Extended Data Fig. 3: Validation of imaging and signal extraction techniques.

(a) Map of identified cells (in green) overlaid on an imaging frame displaying the log of the standard deviation of individual pixels over all preprocessed, aligned and concatenated recording sessions for an example mouse. This visualization provides an intuition for the spatial extent of individual cells, but by itself cannot capture the quality of individual cells (see Extended Data Fig. 4). Scale bar: 100 μm. Five independent repetitions with data from different subjects produced qualitatively similar results. (b) Quantification of session alignment for individual subjects. We calculated pairwise mean squared errors (MSEs) between the cell maps of two aligned sessions. Shifts in the field of view that could not be aligned were visible as high MSE values between a given session and a set of well-aligned sessions. Data exclusions are displayed in red. (c) Analysis of background (BG) activity and the effect of lowpass normalization. For all cells of an example session (example cell displayed in grayscale), we calculated background activity traces using ring filters centered on the cell’s centroid (displayed in purple). Scale bar: 30 μm. (d) Activity traces for the two spatial filters from c with and without lowpass normalization. Without normalization, cell and background activities are highly correlated, indicating contamination of the cellular signal. With normalization, the correlation disappears and calcium transients can be resolved in the cellular signal. (e) Quantification of the correlation between cellular and background activity for all cells of an example session. Without lowpass normalization, most cells showed substantial positive correlations (median Pearson correlation coefficient = 0.60). Using the lowpass normalization, the median Pearson correlation coefficient dropped to 0.02, indicating that the lowpass filtering strategy employed in our pipeline successfully removes the majority of neuropil contamination from the cells’ activities. (f) Analysis of motion-related activity as a control for motion-related artifacts in neural activity. Average response to all ITI shuttles (black) of an example cell with individual example shuttles displayed in gray. This example cell shows motion-related activity based on calcium transients that are not consistent with artifacts based on microscope motion. (g) Quantification of motion-related responses over all recorded cells. Motion score is calculated as the mean z score in the 4 s after motion onset (at time 0). Many cells show positive (green) and negative (red) responses, but many cells (gray) are not strongly modulated by motion, indicating that there is no systematic motion artifact affecting all cells. (h) The time course of positively and negatively modulated cells is consistent with neural responses (slow and asymmetric) rather than motion-related artifacts, which would be expected to be fast and symmetric. (i) Quantification of peak latency for positively and negatively modulated cells. (j) Quantification of response symmetry for positively and negatively modulated cells. Symmetry index is calculated as (activity_{peak + 2 s} + activity _{peak − 2 s})/activity_peak, such that 0 indicates symmetric responses.