Next generation CANCAN focusing for remote stimulation by nanosecond electric pulses (DANDI:000875)

Claims

From the paper Next generation CANCAN focusing for remote stimulation by nanosecond electric pulses

Reproducing Pakhomov et al. (2023) from dandiset 000875 4

Notebook reproduce_pakhomov_2023.ipynb

Reproduced2

ReproducedResults, 3.3, paragraph 1; Figure 6A
Increasing the total number of rotated Packet 2 packets from 4 (1Px4) to 20 (5Px4) and 40 (10Px4) profoundly enhanced YP uptake throughout the electrode array.
Increasing the total number of applied packets from 4 to 20 and 40 profoundly enhanced the electroporative YP dye uptake throughout the electrode array
Results, 3.3, paragraph 1; Figure 6A
Under the primary reading (both diagonals, raw counts, the two central ROIs, the mean of the two ends, the mean of the two side minima) the means for 4, 20 and 40 packets are 544, 3240 and 9712 counts at the periphery, 674, 2631 and 5172 at the center and 479, 1148 and 1880 at the least affected area; 40 exceeds 4 at each place by a two-sided Student's t-test (3 dishes each, p = 0.016, 0.00076 and 0.00079). The order holds in all 162 combinations of the claims file's readings; all three tests pass in 72 and at least one in the other 90. Only the periphery test fails (p up to 0.127), in the 54 combinations on diagonal A alone and the 36 that take the periphery from the negative end alone, so the full result needs diagonal B or both diagonals with the periphery from both ends or the positive end, not from the negative end alone. Saturation of up to 5% of an ROI near the 10Px4 electrodes lowers the 10Px4 periphery and works against the order. With the ROI extent taken from the footprints the primary reading gives p = 0.035, 0.0008 and 0.0008.
How the test differed from the paper’s method
- The paper's ROIs (MetaMorph) are not deposited; ROIs are placed by code along the diagonals of the detected electrode footprints, over +/- 1.65 mm from the array center with 92 ROIs, the small-array values of Methods 2.4 ("92 or 224 regions of interest (ROI), placed along one or both diagonals, and covering a distance of +/- 1.65 mm or +/- 3.65 mm"). The array is taken as the small one from the images, as the claims file asks.
- The files give no pixel size. 0.65 um per pixel (the ORCA-Flash4 camera's 6.5 um pixel through the 10x objective, both named in the file's device description and Methods 2.4) is assumed. With it the small array's ROI extent fits between the footprints and the large array's does not, and the small array measures 3.73 to 4.05 mm between needle edges, against 3.80 mm from the paper's large array (8.4 mm diagonal, Section 3.5) and its ROI extents. The scale of about 0.53 um that would make the footprints 1.5 mm across leaves 2.77 to 3.03 mm between them, too short for the paper's 3.30 mm of ROIs. A variant that ends the ROIs 5% of the diagonal short of the footprints, using no pixel size, gives the same outcome.
- 3 dishes per treatment, where the Fig. 6 legend gives n = 3 to 4.
- The ROI layout, background, center, periphery and least-affected-area readings of the claims file header are all run; with an even number of ROIs two are equally near the center, and their mean is taken for reading c1, a standard choice. The files hold no dark frame, so the background reading b2 ("minus a cell-free or dark offset, where the data hold one") is run with two estimates, the 1st percentile of each image's nonzero 8 x 8 block means (73.5 to 87.1 counts, from blocks more than twice the center-to-footprint distance from the array center; the 1st percentile of the blocks away from the zero padding is within 1.4 counts of it) and the darkest pixel near the array (110 to 200 counts), both standard choices.
ReproducedResults, 3.3, paragraph 1; Figure 6
The central peak appeared narrower (sharper) with increasing number of packets.
Also, the central peak appeared narrower (sharper) with increasing the number of packets
Results, 3.3, paragraph 1; Figure 6
On the primary reading (both diagonals, raw counts, each dish normalized to the mean of its two ends) the width at half height above the least affected area (w1) is 1.051, 0.545 and 0.488 mm for 4, 20 and 40 packets, and with LOESS spans 0.25 and 0.5 it is 0.800, 0.654, 0.560 and 0.992, 0.863, 0.817 mm. Unsmoothed, w1 depends on the center reading (0.591, 0.511 and 0.437 mm under c2, still in order). w1 decreases in 182 of 216 combinations, 40 is narrower than 4 with 20 out of order in 28, and 6, all unsmoothed with c2, are out of order. The distance between the side minima (w2), counted once as it does not depend on the center reading, decreases in 71 of 108, is partial in 28 and out of order in 9, all unsmoothed; on the primary reading it is 2.367, 1.650 and 1.435 mm with LOESS 0.5 and 1.363, 1.722 and 1.148 mm unsmoothed. Per dish, w2 on LOESS 0.5 is 1.722 to 2.367 mm for 1Px4, 1.435 to 1.650 for 5Px4 and 1.363 to 1.578 for 10Px4, so 5Px4 and 10Px4 overlap; the criterion asks for no test. With the ROI extent from the footprints w1 keeps the order under all three smoothings.
How the test differed from the paper’s method
- ROIs placed by code, pixel size assumed, dark offsets estimated and 3 dishes per treatment, as for fig6a-more-packets-more-uptake. Widths in mm rest on the assumed pixel size; their order does not.
- LOESS is fitted with statsmodels lowess without robustness iterations, a standard choice; the paper names LOESS (Methods 2.4) without a span or implementation, and the claims file names spans 0.25 and 0.5.

Partially reproduced2

Partially reproducedResults, 3.3, paragraph 1; Figure 6Brests on an assumption
The central peak was almost twofold brighter than the periphery for 1Px4, and about the same as the periphery for 5Px4 and 10Px4.
Its brightness was almost twofold higher than at the periphery for a single packet was applied 4 times from different directions, or about the same as at the periphery for the rotation of 5 and 10 packets
Results, 3.3, paragraph 1; Figure 6B
Measured values are the primary reading (both diagonals, raw counts, the two central ROIs, mean of the two ends, each dish normalized); per dish 107.2 to 151.6% for 1Px4, 73.2 to 106.2% for 5Px4 and 34.5 to 77.4% for 10Px4. No combination of the 108 readings puts all three values within tolerance. 1Px4 is within 180 to 220% and above both others in 13, which meets the claims file's rule for a partial reproduction; all 13 subtract a dark offset (5 with b2-dark, 8 with b2-min, none with raw counts), use one diagonal and take the peak as the highest ROI between the minima (c2). They are fragile. The center is one 36 um ROI, and across these rows 1Px4 per dish is 170.2 to 380.1% for dish 23, 75.2 to 145.0% for dish 24 and 126.1 to 279.5% for dish 25. On diagonal A with c2, e1 and n2, 1Px4 passes 180% once a common offset of 50 counts is subtracted, less than either dark estimate; on both diagonals with c1 it stays below 155% for offsets up to 300 counts. 1Px4 is above both others in 96 of 108; 10Px4 is within 90 to 110% in 14 and below its periphery under the primary reading. With the ROI extent from the footprints 1 of 108 combinations is partial and none reproduced (primary 116.1, 105.1 and 68.0%). The ranking holds; "almost twofold" for 1Px4 holds only on one diagonal with a dark offset subtracted; the 40-packet center is mostly well below the periphery, not about the same.
How the test differed from the paper’s method
- ROIs placed by code, pixel size assumed and 3 dishes per treatment, as for fig6a-more-packets-more-uptake.
- The paper does not fix the layout, background, center, periphery or normalization order; the claims file names readings for each and a claim holds under any combination. The partial outcome rests on reading b2, a dark offset subtracted, which the Methods do not name. The files hold no dark frame, so the offset is estimated (assumed) as the 1st percentile of each image's nonzero block means, which lie far outside the array, or the darkest pixel near the array. A sweep of a common offset over 1Px4 alone puts it above 180% from 50 counts on diagonal A with c2, e1 and n2 (within 220% up to 175 counts, 222.3% at 200); it does not check the order against 5Px4 and 10Px4. The first round used the median inside the electrode footprints; full-resolution crops show a faint monolayer inside them (median 0.74 to 0.94 of the monolayer outside the array, 3.1 to 5.0 times the dark estimate), so it is not cell-free and is shown only as a variant outside the outcome (7 of 36 partial).
Partially reproducedResults, 3.3, paragraph 1; Figure 6A and 6B
For all packet numbers the central peak was clearly discernible, and the least affected area of the monolayer, between the center and the electrodes, showed about 2-fold lower intensity than the center.
The peak of electroporation in the center was clearly discernable for all tested treatments ... Interestingly, the least affected area of the cell monolayer (between the center and the electrodes) showed about 2-fold lower fluorescence intensity than in the center for all the treatments
Results, 3.3, paragraph 1; Figure 6A and 6B
Center over least affected area under the primary reading is 1.40 for 1Px4, 2.16 for 5Px4 and 2.80 for 10Px4 (no single measured value is given, as the expected value is one number for all three treatments). Over 216 combinations it is 1.32 to 4.04 for 1Px4 (within 1.8 to 2.2 in 52), 1.86 to 4.37 for 5Px4 (in 29) and 2.66 to 10.11 for 10Px4 (in none). No combination puts all three within tolerance; 79 put one or two within it with all above 1, 27 of them with raw counts. The central peak is discernible for all treatments (ratio above 1 in all 216). The ratio grows with packet number, so the 2-fold for all treatments does not hold at 40 packets (per dish 2.52 to 2.92 with raw counts). The ROI extent from the footprints gives 69 partial and none reproduced.
How the test differed from the paper’s method
- ROIs placed by code, pixel size assumed, dark offsets estimated and 3 dishes per treatment, as for fig6a-more-packets-more-uptake; all readings of the claims file are run.

Not tested in a notebook 13

13 data missing from the dataset

Blocked13

BlockedResults, 3.1, paragraph 1; Figure 4· Data missing from the dataset
A single 600-ns unipolar pulse at 15 kV/cm caused by far the strongest YP uptake in single CHO cells (126 +/- 14 a.u. at the end of recording, n = 25), while bipolar signals starting with the same pulse caused significantly weaker uptake (4-phase 20.8 +/- 2.0, 8-phase 6.9 +/- 1.6, 8-phase with 5-us delays 25.4 +/- 2.8 a.u.; p < 0.001); sham caused no gain.
Sham-exposed cells showed no gain in YP fluorescence during the 10 min of observation ... A single 600-ns pulse at 15 kV/cm caused by far the strongest YP uptake, reaching 126±14 a.u. (n=25) by the end of recording. Bipolar signals started with exactly the same pulse (600 ns, 15 kV/cm) but caused significantly weaker effects (p<0.001). By the end of recording, YP signal reached only 20.8±2.0 a.u. for the 4-phasic signal; 6.9±1.6 a.u. for the 8-phasic signal; and 25.4±2.8 a.u. for the 8-phasic signal with 5-μs delays between phases
Results, 3.1, paragraph 1; Figure 4
What the paper reports
- each bipolar signal against the single unipolar pulse, end of recording: not named for single cells (Methods 2.2 names none; Methods 2.4 names a two-sided Student's t-test for the monolayer data), p < 0.001 (n = 21 to 30 cells per group (25 for the unipolar pulse))
The single-cell confocal time-lapse of Fig. 4 (Methods 2.2) is not in the dandiset, which holds nine stitched monolayer images only. The paper names no other repository for it.
BlockedResults, 3.1, paragraph 1; Results, 3.4, paragraph 1; Figure 4 legend· Data missing from the dataset
The 8-phase bipolar signal caused about 20-fold weaker electroporation than the single unipolar pulse, and was the most efficient at cancellation; the 4-phase and delayed 8-phase signals cancelled 4 to 5 fold.
Note that applying the 8-phasic bipolar signal with no inter-phase delays caused about 20-fold weaker electroporation than a single unipolar pulse
Results, 3.1, paragraph 1; Results, 3.4, paragraph 1; Figure 4 legend
Depends on fig4-bipolar-cancellation, whose single-cell data are not in the dandiset.
BlockedResults, 3.2, paragraph 1; Figure 5A· Data missing from the dataset
Packet 2 applied 40 times from e1 caused much weaker electroporation throughout the large array than 40 unipolar trains from e1, and enhanced electroporation in the center, but electroporation near e1 stayed stronger than in the center.
The lack of cancellation caused a much stronger electroporation throughout the array ... In contrast, Packet 2 enhanced electroporation in the center of the array, despite gradual weakening of the electric field with distance from e1. However, the bipolar cancellation only partially offset the electric field weakening, and electroporation near e1 remained stronger than in the center
Results, 3.2, paragraph 1; Figure 5A
The Fig. 5 dishes (40 packets or 40 trains from e1, 6.4 kV) are not in the dandiset; all nine files are rotated Packet 2 at 3.2 kV.
BlockedResults, 3.2, paragraph 1; Figure 5B and 5C· Data missing from the dataset
Rotating the first-energized electrode with Packet 2 (1Px40) emphasized the remote effect in the center of the array, while the same rotation with the pulse train (1Tx40) left a void of electroporation in the center.
Rotation with Packet 2 “smeared” the electroporation effect at the perimeter and emphasized the remote effect in the center. This remote effect contrasted with a void of electroporation in the center when the pulse train was applied in the same manner instead of the packet
Results, 3.2, paragraph 1; Figure 5B and 5C
The Fig. 5 dishes (1Px40 and 1Tx40, 6.4 kV) are not in the dandiset.
BlockedResults, 3.3, paragraph 2; Figure 7A· Data missing from the dataset
With 1Px4 in the small array, raising the first-pulse amplitude from 2.4 to 5.6 kV raised YP fluorescence 15.5 times at the most peripheral point, 16.2 times half-way to the center and 16.8 times in the center.
when the first pulse amplitude was increased from 2.4 to 5.6 kV ... YP fluorescence increased 15.5 times at the most peripheral point; 16.2 times half-way to the center; and 16.8 times in the center
Results, 3.3, paragraph 2; Figure 7A
Needs 1Px4 profiles at 2.4 and 5.6 kV in the small array (Fig. 7A). The dandiset holds 1Px4 only at 3.2 kV.
BlockedResults, 3.3, paragraph 2; Figure 7B and 7D· Data missing from the dataset
Normalized to the most peripheral ROI, the profiles for different nsEP amplitudes fully overlapped, in the small array with 1Px4 and in the large array with 10Px4, so NG-CANCAN targeting did not depend on amplitude within the range studied.
When the YP emission was normalized to the most peripheral ROI, the data for different nsEP amplitudes fully overlapped ... The findings held true for rotating multiple packets and when using the larger array ( Fig. 7C and D ). Overall, our results show that NG-CANCAN targeting of electroporation does not depend on nsEP amplitude
Results, 3.3, paragraph 2; Figure 7B and 7D
Needs several first-pulse voltages in each array (Fig. 7B and 7D). The dandiset holds one voltage (3.2 kV) and the small array only.
BlockedDiscussion, paragraph 4; Figure 7· Data missing from the dataset
Remote targeting, at least within the limits studied, did not depend on array size; the authors infer that findings from small arrays can be extrapolated to larger inter-electrode distances.
The electric field distribution and strength achieved within each array at a given pulse voltage were obviously different, but the remote targeting, at least within studied limits, did not depend on either the electric pulse amplitude or array size ( Fig. 7 )
Discussion, paragraph 4; Figure 7
Needs large-array dishes (Fig. 7C and 7D), which are not in the dandiset; it also depends on fig7-targeting-independent-of-amplitude, blocked for the same reason.
BlockedResults, 3.4, paragraph 2; Figure 8A· Data missing from the dataset
At 4 kV and one rotation, Packets 1, 2 and 3 produced the same YP uptake in the center (215 to 235 a.u.), but Packets 1 and 3 caused much more uptake near the electrodes (250 to 350 a.u.) than Packet 2 (about 120 a.u., p < 0.01).
All packets produced the same YP uptake in the center (215–235 a.u.), but Packets 1 and 3 caused drastically more YP uptake near the electrodes (250–350 a.u. vs ~120 a.u. for packet 1, p<0.01, Fig. 8A )
Results, 3.4, paragraph 2; Figure 8A
What the paper reports
- Packets 1 and 3 against Packet 2, near the electrodes: two-sided Student's t-test, p < 0.01 (n = 3 to 4 dishes per packet)
Needs Packets 1, 2 and 3 at 4 kV (Fig. 8); the dandiset holds Packet 2 at 3.2 kV only.
BlockedResults, 3.4, paragraph 2; Figure 8B· Data missing from the dataset
Normalized YP emission in the center was 183% for Packet 2, against 83% for Packet 3 and 74% for Packet 1.
The superiority of Packet 2 for CANCAN targeting is highlighted by the drop of normalized YP emission in the center from 183% for Packet 2 to 83% and 74% for Packets 3 and 1, respectively
Results, 3.4, paragraph 2; Figure 8B
Needs Packets 1, 2 and 3 at 4 kV (Fig. 8B); not in the dandiset.
BlockedDiscussion, paragraph 3; Figures 4 and 8· Data missing from the dataset
The bipolar signal with the strongest cancellation between two electrodes (8-phase, Fig. 4) gave the most efficient remote targeting in the quadrupole (Packet 2, Fig. 8).
Three packets for CANCAN targeting were designed to produce three types of bipolar signals near electrodes and we showed that, indeed, the strongest bipolar cancellation has led to the most efficient remote targeting ( Figs. 4 and 8 )
Discussion, paragraph 3; Figures 4 and 8
Depends on fig4-twenty-fold-cancellation and fig8b-normalized-center, both blocked for missing data.
BlockedResults, 3.5, paragraph 1; Figure 9A· Data missing from the dataset
With the large array lifted 1, 2 and 3 mm above the monolayer, YP emission near the electrodes fell 2.4, 6.4 and 9 times, and in the center only 1.02, 1.4 and 3.8 times, relative to electrodes touching the monolayer.
Specifically, YP emission near the electrodes decreased 2.4, 6.4, and 9 times with 1-, 2- and 3-mm gaps (as compared to electrodes being in the same plane with cells); in the center, it decreased just 1.02, 1.4, and 3.8 times, respectively
Results, 3.5, paragraph 1; Figure 9A
Needs large-array dishes with the electrodes 0 to 3 mm above the monolayer (Fig. 9); not in the dandiset.
BlockedResults, 3.5, paragraph 1; Figure 9B and 9C· Data missing from the dataset
The faster fading near the electrodes raised targeting efficiency (center over periphery) to as much as 600% with a 2-mm gap, and electroporation was restricted almost entirely to the center with the electrodes above the monolayer.
Such disproportionally faster fading of the effect near electrodes boosted the CANCAN targeting efficiency (measured as a ratio of the effect in the center and at the periphery) to as much as 600% for the 2-mm gap ( Fig. 9B )
Results, 3.5, paragraph 1; Figure 9B and 9C
Depends on fig9-gap-fades-periphery; the Fig. 9 dishes are not in the dandiset.
BlockedMethods, 2.3, paragraph 3; Figure 3· Data missing from the datasetuses data from outside the dataset
In the Sim4Life simulation, 1 kV on one electrode gives about 1.8 kV/cm (small array) and 0.8 kV/cm (large array) in the center and about 4-fold more near the energized electrode (7 and 3.5 kV/cm), while energizing a diagonal gives no field in the center.
applying 1 kV to one electrode produced the electric field of approximately 1.8 kV/cm and 0.8 kV/cm in center of the small and large arrays ... The electric field strength was about 4-fold higher near the energized electrode (7 kV/cm and 3.5 kV/cm per 1 kV applied, for the small and large arrays, respectively) Note that energizing either electrode diagonal produced no electric field in the center of the array
Methods, 2.3, paragraph 3; Figure 3
Needs the Sim4Life model of the two arrays. The paper deposits none and says the simulation follows reference 15 (doi:10.1016/j.bioelechem.2021.107876), which names no public model either; the dandiset holds images only. No public model was found to pin.

Findings about the data

What attempts found about the dataset itself, for its owner and the archive.

Data missingThe dandiset holds only the experiment of Fig. 6 (nine dishes, 1Px4, 5Px4 and 10Px4 at 3.2 kV, three each, against n = 3 to 4 in the legend). The single-cell time-lapse of Fig. 4 and the monolayer experiments of Figs. 5, 7, 8 and 9 are not in it, so 13 of the paper's 17 claims cannot be tested.
Description missingThe files give no pixel size or field of view, no electrode positions, no ROIs or measured profiles, no dark frame, and do not say which of the two arrays was used. At the camera scale (0.65 um per pixel) the geometry fits the small array and the paper's ROI extents; the dark electrode footprints are then 1.66 to 2.09 mm across, wider than the 1.5 mm needles of Methods 2.3, and they hold a faint monolayer, so they give no cell-free background.
Metadata errorThe subject species is NCBITaxon 9913 (Bos taurus, cattle) in every file and in the archive's summary, while the strain field and the paper say CHO cells (Chinese hamster). The age is "P1D", which has no meaning for a cell line, and the archive's summary counts 18 subjects for 9 files and 9 subject ids.
Metadata errorTwo pairs of dishes share a session start time to the second (dishes 20 and 22 of 5Px4; dish 09 of 5Px4 and dish 25 of 1Px4). The images differ (no two identical, array position different in each), so the start times are not per-dish acquisition times.
Description missingThe ImageSeries has unit "n.a." and a rate of 2 Hz for a single frame, and the "200ms" in every file and session name is not explained; the description says packets were given at 1 Hz.
Methods not availableQuestion for the paper's authors. Methods 2.4 says the ROIs were "placed along one or both diagonals" but not which for Fig. 6, nor the ROI size, whether a background was subtracted, which ROI is the center, or whether dishes were normalized before averaging. The Fig. 6B normalized center depends on these (1Px4 is 121.7 to 301.3% over the readings, near 200% only on one diagonal with a dark offset subtracted). Which layout, ROI size and background were used for Fig. 6?

Notebooks

Every registered entry that uses this dataset. An entry can hold several notebooks, and can use more than one dataset.

Reproduction NG-CANCAN focusing of electroporation by unipolar pulse packets (Pakhomov et al. 2023)

Tests the findings of Pakhomov, Gudvangen, Mangalanathan, Kondratiev, Redondo and Semenov (2023) on dandiset 000875, which holds nine stitched YO-PRO-1 images of CHO monolayers after 4, 20 or 40 rotated Packet 2 packets in a quadrupole electrode array (Fig. 6 of the paper). It measures dye uptake along the array diagonals and compares the profiles between packet numbers. The single-cell data of Fig. 4 and the monolayer data of Figs. 5, 7, 8 and 9 are not in the dandiset. All files

Reproducing Pakhomov et al. (2023) from dandiset 000875
reproduce_pakhomov_2023.ipynbExecuted, with outputs · Source

Verification

Round 1
2026-10-31Agreed with 17 of 17 outcomes. Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook). Reviewed: The handover of the first run, the executed notebook with the conclusions and summary removed, the helper module and the claims file, with the paper. The verifier checked the archive's assets and metadata itself for the blocked claims.
Round 2
2026-10-31Agreed with 17 of 17 outcomes. Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook). Reviewed: The full executed notebook of the second run with its conclusions and summary, the helper module, the README, this attempt record and the claims file, with the paper (round 2).

Method differences

How the tests differed from the paper’s method or data, and the claims each applies to.

The paper's ROIs (MetaMorph) are not deposited; ROIs are placed by code along the diagonals of the detected electrode footprints, over +/- 1.65 mm from the array center with 92 ROIs, the small-array values of Methods 2.4 ("92 or 224 regions of interest (ROI), placed along one or both diagonals, and covering a distance of +/- 1.65 mm or +/- 3.65 mm"). The array is taken as the small one from the images, as the claims file asks.fig6a-more-packets-more-uptake
The files give no pixel size. 0.65 um per pixel (the ORCA-Flash4 camera's 6.5 um pixel through the 10x objective, both named in the file's device description and Methods 2.4) is assumed. With it the small array's ROI extent fits between the footprints and the large array's does not, and the small array measures 3.73 to 4.05 mm between needle edges, against 3.80 mm from the paper's large array (8.4 mm diagonal, Section 3.5) and its ROI extents. The scale of about 0.53 um that would make the footprints 1.5 mm across leaves 2.77 to 3.03 mm between them, too short for the paper's 3.30 mm of ROIs. A variant that ends the ROIs 5% of the diagonal short of the footprints, using no pixel size, gives the same outcome.fig6a-more-packets-more-uptake
3 dishes per treatment, where the Fig. 6 legend gives n = 3 to 4.fig6a-more-packets-more-uptake
The ROI layout, background, center, periphery and least-affected-area readings of the claims file header are all run; with an even number of ROIs two are equally near the center, and their mean is taken for reading c1, a standard choice. The files hold no dark frame, so the background reading b2 ("minus a cell-free or dark offset, where the data hold one") is run with two estimates, the 1st percentile of each image's nonzero 8 x 8 block means (73.5 to 87.1 counts, from blocks more than twice the center-to-footprint distance from the array center; the 1st percentile of the blocks away from the zero padding is within 1.4 counts of it) and the darkest pixel near the array (110 to 200 counts), both standard choices.fig6a-more-packets-more-uptake
ROIs placed by code, pixel size assumed and 3 dishes per treatment, as for fig6a-more-packets-more-uptake.fig6b-peak-relative-to-periphery
The paper does not fix the layout, background, center, periphery or normalization order; the claims file names readings for each and a claim holds under any combination. The partial outcome rests on reading b2, a dark offset subtracted, which the Methods do not name. The files hold no dark frame, so the offset is estimated (assumed) as the 1st percentile of each image's nonzero block means, which lie far outside the array, or the darkest pixel near the array. A sweep of a common offset over 1Px4 alone puts it above 180% from 50 counts on diagonal A with c2, e1 and n2 (within 220% up to 175 counts, 222.3% at 200); it does not check the order against 5Px4 and 10Px4. The first round used the median inside the electrode footprints; full-resolution crops show a faint monolayer inside them (median 0.74 to 0.94 of the monolayer outside the array, 3.1 to 5.0 times the dark estimate), so it is not cell-free and is shown only as a variant outside the outcome (7 of 36 partial).fig6b-peak-relative-to-periphery
ROIs placed by code, pixel size assumed, dark offsets estimated and 3 dishes per treatment, as for fig6a-more-packets-more-uptake; all readings of the claims file are run.fig6-trough-half-of-center
ROIs placed by code, pixel size assumed, dark offsets estimated and 3 dishes per treatment, as for fig6a-more-packets-more-uptake. Widths in mm rest on the assumed pixel size; their order does not.fig6-peak-narrower
LOESS is fitted with statsmodels lowess without robustness iterations, a standard choice; the paper names LOESS (Methods 2.4) without a span or implementation, and the claims file names spans 0.25 and 0.5.fig6-peak-narrower

Provenance

Attempt record: 000875-pakhomov-2023-01
Paper: Next generation CANCAN focusing for remote stimulation by nanosecond electric pulses
Date: 2026-10-31
Made by: An agent, run by CatalystNeuro
Dataset: DANDI:000875
Version: 0.240515.1925 (doi:10.48324/dandi.000875/0.240515.1925)
Run on: GitHub Actions
Harness: Claude Code
Model: claude-opus-5-5
Protocol: pilot-12

Transcripts of the agent runs

Each run’s prompt, messages and tool calls, with long output cut and the text of the paper left out, and its final report.

Other datasets that name the same paper

← DANDI:000774 DANDI:000897 →