Schematic depiction of drift:

a. Mice were implanted with a 4-shank Neuropixels 2.0 probe in visual cortex area V1. b. he location of a unit recorded on the probe. In this hypothetical case, the same color indicates unit The black unit is missing on day 48, while the turquoise star is an example of a new unit. Tracking and blue units across all datasets and determine that the black unit is undetected on day 48. c. Two eforms of units recorded in two datasets that likely represent the same neuron, based on similar the average waveform on one channel across 2.7 milliseconds. The blue traces are waveforms on channels (two rows above, two rows below, and one in the same row) from the first dataset (Day ected, are from the second dataset. Waveforms are aligned at the electrodes with peak amplitude, different on the two days.

The EMD can detect the displacement of single units:

a. Schematic of EMD unit matching. Each blue unit in day 1 is. Dashed lines indicate the matches to be found by minimizing the weighted sum of physical and nd filled circles show positions of units in days 1 and 2, respectively. Arrows indicate matching using ts the match direction; upward matches found with the EMD are in red and downward in black. Solid ce within 15μm, while a dashed line indicates a z distance > 15μm. Expanded view shows probe area gram of z-distances of matches (black and red bars) and kernel fit (light blue solid curve). The light ode (dm = 15.65μm). The dark blue dashed line shows the imposed drift (di = 12μm). The red region m of the mode. The EMD needs to detect the homogeneous movement against the background, i.e. re unlikely to be the real matches due to biological constraints.

The ROC curve of matching accuracy vs. distance.

The blue curve shows the accuracy for reference units. The red line nce units included. The solid vertical line indicates the average z distance across all reference pairs dashed vertical black line indicates a z-distance threshold at z = 10μm.

Recovery rate, accuracy and putative pairs:

a. The histogram distribution fit for all KS-good units (top) and reference units alone (middle). False positives for reference units are defined as units matched by EMD but not matched when using receptive fields. The false positive fraction for the set of all KSgood units is obtained by integration. z = 10μm threshold has a false positive rate = 27% for KSgood units. b. Light blue bars represent the number of reference units successfully recovered using only unit location and waveform. The numbers on the bars are the recovery rate of each datatset, and the red portion indicates incorrect matches. Incorrect matches are cases where units with a known match from receptive field data are paired with a different unit by EMD; these errors are false positives. The green bars show matching accuracy for the set of pairs with z-distance less than the 10μm threshold. The orange portion indicates incorrect matches after thresholding. The false positives are mostly eliminated by adding the threshold. Purple bars are the number of putative units (unit with no reference information) inferred with z-threshold = 10μm.

Number of reference units (deep blue, dark orange and green for different subjects), putative (medium green, medium orange and blue) units, and mixed units (light green, yellow, and light blue) tracked for different durations. The loss rate is similar for different chain types in the same subject. Note that chains can start on any day in the full set of recordings, so the different sets of neurons have chains with different spans between measurements.

Example mixed chain:

a. Above: Firing rates of this neuron on each day (Day 1, 2, 13, 23, 48). Below: Firing rate fractional change compared to the previous day. b. Visual response similarity (yellow line), PSTH correlation (orange line), and visual fingerprint correlation (blue line). The similarity score is the sum of vfp and PSTH. The dashed black line shows the threshold to be considered a reference unit. c. Spatial-temporal waveform of a trackable unit. Each pair of traces represents the waveform on a single channel. d. Estimated location of this unit on different days. Each colored dot represents a unit on one day. The orange squares represent the electrodes. e. The pairwise vfp and PSTH traces of this unit.

Summary of dataset:

a. The recording intervals for each animal. A black dash indicates one recording on that day. b. All tex V1 with a 720 μm section of the probe containing 96 recording sites. The blue arrow indicates mples of visual fingerprint(vfp) and peri-stimulus time histogram(PSTH) from a high correlation (left shold (right column) correlation unit. Both vfp and PSTH values vary from [-1,1]. d. Kilosort-good and al AL032, including units from all four shanks.

Distribution of waveform L2 similarity change per dataset for each neuron group and across all neurons.

Box plots ns, and 75% percentile. Whiskers at the ends of the box plot show maximum and minimum values. t comparisons, i.e. (number of units)×(number of datatsets - 1).

Distributions of individual unit location changes over whole chains (top) and unit location changes between pairs of datasets (bottom), for each neuron group and across all neurons. Box plots indicate 25% percentile, medians, and 75% percentile. Whiskers at the ends of the box plot show maximum and minimum values. In the top plot, n and N are the number of units. In the bottom plot, n and N are the number of unit comparisons, i.e. (number of units)×(number of datatsets - 1).

Distribution of firing rate fold change per dataset for each neuron group and across all neurons. Box plots indicate 25% percentile, medians, and 75% percentile. Whiskers at the ends of the box plot show maximum and minimum values. n and N represent the number of units.

The visual fingerprint and PSTH change distributions per dataset for each neuron group and across all neurons. Box plots indicate 25% percentile, medians, and 75% percentile. Whiskers at the ends of the box plot show maximum and minimum values. n and N are the number of unit comparisons, i.e.(number of units)×(number of datatsets - 1).

The similarity score distribution per dataset for each neuron group and across all neurons.

Box plots indicate 25% percentile, medians, and 75% percentile. Whiskers at the ends of the box plot show maximum and minimum values. n and N are the number of observations of the units, i.e. Σunits (observations of this unit)

An example similarity score (vfp + PSTH) heatmap from animal AL032 shank 2 Kilosort-good units between day 1 and 2.

Each small square represents the similarity score (value range from [-2,2]) between one unit from day 1 and one unit from day 2. A warm colored square indicates a higher score. The clusters are ordered by their physical locations on the probe. There is a diagonal line with brighter color blocks, indicating that units with more similar visual responses across days tend to be physically close. This confirms our assumption that neurons are physically stable over time. Also notice that, on each column, there might be more than one bright block in the more distant clusters. We minimize the effect of distant units by constraining the feasible region during selection of reference units. There are also columns without bright yellow blocks; these units do not respond to the stimulus and are not included in the reference set.

The effect of drift correction on reference unit yield for all three animals.

Note that drift correction improves the recovery rate for most cases; the degree of improvement is a function of the magnitude of the drift.

P (H ∣ R) can be estimated from the reference units recovery rate 0.86, and P (R) can be estimated from the ratio of reference units, which is 0.29. P (∼ R) = 1 − P (R) = 0.73. Then

Fits of z-distance distributions from the Monte Carlo simulations. The five panels correspond to: f = 0.23, 0.5, 0.6, 0.7 and 0.96.

z-distance distribution fit comparison: a. Distribution fit with 3 parameters, where the z-distribution for true hits is estimated from the reference units. The same as figure 4a. b. Distribution fit with 4 parameters, using no reference information. c. Distribution fit of a dataset in prefrontal cortex using Neuropixels 1.0, using no reference information.35

The reference unit recovery rate for recordings spanning durations.

Each triangle represents the matching results of two datasets. Animal AL031 has 6 sets of matched units, with one outlier removed. Animal AL032 has 24 sets of matched units. Animal AL036 has 60 sets of matching. The recovery rate is lower for longer durations.

Reference unit counts and normalized EMD cost for each pair of datasets recorded by the same shank.

For animal AL036 (left), we excluded the first two datasets and all of their matching results (first two rows of each matrix on the left) based on the low reference unit counts. Following analysis on their matching EMD cost, location-only cost and waveform-only cost suggest a significant difference compared to the following days (datasets in the red rectangles). We infer that the first two datatsets were recorded from a different population than later days. The other matrices show similar information for animal AL032 for reference. To show the relative magnitude of EMD cost in related datasets versus unrelated datasets, we calculated the cost between unrelated datasets with similar unit count (AL032 shank 1 and AL036 shank 1: EMD cost = 78, location cost = 67, and waveform cost = 32). The EMD cost is between 70-80, much larger than those between related datasets (between 20-30).

The normalized EMD cost (unitless), z distance (μm), physical distance (μm), and waveform distance (unitless) and the corresponding recovery rate in pairwise matches of all to all pairs of recordings, on each shank. Each triangle represents the recovery rate in a pair of datasets. Animal AL031 has 6 sets of matching, with one outlier removed. Animal AL032 has 24 sets of matching. Animal AL036 has 60 sets of matched units. Overall, most of the datatsets with high recovery rates have per-unit EMD cost in the range 20-30. Note that the EMD cost is not predictive of recovery rate.

The ratio of number of reference units to number of KSgood units decreases for pairs of datasets with larger time intervals. However, the variability of the number of reference units is generally large for all time intervals.

We varied the weight ω in Equation 4 used to combine the physical and waveform distances in increments of 500. The vertical line indicates weight = 1500, where the overall recovery rate = 86.29%. The maximum recovery rate = 87.68% occurs at weight = 3000. We chose weight = 1500 for all subsequent analysis.

The Kilosort-good and reference unit counts for the animals AL031 and AL036, as shown for animal AL032 in Figure 5.

An example of reference chain.

a. Above: Firing rates of this neuron on each day. Below: Firing rate fractional change compared to the previous day. b. Visual response similarity (yellow line), PSTH correlation (orange line), and visual fingerprint correlation (blue line). The similarity score is the sum of vfp and PSTH. The dashed black line shows the threshold to be considered a reference unit. c. Spatial-temporal waveform of a trackable unit. Each pair of traces represent the waveform on a single channel. d. Estimated location of this unit on different days. Each colored dot represents a unit on one day. The orange squares represent the electrodes. e. The pairwise vfp and PSTH traces of this unit.

An example of putative chain.

Order is the same as above.