Extended Data Fig. 4: Quality control for identified neurons.
From: Population-level coding of avoidance learning in medial prefrontal cortex
(a) Features of an example cell that was accepted in the annotation process. A cell is defined by its activity trace over all sessions (top left, individual imaging sessions are indicated as different gray shades, see top right for zoom-in) and its spatial filter (bottom middle). Scale bar: 100 μm. We detect events (red dots) as peaks in the activity trace that deviate 3 standard deviations from the mean. We then use these detected events to calculate a mean transient (bottom left in red, individual events in black) and to display snapshots of the images that caused the peaks in the activity trace (bottom right). Cells are accepted if they have (1) a clear and appropriately shaped spatial extent, (2) a stable activity trace with well-identified peaks, (3) a mean transient with fast rise and slow decay as expected from calcium indicator kinematics and (4) if the event snapshots consistently resemble the spatial filter (displayed on the upper left of the snapshot matrix). Snapshots that do not resemble the filter indicate contamination through another cell. (b) Example of a rejected cell with a noisy spatial filter and activity trace, a symmetric mean transient and inconsistent event snapshots. (c) Distributions of mean numbers of events per day. (d) Distributions for the coefficient of variation of the number of events per day (ratio of s.d. and mean). The CV tends to be substantially below 1 (median = 0.52 for accepted cells), indicating that the distribution of the number of events per day does not fluctuate far from the mean over days (displayed in c), which suggests stable activity levels over sessions. (e) Distributions of symmetry index calculated using the mean transient as (activitypeak + 1 s + activitypeak − 1 s)/activitypeak. (f) Distribution of filter diameter for accepted (green) and rejected cells (red). (g) Distributions of snapshot dissimilarity values. Snapshot dissimilarity is calculated as the mean of the mean squared error (MSE) between the filter and each event snapshot.
