Fig. 5: Tracking neurons over many recordings.

a, The match probability P(match) of every pair of neurons in every pair of sessions, for an example mouse (ID1). The ticks on the axes indicate different recording sessions. b, Top: an illustration of unit tracking for four example units across three recordings, with match probabilities ranging from 0.00 to 0.99, sorted from high to low (brackets). Bottom: the outputs of the three versions of the algorithm to track neurons across many recordings, showing connections that support (green) or block (red) units being given the same identity (same color). For example, the ‘default’ version of the algorithm finds two distinct units, one appearing in all 3 days (black) and one appearing only on day 2 (blue). c, A population of neurons that was tracked over 195 days in the example mouse. Other neurons on the probe are not shown. Note that some neurons can disappear on some days and reappear on other days. d, The presence (black) of a unique neuron across recordings, sorted by first appearance. e, The average probability of tracking a unit P(track) (±s.e.m.) as a function of days between recordings. The number of tracked neurons was divided by the total number of neurons available in the future recording (negative Δdays) or past recording (positive Δdays). The number of datasets per bin is indicated in g. f, The average AUC values (±s.e.m. across datasets) when comparing the functional similarity scores of tracked versus nontracked neurons, for ISI histogram correlations (black), correlation with reference population (red) and responses to natural images (blue). The number of datasets per bin is indicated in g. g, The number of datasets per bin.