) Classification results for data obtained from electrical stimulation experiments. The different colors correspond to different experimental days. The saturation is reached at 64 μm for all datasets considered. However, for some experimental days, we note particularly high values for the 32 μm grating width as well. (B,D) Feature importance plot showing the most significant 20 features (vectors containing firing rate values for individual stimulus repetitions) for the classifier. In panel (B) there is an exemplary plot for good classification results, for the case of the 50 μm grating width, with a 10 Hz pattern reversal from an optogenetic stimulation dataset, while in panel (D) the plot shows an example of the feature importance for the 32 μm grating width, in the case of an electrical stimulation dataset.