{ "dataset": { "archive": "dandi", "id": "000351", "label": "DANDI:000351" }, "page": "datasets/dandi/000351/", "generated": "2026-11-05T07:41:23+00:00", "caution": "A failure to reproduce is not a finding that the paper is wrong. It may be a failure of the attempt.", "outcome_counts": { "reproduced": 27, "partially_reproduced": 0, "not_reproduced": 2, "blocked": 8 }, "outcome_counts_with_outside_data": { "reproduced": 0, "partially_reproduced": 0, "not_reproduced": 0, "blocked": 0 }, "outcome_counts_elsewhere": { "reproduced": 0, "partially_reproduced": 0, "not_reproduced": 0, "blocked": 0 }, "outcome_counts_links": { "reproduced": 0, "partially_reproduced": 1, "not_reproduced": 0, "blocked": 0 }, "attempts": [ { "id": "000351-jeong-2022-01", "dataset": { "archive": "dandi", "id": "000351", "version": "0.260828.1905", "doi": "10.48324/dandi.000351/0.260828.1905" }, "key": "dandi:000351", "label": "DANDI:000351", "dataset_version": "0.260828.1905", "paper": "10.1126/science.abq6740", "paper_title": "Mesolimbic dopamine release conveys causal associations", "date": "2026-10-30", "kind": "agent", "accountable": "CatalystNeuro", "agent": { "harness": "Claude Code", "model": "claude-opus-5-5", "protocol": "pilot-12" }, "runner": { "platform": "GitHub Actions", "url": "https://github.com/catalystneuro/encore/actions/runs/40984158261" }, "tier": "agent_verified", "deprecated": "", "deprecated_on": null, "entry": "000351-jeong-2022", "entry_title": "Mesolimbic dopamine release conveys causal associations (Jeong et al. 2022)", "description": "Tests the findings of Jeong, Taylor, Floeder et al. (2022) on dandiset 000351, which holds fiber photometry of dopamine release in the nucleus accumbens and the event logs of the random reward, Pavlovian, trial-less and sequential conditioning experiments, and the licking of a separate group given random rewards at three intervals. It repeats the paper's tests of how the reward and cue responses change with experience, the comparisons of dopaminergic and behavioural learning, the intermediate cue test and the optogenetic inhibition experiment, with the session lists of the paper's code. The model simulations are not tested.", "notebooks": [ { "path": "reproduce_jeong_2022.ipynb", "title": "Reproducing Jeong et al. (2022) from dandiset 000351", "executed": "notebooks/000351-jeong-2022/reproduce_jeong_2022.html" } ], "outcomes": [ { "claim": "session-day-link", "statement": "The subjects and \"Day\" numbers that the paper's analysis code names for each experiment are sessions of the dandiset, so that the code's selection of sessions can be applied to it.", "source": "Data and materials availability; README.md and functions/load.py of github.com/namboodirilab/ANCCR at commit cff4414; dandiset description (related resource \"Subject and session information\")", "quote": "", "statistics": [], "kind": "link", "inferred": "", "outcome": "partially_reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "All 28 subjects of the code's lists are in the dandiset, matched with _ read as -. Of the 119 listed days (startday, endday, learnedday and the single days of omission.m), 117 are sessions with the condition the script names; HJ-FP-M4 Day38, the first background reward day in cumsum_pavlovian.m (condition 4) and cumsum_bgdrw_vs_extinction.m, is not in the dandiset. All 320 sessions inside the ranges have the condition (3 s or 9 s cue-reward delay, background rewards, no delivered reward, trial-less task). Counted literally, 320 of the 331 integer days in the ranges are sessions; the 7 missing days inside the Pavlovian ranges are M2 Day23, M3 Day16 and Day31, M4 Day30 and Day38, F2 Day16 and M6 Day49. The subjects match and more than 90% of the listed days do, so the link holds in part. The dependent claims are tested on the sessions present; the missing sessions bear on fig4p-background-faster-decay (M4's first background session) and on the session counts. The spreadsheet named in the dandiset's description was not used, as it has no version to pin. The notebook reads the six scripts at the pinned commit and finds that every mouse and day list in jeong.py equals the script's, and that cue_acquisition.m and backpropagation_pavlovian.m use the initial learning days of cumsum_pavlovian.m.", "deviations": [ "Day numbers are read from the file names as load_dandi_url does; the Matlab scripts read them from folder names of the authors' local copies, which the dandiset does not hold.", "The criterion's every Day number from startday to endday is reported both for the listed days and literally for every integer day; the outcome uses the listed days, since integer days without a session can be days without training." ], "assumed": false, "disagreed": false, "data": "with_labels", "where": null, "proposed": null }, { "claim": "methods-session-counts", "statement": "Animals were trained for 7 +/- 2.1 sessions of random rewards, 13.3 +/- 2.7 sessions of initial conditioning, 4.3 +/- 0.8 sessions with the 8 s cue, 8.1 +/- 2.5 sessions with background rewards and 2.6 +/- 0.8 sessions of extinction; in sequential conditioning, 5.6 +/- 2.4 (wild type) and 7.1 +/- 0.7 (DAT-Cre) sessions (mean +/- SD).", "source": "Methods, Behavioral task, paragraphs 1 to 5", "quote": "Animals were trained on Experiment 1 for 4-11 sessions (7 \u00b1 2.1 sessions, mean \u00b1 std) ... (13.3 \u00b1 2.7 sessions in total) ... (4.3 \u00b1 0.8 sessions) ... (8.1 \u00b1 2.5 sessions) ... (2.6 \u00b1 0.8 sessions) ... (5.6 \u00b1 2.4 and 7.1 \u00b1 0.7 sessions for WT and DAT-Cre, respectively)", "statistics": [], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Counted from the dandiset, the means per animal are 7.0 random reward sessions (4 to 11, SD 2.14), 12.71 of Experiment 2, 4.29 of Experiment 3, 7.86 of Experiment 5, 2.43 of Experiment 4, and 5.43 (WT) and 7.0 (DAT-Cre) sequential conditioning sessions, all within the stated SDs. The counts of integer days in the code's ranges are 13.29, 4.29, 8.14 and 2.57, which equal the paper's means for Experiments 2, 5 and 4; this suggests that the paper's counts include sessions on the 7 integer days in the ranges that have no session in the dandiset. The SDs of the Experiment 7 counts, 0.79 (wild type) and 2.45 (DAT-Cre, from datHT-M7's 12 sessions), are close to the paper's 2.4 and 0.7 in the other order, so the paper may have swapped them; the criterion tests only the means.", "deviations": [ "Experiment 5 is counted as the sessions with background rewards in the event log, which equals the count in the range of cumsum_bgdrw_vs_extinction.m (7.86); condition 4 of cumsum_pavlovian.m gives 7.14, also within tolerance." ], "assumed": false, "disagreed": false, "data": "with_labels", "where": null, "proposed": null }, { "claim": "exp1-excluded-rewards", "statement": "In Experiment 1, 22.6 +/- 2.4% of rewards came less than 3 s after the previous reward and 5.7 +/- 0.1% had no lick before the next reward; both were excluded.", "source": "Methods, Data analysis, Experiment 1, paragraph 2", "quote": "we excluded rewards with less than 3 s IRI from the previous reward (22.6\u00b12.4%) for the above analyses. Rewards without lick until the next reward (5.7\u00b10.1%) were also excluded from analyses.", "statistics": [], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Reading (a), mean over the 8 animals, 22.59% of rewards with IRI under 3 s and 5.66% without a lick before the next reward; reading (b), over 56 sessions, 22.59% and 5.68%. Both within tolerance under both readings. The SD over animals is 2.39 and 1.04, and over sessions 7.04 and 4.27; the paper's 0.1 matches neither.", "deviations": [ "The paper does not say how the percentages are averaged; both readings of the criterion are shown and both match." ], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig3e-reward-response-increases", "statement": "In experimentally naive mice given unpredicted sucrose, the dopamine response to reward increases with the number of rewards experienced, in every animal (Test 1).", "source": "Figure 3D and 3E; Results, Tests 1 and 2, paragraph 3", "quote": "Every animal showed an increasing sucrose response that reached a high positive asymptote.", "statistics": [ { "test": "one sample t-test against zero, of per-animal correlations", "statistic": "t(7) = 4.40", "p": 0.0031, "n": "8 animals" } ], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Per animal r between response and reward number, 0.068 to 0.677 (mean 0.308), positive in 8 of 8; t(7) = 4.41, p = 0.0031 (paper t = 4.40, p = 0.0031). Of 5600 rewards, 4064 are kept by the exclusions, 4013 of them with dF/F.", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig3g-response-iri-positive", "statement": "The dopamine response to an unpredicted reward is positively correlated with the interval since the previous reward (Test 2).", "source": "Figure 3G; Results, Tests 1 and 2, paragraph 5", "quote": "The experimentally observed correlation between dopamine response to sucrose and the previous IRI was positive, thereby being consistent with ANCCR but not RPE.", "statistics": [ { "test": "one sample t-test against zero, of per-animal correlations", "statistic": "t(7) = 5.95", "p": 0.00057, "n": "8 animals" } ], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Per animal mean of per-session r between response and previous IRI, 0.029 to 0.405; t(7) = 5.95, p = 5.7e-4 (paper t = 5.95, p = 5.7e-4).", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "figs8k-iri-correlation-stable", "statement": "The positive correlation between reward response and previous IRI holds across sessions and does not change over the first 7 sessions.", "source": "Figure S8K; Results, Tests 1 and 2, paragraph 5; Table S1", "quote": "The correlation between dopamine response to a reward and the previous IRI remains positive across sessions.", "statistics": [ { "about": "change of the per-session correlation over the first 7 sessions", "test": "linear regression", "statistic": "t = -0.04", "p": 0.9707, "n": "50 sessions from 8 animals" } ], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Linear regression of per-session r on session number over 50 sessions, slope -0.0005, t = -0.04, p = 0.969 (paper t = -0.04, p = 0.9707, n = 50). Mean r over animals is positive in each of sessions 1 to 7 (0.134 to 0.288).", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "figs8b-lick-rates-constant", "statement": "Total, consummatory and non-consummatory lick rates do not change across the first 7 random reward sessions.", "source": "Figure S8B; Supplementary Note 3, point 1; Table S1", "quote": "The lick rates generally remained constant across repeated sessions.", "statistics": [ { "about": "total lick rate across sessions", "test": "linear regression", "statistic": "t = -0.79", "p": 0.4389, "n": "50 sessions from 8 animals" }, { "about": "consummatory lick rate across sessions", "test": "linear regression", "statistic": "t = -0.20", "p": 0.8397, "n": "50 sessions from 8 animals" }, { "about": "non-consummatory lick rate across sessions", "test": "linear regression", "statistic": "t = -0.58", "p": 0.5618, "n": "50 sessions from 8 animals" } ], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Over 50 sessions, total rate slope -0.038 licks/s per session (t = -0.78, p = 0.4389), consummatory -0.008 (t = -0.20, p = 0.8397), non-consummatory -0.027 (t = -0.58, p = 0.5618); the paper's p values.", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "figs8d-consummatory-lick-uncorrelated", "statement": "The consummatory lick rate after a reward is not correlated with the dopamine response to that reward.", "source": "Figure S8C and S8D; Table S1", "quote": "The consummatory lick rate is not correlated with dopamine reward response.", "statistics": [ { "test": "one sample t-test of per-animal correlations", "statistic": "t = 0.23", "p": 0.8245, "n": "8 animals" } ], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Per animal r from -0.20 to 0.15 (mean 0.010) over 281 to 686 rewards; t(7) = 0.23, p = 0.824 (paper t = 0.23, p = 0.8245). 0 to 14 rewards per animal with an infinite or missing lick rate are dropped, as the code drops infinite values.", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "figs8e-lick-bout-responses", "statement": "Dopamine rises at the onset of consummatory lick bouts but falls slightly at the onset of non-consummatory (dry) lick bouts, so the rising reward response is not a general response to licking.", "source": "Figure S8E; Table S1", "quote": "The average lick bout onset response was positive for consummatory lick bouts (during sucrose consumption) but was slightly negative for dry lick bout onsets away from reward consumption.", "statistics": [ { "about": "consummatory against non-consummatory bout onset response", "test": "paired t-test", "statistic": "t = 3.28", "p": 0.0134, "n": "8 animals" }, { "about": "consummatory bout onset response against zero", "test": "one sample t-test", "statistic": "t = 3.03", "p": 0.0191, "n": "8 animals" }, { "about": "non-consummatory bout onset response against zero", "test": "one sample t-test", "statistic": "t = -2.55", "p": 0.0379, "n": "8 animals" } ], "kind": "finding", "inferred": "", "outcome": "blocked", "blocked_by": "methods_underspecified", "notebook": "reproduce_jeong_2022.ipynb", "notes": "Under every reading the consummatory onset response is above 0 (mean z 1.76 to 3.07, t = 6.30 to 6.49, p <= 4.0e-4) and above the non-consummatory one (paired t = 7.17 to 8.58, p <= 1.8e-4). The non-consummatory response is negative on average under all readings but significant only under (c), 0 to 0.5 s (z = -0.35, t = -2.56, p = 0.0376, against the paper's t = -2.55, p = 0.0379); under (a) 0 to 1 s z = -0.15, t = -1.65, p = 0.14, and under (b) -0.5 to 1 s z = -0.02, p = 0.80. The window is open (Methods, legend and code do not give it) and the readings give different outcomes, so under the claims file's rule for readings of an open method detail the claim is blocked; the question is put to the authors under findings. The consummatory t values are about twice the paper's (3.03), so the paper's response measure differs from these readings.", "deviations": [ "The window of the tests is given by neither the Methods, the legend nor the code; the three readings of the criterion are run, and they give different outcomes for the non-consummatory test." ], "assumed": false, "disagreed": true, "data": "dataset", "where": null, "proposed": null }, { "claim": "figs8g-baseline-unchanged", "statement": "Spontaneous baseline dopamine events, away from rewards and licks, do not change in size across random reward sessions.", "source": "Figure S8F and S8G; Methods, Data analysis, Experiment 1, paragraph 3; Table S1", "quote": "There was no change in the baseline dopamine activity across sessions.", "statistics": [ { "test": "one sample t-test of per-animal correlations", "statistic": "t = 0.11", "p": 0.919, "n": "8 animals" } ], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "With peaks kept by the Methods' rule (more than 0.5 s from any lick and more than 2 s from any reward), mean r of peak size with session -0.034, t(7) = -0.13, p = 0.90; with the code's rule, mean r 0.018, t(7) = 0.07, p = 0.95 (paper t = 0.11, p = 0.919). Per animal r from -0.98 to 0.97. Sessions keep a median of 194 peaks (5 to 516) under the Methods' rule.", "deviations": [ "Peaks are found with scipy.signal.find_peaks (distance 50 samples, height and prominence as the code), in place of Matlab's findpeaks; NaN samples are set to the session minimum so that they form no peak." ], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "figs8h-baseline-uncorrelated-with-reward-change", "statement": "Across animals, the change of baseline dopamine activity over sessions is not correlated with the change of the reward response over sessions.", "source": "Figure S8H; Supplementary Note 3, point 4", "quote": "Dopamine baseline activity change across sessions is uncorrelated with dopamine reward response change across sessions.", "statistics": [], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Across the 8 animals, r between the baseline peak size correlation and the reward response correlation is -0.10 (p = 0.81) with the Methods' rule and -0.09 (p = 0.83) with the code's. The reward response correlation with session is positive in every animal (0.23 to 0.92).", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "figs8l-lick-rate-tracks-iri", "statement": "In a separate group of 6 mice, the non-consummatory lick rate depends on the session's mean IRI (6, 12 or 30 s) while the consummatory lick rate does not, so animals learn the average reward rate within at most two sessions.", "source": "Figure S8L; Table S1", "quote": "The non-consummatory exploratory licks made by the animals depended on the average IRI for a given session. Thus, animals learn the average reward rate of an environment within at most two sessions.", "statistics": [ { "about": "non-consummatory lick rate against IRI condition", "test": "linear regression", "statistic": "t = -2.73", "p": 0.0147, "n": "18 sessions from 6 animals" }, { "about": "consummatory lick rate against IRI condition", "test": "linear regression", "statistic": "t = 1.00", "p": 0.3338, "n": "18 sessions from 6 animals" } ], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "With the IRI in seconds, non-consummatory slope -0.0195 licks/s per s, t = -2.73, p = 0.0147, and consummatory t = 1.00, p = 0.334 (the paper's values, n = 18); with the rank of the condition, t = -2.61 (p = 0.019) and t = 1.34 (p = 0.199). Every animal has two sessions at each of 6, 12 and 30 s.", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig4ef-dopamine-leads-behavior", "statement": "During initial cue-reward learning, the dopamine response to CS+ develops before anticipatory licking, and behavioral learning is much more abrupt than dopaminergic learning (Test 3).", "source": "Figure 4C to 4F; Results, Tests 3-7, paragraph 1", "quote": "Dopamine cue response considerably leads behavior across animals. ... Behavioral learning is much more abrupt than dopaminergic learning (paired t test for abruptness of change; t(6) = 9.06; two-tailed p = 1.0\u00d710-4; paired t test for change trial; t(6) = -2.93; two-tailed p = 0.0263; n=7 animals)", "statistics": [ { "about": "abruptness of change, lick against dopamine", "test": "paired t-test", "statistic": "t(6) = 9.06", "p": 0.0001, "n": "7 animals" }, { "about": "change trial, lick against dopamine", "test": "paired t-test", "statistic": "t(6) = -2.93", "p": 0.0263, "n": "7 animals" } ], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Methods' lick measure (0 to 2 s), lick abruptness 0.85 against dopamine 0.26, paired t = 5.06, p = 0.0023; change trial lick 337 against dopamine 177, t = 3.31, p = 0.016. Code's lick measure (cue and trace), t = 9.09, p = 1.0e-4 and t = 2.93, p = 0.026, the paper's values with the change-trial t as lick minus dopamine (the paper prints -2.93). Trials used, round(1.5 x trials before learnedday), 191 to 900. Of them 0 to 3 per animal have a NaN smoothed dF/F sample from -1 to 2 s and keep the nansum of the others, as aucsignal.m does; leaving them out gives t = 5.15, 3.31, 9.15 and 2.93.", "deviations": [], "assumed": false, "disagreed": false, "data": "with_labels", "where": null, "proposed": null }, { "claim": "figs10c-cue-response-at-acquisition", "statement": "By the session at which behavior was acquired, the dopamine cue response had reached 93% of its value on the final session (about 100% relative to the mean of the last two sessions).", "source": "Figure S10C; Supplementary Note 5, paragraph 2", "quote": "Dopamine cue response reaches 93% of its response on the final session (~100% if normalized relative to average cue response on session n and n-1).", "statistics": [], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Reading (a), mean of per-animal ratios, cue response on the learned session over the last session 0.934 and over the mean of the last two 0.934; reading (b), ratio of means, 0.913 and 0.947. All within tolerance. Per animal the first ratio ranges from 0.59 to 1.33. The Fig 4 measure gives 0.985 and 0.955. One trial of the sessions compared has a NaN dF/F sample; without it reading (a) gives 0.938 and 0.938.", "deviations": [], "assumed": false, "disagreed": false, "data": "with_labels", "where": null, "proposed": null }, { "claim": "fig4hi-delay-change-behavior-changes", "statement": "When the cue is lengthened from 2 s to 8 s (cue-reward delay 3 s to 9 s), animals change the timing of anticipatory licking abruptly, and this change is better fit by a changing model than by a constant one.", "source": "Figure 4G to 4I; Results, Tests 3-7, paragraph 2", "quote": "Behavior is learned abruptly by all animals, but the dopaminergic cue response shows little to no change. ... Paired t test for abruptness of change; t(6) = 22.92; two-tailed p = 4.52\u00d710-7; one-sample t test for \u0394AIC against a null of zero; t(6) = 7.49 for lick", "statistics": [ { "about": "abruptness of change, lick against dopamine", "test": "paired t-test", "statistic": "t(6) = 22.92", "p": 4.52e-07, "n": "7 animals" }, { "about": "delta AIC for licking against zero", "test": "one sample t-test", "statistic": "t(6) = 7.49", "p": 0.00029, "n": "7 animals" } ], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Lick timing abruptness 0.233 against dopamine 0.064, paired t = 15.34, p = 4.9e-6 (paper t = 22.92). Changing model better than constant for lick timing in all 7 animals, mean delta AIC 382.8, t = 7.49, p = 2.9e-4 with k = 4 (the paper's t), and t = 7.53 with k = 3. 12 to 77 trials per animal without licks in cue and trace are dropped, as the code does. 0 to 2 trials per animal have a NaN smoothed dF/F sample from -1 to 2 s; leaving them out gives abruptness t = 15.40 and delta AIC t = 7.46. The abruptness t is smaller than the paper's while the delta AIC t equals it, for a reason not found here; HJ-FP-M4 lacks Day30, one of its three sessions before the change, and has 88 trials before it where the others have 143 to 150.", "deviations": [ "The changing model is fitted by bounded least squares (scipy) from 20 random start points with a fixed seed, keeping the best fit; Matlab's fit starts custom models at random points, so the code's fits are not repeated exactly.", "dF/F is divided by 100 before the dopamine AUC, as the files store percent and the Methods define dF/F as the fraction (470 - fitted 405)/fitted 405 (Methods, Fiber photometry). The scale for which the code's bounds were set is not stated; see fig4hi-delay-change-dopamine-unchanged." ], "assumed": false, "disagreed": false, "data": "with_labels", "where": null, "proposed": null }, { "claim": "fig4hi-delay-change-dopamine-unchanged", "statement": "After the cue-reward delay is lengthened, the dopamine cue response shows no significant change (Test 4).", "source": "Figure 4G to 4I; Results, Tests 3-7, paragraph 2", "quote": "Experimentally, we observed that while the animals learned the new delay rapidly, dopaminergic cue response showed no significant change (Fig 4 G-I).", "statistics": [ { "about": "delta AIC for dopamine against zero", "test": "one sample t-test", "statistic": "t(6) = -0.86", "p": 0.4244, "n": "7 animals" } ], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Dopamine delta AIC (constant minus changing) per animal -168.5 to 32.7, mean -18.7; t(6) = -0.73, p = 0.49 with k = 4 (paper t = -0.86, p = 0.42), and t = -0.65, p = 0.54 with k = 3. No change is found. Without HJ-FP-M3 (delta AIC -168.5), t(5) = 0.99, p = 0.37. Leaving out trials with a NaN dF/F sample gives t = -0.73. With the code's bounds, which the criterion fixes, this test cannot detect a change in these data. On dF/F as a fraction, the scale used, b is at most 10 where 5 of 7 animals have mean responses of 13.2 to 44.6; three animals have delta AIC of exactly -6.0, the changing model collapsed to the constant one, and in HJ-FP-M3 the changing model fits worse than the constant one. On dF/F in percent, the stored scale, the bounded model cannot fit the data (delta AIC -272 to -817 in every animal, t = -7.74), so that variant does not test the scale. Fitted without bounds the changing model is better in every animal (delta AIC 14 to 38, t = 7.84, p = 2.3e-4). The outcome rests on the assumption that the paper's fit was bounded as here; the paper's t of -0.86, close to -0.73, suggests it was. Which dF/F scale the bounds were set for is asked of the authors (findings).", "deviations": [ "The model fit and the dF/F scale are as in fig4hi-delay-change-behavior-changes.", "Assumed that the paper's fit applied the code's bounds (A -2 to 50, b -10 to 10) to dopamine values on a scale where they bind as they do here. Neither the Methods nor the code states the scale of the dF/F the Matlab scripts read from the authors' local files. The criterion fixes the bounds; the two scales the dandiset supports (fraction and percent) both give no change, and the paper's t (-0.86) is close to the one found here (-0.73), while the unbounded fit, outside the criterion, finds a change in every animal." ], "assumed": true, "disagreed": false, "data": "with_labels", "where": null, "proposed": null }, { "claim": "figs10b-no-omission-response", "statement": "After the delay change, there is no suppression of dopamine at the old reward time (3 s) in any animal, whereas omission of the reward in extinction does suppress dopamine at the reward time (Test 5).", "source": "Figure 4J and S10B; Results, Tests 3-7, paragraph 2; Methods, Experiments 2-5, paragraph 4", "quote": "Omission response is not negative with respect to pre-cue baseline for any animal after cue duration change but was negative during extinction.", "statistics": [ { "about": "cumulative omission-window response, delay change against extinction", "test": "paired t-test", "statistic": "t = 3.37", "p": 0.015, "n": "7 animals" } ], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Cumulative response at the old reward time on the first 8 s session is at or above 0 in 7 of 7 animals; the extinction omission response is negative on average (Methods' measure, mean -87.25, 6 of 7 negative; code's, -45.75, 5 of 7). Paired t = 3.03, p = 0.023 (Methods) and t = 3.42, p = 0.014 (code; paper t = 3.37, p = 0.0150). The criterion's reading (b), negative on average, is met; reading (a), in every animal, is not (F1 is positive under both measures). Leaving out the 0 to 4 trials per session with a NaN dF/F sample in the windows gives t = 3.01, p = 0.024.", "deviations": [], "assumed": false, "disagreed": false, "data": "with_labels", "where": null, "proposed": null }, { "claim": "fig4kl-extinction-abrupt-behavior", "statement": "Behavioral learning of extinction is abrupt, much more so than the change of the dopamine cue response.", "source": "Figure 4J to 4L", "quote": "Across all animals, the dopaminergic cue response remains significantly positive despite abrupt behavioral learning of extinction (paired t test for abruptness of change; t(6) = 5.67; two-tailed p = 0.0013; paired t test for change trial; t(6) = -2.40; two-tailed p = 0.0531; n=7 animals).", "statistics": [ { "about": "abruptness of change, lick against dopamine", "test": "paired t-test", "statistic": "t(6) = 5.67", "p": 0.0013, "n": "7 animals" }, { "about": "change trial, lick against dopamine", "test": "paired t-test", "statistic": "t(6) = -2.40", "p": 0.0531, "n": "7 animals" } ], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Lick abruptness exceeds dopamine abruptness with the Methods' lick measure (1.09 against 0.13, t = 3.28, p = 0.017) and the code's (0.47 against 0.13, t = 9.73, p = 6.8e-5; paper t = 5.67, p = 0.0013). The change-trial comparison gives t = -2.95, p = 0.026 (Methods) and t = -2.40, p = 0.0531 (code, the paper's values). 125 trials per animal, 0 to 4 of them with a NaN dF/F sample from -1 to 2 s; without them the abruptness t values are 3.43 and 9.39. The code variant's abruptness t (9.7) is larger than the paper's 5.67 while its change-trial t equals the paper's, for a reason not found here.", "deviations": [], "assumed": false, "disagreed": false, "data": "with_labels", "where": null, "proposed": null }, { "claim": "fig4jl-extinction-cue-response-positive", "statement": "The dopamine cue response remains significantly positive well after animals stop licking to the cue in extinction (Test 6).", "source": "Figure 4J to 4L; Results, Tests 3-7, paragraph 3", "quote": "As predicted by ANCCR but not RPE, dopamine cue response remained significantly positive well after animals cease to behaviorally respond to the cues (Fig 4J-L)", "statistics": [], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "(a) Last extinction session, mean CS+ response positive across animals, t(6) = 4.31, p = 0.0050 (4.33 without the 2 trials with a NaN dF/F sample); (b) trials after the lick change trial, t(6) = 4.16, p = 0.0060 with the code's lick measure and t(6) = 3.78, p = 0.0092 with the Methods'; (c) positive at p < 0.05 within the last session in 7 of 7 animals.", "deviations": [ "For reading (b) the trials after the lick change trial run to the end of the extinction sessions; the change trial is taken from both the code's and the Methods' lick measures, and both give every animal a positive mean after it (2.2 to 30.6)." ], "assumed": false, "disagreed": false, "data": "with_labels", "where": null, "proposed": null }, { "claim": "fig4p-background-faster-decay", "statement": "When unpredicted rewards are added during the ITI, the dopamine cue response stays positive but decreases faster over trials than during extinction (Test 7).", "source": "Figure 4M to 4P; Results, Tests 3-7, paragraph 4", "quote": "The dopamine cue response remained significantly positive but decayed across trials faster than during extinction (Fig 4M-P).", "statistics": [ { "about": "background rewards against extinction", "test": "paired t-test", "statistic": "t(6) = -3.51", "p": 0.0126, "n": "7 animals" } ], "kind": "finding", "inferred": "", "outcome": "not_reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "The background reward cue response stays positive after the change (mean of the 90 post-change trials over animals, t(6) = 3.40, p = 0.015). It decays further than in extinction on average under every reading, but not significantly; (a) normalized cumulative sum at trial 140, 1.60 against 2.13, paired t = -2.40, p = 0.053; (b) mean over the 90 trials, t = -1.97, p = 0.097; (c) post over pre mean, t = -2.40, p = 0.053 (paper t = -3.51, p = 0.0126). Background is below extinction in 5 of 7 animals. The first background reward session of HJ-FP-M4 (Day38) is not in the dandiset, so that animal's 90 post-change trials start a session later; without M4, reading (a) gives t(5) = -1.85, p = 0.12. Readings (a) and (c) are one affine function of the other ((a) = 1 + 1.8 (c)), so only two independent readings were tested. The criterion asks for background below extinction at p < 0.05 under every reading, which none meets, so the main part of the claim, the faster decay, is not reproduced. Leaving out the 0 to 4 trials per animal and condition with a NaN dF/F sample gives p = 0.048 for (a), 0.047 for (c) and 0.088 for (b). With the sessions of cumsum_pavlovian.m (conditions 4 and 3, one session before each) no reading shows a difference (t = 0.69 and 0.72); there HJ-FP-M6's background sum over the 50 trials before the change is 37.3 (288.1 with the sessions of cumsum_bgdrw_vs_extinction.m), giving it a reading (a) of 12.15, and HJ-FP-M2's is -492.9.", "deviations": [ "What the paired t-test compares is not in the Methods, legend or code; the three readings of the criterion are run and agree that background is not below extinction at p < 0.05.", "HJ-FP-M4's first Experiment 5 session is missing from the dandiset (session-day-link)." ], "assumed": false, "disagreed": true, "data": "with_labels", "where": null, "proposed": null }, { "claim": "fig5d-intermediate-cue-positive", "statement": "In the trial-less task, the dopamine response to an intermediate cue (one arriving before the previous cue's reward) is positive but weaker than the response to the previous cue (Test 8).", "source": "Figure 5C and 5D; Results, Test 8, paragraph 3", "quote": "We did not observe any negative dopamine response to the intermediate cue regardless of how baseline was measured, and instead observed a positive but weaker response", "statistics": [ { "about": "ratio of intermediate to previous cue response against zero", "test": "one sample t-test", "statistic": "t(6) = 6.64", "p": 0.00056, "n": "7 animals" } ], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Per animal ratio of intermediate to previous cue response 0.12 to 0.68, mean 0.435; t(6) = 6.66, p = 5.6e-4 against 0 (paper t = 6.64), and below 1 (t = -8.66, p = 1.3e-4). 7 to 18 pairs per animal over 2 to 4 sessions. None of the 190 cues of the pairs has a NaN dF/F sample in its windows.", "deviations": [], "assumed": false, "disagreed": false, "data": "with_labels", "where": null, "proposed": null }, { "claim": "fig5d-intermediate-reward-larger", "statement": "The dopamine response to the reward paired with an intermediate cue is larger than the response to the reward of the previous cue (the ratio is above 1).", "source": "Figure 5D, right", "quote": "and reward response (One sample t test against a null of one; t(6) = 2.95; two-tailed p value = 0.0256; n=7 animals)", "statistics": [ { "about": "ratio of intermediate to previous reward response against one", "test": "one sample t-test", "statistic": "t(6) = 2.95", "p": 0.0256, "n": "7 animals" } ], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Per animal ratio of intermediate to previous reward response 1.08 to 3.25, mean 1.85; t(6) = 2.96, p = 0.025 against 1 (paper t = 2.95, p = 0.0256).", "deviations": [], "assumed": false, "disagreed": false, "data": "with_labels", "where": null, "proposed": null }, { "claim": "figs11de-intermediate-cue-local-baseline", "statement": "With the baseline taken immediately before the intermediate cue and the response measured as a peak, the intermediate cue response is still positive.", "source": "Figure S11D and S11E; Methods, Experiment 6, paragraph 2", "quote": "Here, we measure baseline as the dopamine fluorescence immediately prior to the intermediate cue delivery, as opposed to the baseline prior to the first cue in Fig 5C, D. Cue response was defined as baseline subtracted peak response. The intermediate cue response was still positive.", "statistics": [ { "test": "one sample t-test of the cue response ratio", "statistic": "t = 10.42", "p": 4.6e-05, "n": "7 animals" } ], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Under all four readings (peak over 0 to 0.5 s or 0 to 1 s, baseline the last sample or the mean of the 0.5 s before the cue) the ratio of intermediate to previous peak response is positive in every animal (minimum 0.135), mean 0.25 to 0.45, t = 5.21 to 8.35, p <= 0.002 (paper t = 10.42).", "deviations": [ "The peak window and the baseline are not fixed by the paper; the four readings of the criterion are run and agree." ], "assumed": false, "disagreed": false, "data": "with_labels", "where": null, "proposed": null }, { "claim": "fig6b-no-backpropagation", "statement": "During initial cue-reward learning, dopamine activity does not shift backward from just before reward to the cue; the late (pre-reward) response does not differ from the early (cue) response in the first 100 trials (Test 9).", "source": "Figure 6A and 6B; Results, Tests 9-11, paragraph 2; Supplementary Note 9", "quote": "Our observations were not consistent with a backpropagating bump of activity and were instead consistent with an increase in cue response over trials of learning (Fig 6B)", "statistics": [ { "about": "late minus early normalized response, first 100 trials", "test": "paired t-test", "statistic": "t = -0.16", "p": 0.8771, "n": "7 animals" } ], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Normalized late minus early response over the first 100 CS+ trials, -0.008 with the Methods' baseline (t = -0.23, p = 0.83) and -0.006 without (t = -0.16, p = 0.876; paper t = -0.16, p = 0.8771). 473 to 900 trials per animal, so all have the 400 the normalizer needs. Leaving out the 0 to 4 trials per animal with a NaN dF/F sample from -1 to 3 s gives t = -0.22, p = 0.84.", "deviations": [], "assumed": false, "disagreed": false, "data": "with_labels", "where": null, "proposed": null }, { "claim": "fig6d-cues-increase-together", "statement": "In sequential conditioning without inhibition, the dopamine responses to CS2 and CS1 increase together early in learning (no difference in the first 50 trials) (Test 10).", "source": "Figure 6C and 6D; Results, Tests 9-11, paragraph 2", "quote": "Consistent with ANCCR, we experimentally found that cue2 and cue1 responses increased together early in learning prior to separating later in learning (Fig 6D).", "statistics": [ { "about": "CS2 minus CS1 normalized response, early in learning", "test": "paired t-test", "statistic": "t = -0.03", "p": 0.9752, "n": "7 animals" } ], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Normalized CS2 minus CS1 over the first 50 trials, mean -0.003, t = -0.03, p = 0.975 without a baseline (the paper's values), -0.002 (p = 0.98) and -0.001 (p = 0.99) with the 1 s and 1.5 s baselines. 225 to 350 trials per control animal.", "deviations": [ "The Methods' baseline window is not given; both readings of the criterion are run and agree with the code's unsubtracted measure." ], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig6d-cs2-below-cs1-late", "statement": "Later in sequential conditioning, the CS2 response falls below the CS1 response (the two separate).", "source": "Figure 6D, right; Results, Tests 9-11, paragraph 2", "quote": "cue2 and cue1 responses increased together early in learning prior to separating later in learning", "statistics": [], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "CS2 over CS1 response over trials 151 to 200, mean 0.549 (0.30 to 0.71) without a baseline, t(6) = -8.01, p = 2.0e-4 against 1; 0.573 and 0.574 with the 1 s and 1.5 s baselines (p <= 1.1e-4).", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig6g-inhibition-magnitude", "statement": "Optogenetic inhibition of VTA dopamine neurons strongly suppresses NAcc dopamine during the CS2-to-reward delay in DAT-Cre mice (to about -0.6 times the reward response on day 1) and reduces their reward response, compared with controls.", "source": "Figure 6F and 6G; Results, Tests 9-11, paragraph 2; Supplementary Note 10", "quote": "we first verified robust inhibition of mesolimbic dopamine release during the cue2\u2192reward delay in the experimental group (~0.6 times the reward response on day 1 of conditioning)", "statistics": [ { "about": "dopamine in the last 0.5 s before reward, DAT-Cre against controls", "test": "two sample t-test", "statistic": "t = -5.74", "p": 0.00013, "n": "13 animals" }, { "about": "reward response on day 1, DAT-Cre against controls", "test": "two sample t-test", "statistic": "t = -3.99", "p": 0.0021, "n": "13 animals" } ], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "With the baseline before CS1, dopamine 1 to 1.5 s after CS2 over the reward response is lower in the 6 DAT-Cre mice than in the 7 controls under every reading (t = -8.62, p = 3.2e-6 for (a) and (c); t = -6.86, p = 2.7e-5 for (b)), and the day 1 reward response over the random reward session's is lower (0.53 against 1.48, t = -4.13, p = 0.0017 with the 1 s window; t = -3.76, p = 0.0032 with 0.5 s). The DAT-Cre mean inhibition is -0.565 under reading (b), within tolerance of -0.6; (a) -0.247 and (c) -0.494 are not. With the code's baseline over every code-15 event (CS1 and CS2), (b) gives -0.632, t = -5.74, p = 1.3e-4, and the day 1 reward test with the 0.5 s window t = -3.99, p = 0.0021, both the paper's values. The excluded DAT-Cre mouse M8 has a negative day 1 reward response (-0.43), as Fig S12D says.", "deviations": [ "The claim's baseline is the 0.5 s before CS1; measure_inhibition_sequential.py takes it before every code-15 event, CS1 and CS2 alike (reftimes). Both are shown; the code's reproduces the paper's t values exactly.", "For the 1 s reward window the 0.5 s baseline AUC is doubled, so that the baseline covers a window of equal length (claims header); a choice made here, which reading (b) does not involve." ], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig6i-learning-despite-inhibition", "statement": "Despite inhibition of dopamine from CS2 to reward, every DAT-Cre animal learns anticipatory licking and acquires a positive dopamine response to CS1, both increasing over sessions as in controls (Test 11).", "source": "Figure 6I; Results, Tests 9-11, paragraph 2; Table S1", "quote": "Consistent with ANCCR, we observed that every experimental animal learned the task and that mesolimbic dopamine acquired positive responses to cue1 in all experimental animals (Fig 6I).", "statistics": [ { "about": "CS1 response across sessions, wild type", "test": "linear regression", "statistic": "t = 5.49", "p": 4.35e-06, "n": "35 sessions from 7 WT animals" }, { "about": "CS1 response across sessions, DAT-Cre", "test": "linear regression", "statistic": "t = 2.55", "p": 0.0166, "n": "30 sessions from 6 DAT-Cre animals" }, { "about": "anticipatory licks across sessions, wild type", "test": "linear regression", "statistic": "t = 3.61", "p": 0.00099, "n": "35 sessions from 7 WT animals" }, { "about": "anticipatory licks across sessions, DAT-Cre", "test": "linear regression", "statistic": "t = 4.92", "p": 3.4e-05, "n": "30 sessions from 6 DAT-Cre animals" } ], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Over the first three and last two sessions, CS1 response slope 0.134 per session in controls (t = 4.17, p = 2.1e-4, n = 35) and 0.076 in DAT-Cre mice (t = 2.87, p = 0.0077, n = 30); lick slopes 1.53 (t = 3.67, p = 8.4e-4) and 1.28 (t = 4.22, p = 2.3e-4) (paper t = 5.49, 2.55, 3.61, 4.92). In the last session all 6 DAT-Cre animals have a positive CS1 response (0.12 to 0.65 of the day 1 reward response) and positive anticipatory licking (0.52 to 10.78 licks). HJ-FP-WT-stGtACR-M2's last session has no dF/F for its CS1 trials 1 to 17, and its mean is over the other 33.", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig6i-lower-cs1-asymptote", "statement": "With inhibition, the CS1 dopamine response reaches a lower level in the last session than in controls.", "source": "Figure 6I; Supplementary Note 10, paragraph 2; Table S1", "quote": "mesolimbic dopamine acquired positive responses to cue1 in all experimental animals, albeit at a lower asymptote relative to the control group (Fig 6I)", "statistics": [ { "test": "two sample t-test", "statistic": "t = 2.80", "p": 0.0171, "n": "13 animals" } ], "kind": "finding", "inferred": "", "outcome": "not_reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Last-session normalized CS1 response, controls 0.690 (n 7) against DAT-Cre 0.359 (n 6), lower on average but t(11) = 1.70, p = 0.117 (paper t = 2.80, p = 0.0171). One control, WT-M3, is at 0.077, below every DAT-Cre value; its last session has 25 trials, where every other animal's has 50. The outcome turns on this session and on the normalisation; with WT-M3's previous session in its place t = 2.43 (p = 0.033), normalised by the random reward session's reward response t = 2.84 (p = 0.016), and not normalised t = 2.85 (p = 0.016). Supplementary Note 10 and learnign_curve_sequential.py normalise by the reward response on day 1 of conditioning, as the criterion does, and no passage excludes a short session, so the outcome is kept on the criterion's measure. HJ-FP-WT-stGtACR-M2's last session (Day6) has no dF/F for its first 1117 s, so its CS1 trials 1 to 17 have no value and its mean is over the other 33; with its previous session in place of it, its value is 0.462 against 0.459 and t = 1.70, p = 0.117. No other Experiment 7 CS1 trial lacks a value.", "deviations": [ "The measure follows learnign_curve_sequential.py (session means of the CS1 AUC over 0 to 1.5 s minus the 1.5 s before, over the first conditioning session's mean reward response)." ], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig6i-last-session-licking-similar", "statement": "In the last session, anticipatory licking does not differ between DAT-Cre and control animals.", "source": "Figure 6I, right; Table S1", "quote": "", "statistics": [ { "test": "two sample t-test", "statistic": "t = 0.88", "p": 0.3999, "n": "13 animals" } ], "kind": "finding", "inferred": "Only Table S1 gives this comparison (Behavior (right), t = 0.88, p = 0.3999); no sentence states it. The Supplementary Note 10 says the experimental animals learned \"albeit slower than control animals\".", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Last-session anticipatory licks 7.25 in controls against 4.84 in DAT-Cre mice, difference 2.41, t(11) = 1.17, p = 0.27 (paper t = 0.88, p = 0.40). No difference is found.", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "figs12f-pre-cs2-licking", "statement": "Anticipatory licking between CS1 and CS2, before inhibition starts, increases from the first to the last session (main effect of session), with no effect of genotype and no genotype by session interaction.", "source": "Figure S12F; Table S1", "quote": "Quantification of pre-CS2 anticipatory licks showing reward-seeking prior to CS2 (onset of dopamine neuron inhibition).", "statistics": [ { "about": "session effect", "test": "two-way mixed ANOVA", "statistic": "F = 9.07", "p": 0.0118, "n": "13 animals" }, { "about": "genotype effect", "test": "two-way mixed ANOVA", "statistic": "F = 0.25", "p": 0.6241, "n": "13 animals" }, { "about": "genotype by session interaction", "test": "two-way mixed ANOVA", "statistic": "F = 0.07", "p": 0.7915, "n": "13 animals" } ], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_jeong_2022.ipynb", "notes": "Pre-CS2 licks rise from 0.33 to 1.62 between the first and last sessions; session F(1, 11) = 7.39, p = 0.020 (paper F = 9.07, p = 0.0118), genotype F = 0.57, p = 0.47, interaction F = 0.0003, p = 0.99. With the code's baseline window (1500 s before CS1), session F = 5.23, p = 0.043, genotype p = 0.57, interaction p = 0.60.", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig2-anccr-classic-results", "statement": "Simulated ANCCR reproduces classic findings taken as support for RPE coding, including conditioned inhibition, blocking, overexpectation, trial-by-trial behavioral changes after dopamine inhibition at reward, and unblocking by dopamine activation.", "source": "Figure 2D to 2I; Table S1", "quote": "Simulations of ANCCR learning produces behavior consistent with conditioned inhibition (D), blocking (E), and overexpectation (F). ... H. Simulation of dopamine inhibition at reward time produces trial-by-trial changes in behavior (similar to (81)). I. Simulation of unblocking due to dopamine activation at reward during blocking (similar to (14)).", "statistics": [ { "about": "conditioned inhibition, probability of lick to CS", "test": "paired t-test", "statistic": "t = 4.42", "p": 0.0003, "n": "20 iterations" }, { "about": "blocking, probability of lick to CS2", "test": "two sample t-test", "statistic": "t = 7.32", "p": 9.2e-07, "n": "20 iterations" }, { "about": "overexpectation, probability of lick to CS1", "test": "paired t-test", "statistic": "t = 4.09", "p": 0.00062, "n": "20 iterations" }, { "about": "inhibition at reward, probability of Action 1", "test": "paired t-test", "statistic": "t = 7.06", "p": 1e-06, "n": "20 iterations" }, { "about": "unblocking, probability of lick to CS2", "test": "two sample t-test", "statistic": "t = -6.33", "p": 2e-07, "n": "20 iterations" } ], "kind": "finding", "inferred": "", "outcome": "blocked", "blocked_by": "not_runnable", "notebook": null, "notes": "A simulation of the ANCCR and TDRL models, which uses no recorded data. The paper's simulation code is Matlab (simulation/ and functions/calculateANCCR.m of the ANCCR repository at the commit the claims file cites), and Matlab cannot be installed in the entry's locked Python environment; GNU Octave is not on the runner and cannot be installed from the lock file either. A port of the models to Python was not attempted. Nothing in the dandiset bears on it.", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig3cf-simulated-unpredicted-rewards", "statement": "In simulations of random rewards, TDRL RPE (CSC and microstimulus) predicts a reward response that decreases with experience and is negatively correlated with the previous IRI, while ANCCR predicts an increasing response positively correlated with the previous IRI.", "source": "Figure 3B, 3C and 3F", "quote": "C. Simulations confirming the intuitive reasoning from B for Test 1. ... F. Simulations confirming the intuitive reasoning from B for Test 2", "statistics": [ { "about": "Test 1, RPE (CSC)", "test": "one sample t-test against zero", "statistic": "t(99) = -65.74", "p": 1.7e-83, "n": "100 simulations" }, { "about": "Test 1, RPE (MS)", "test": "one sample t-test against zero", "statistic": "t(99) = -27.57", "p": 3e-48, "n": "100 simulations" }, { "about": "Test 1, ANCCR", "test": "one sample t-test against zero", "statistic": "t(99) = 18.60", "p": 4.5e-34, "n": "100 simulations" }, { "about": "Test 2, RPE (CSC)", "test": "one sample t-test against zero", "statistic": "t(99) = -1.7\u00d710^3", "p": 5e-223, "n": "100 iterations" }, { "about": "Test 2, RPE (MS)", "test": "one sample t-test against zero", "statistic": "t(99) = -151.28", "p": 6.3e-119, "n": "100 iterations" }, { "about": "Test 2, ANCCR", "test": "one sample t-test against zero", "statistic": "t(99) = 335.03", "p": 4.8e-153, "n": "100 iterations" } ], "kind": "finding", "inferred": "", "outcome": "blocked", "blocked_by": "not_runnable", "notebook": null, "notes": "A simulation of the ANCCR and TDRL models, which uses no recorded data. The paper's simulation code is Matlab (simulation/ and functions/calculateANCCR.m of the ANCCR repository at the commit the claims file cites), and Matlab cannot be installed in the entry's locked Python environment; GNU Octave is not on the runner and cannot be installed from the lock file either. A port of the models to Python was not attempted. Nothing in the dandiset bears on it.", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig5b-simulated-intermediate-cue", "statement": "In simulations of the trial-less task, TDRL RPE predicts a negative response to an intermediate cue relative to the previous cue, while ANCCR predicts a positive one; both predict a reward response ratio (intermediate over previous reward) above 1.", "source": "Figure 5A and 5B; Results, Test 8, paragraphs 2 and 3", "quote": "B. Confirmation of these intuitions based on simulations (Methods) (One sample t test against a null of zero; t(99) = RPE (CSC), -114.74; RPE (MS), -181.32; ANCCR, 322.53 ... Reward responses are predicted to be positive by both models (One sample t test against a null of one", "statistics": [ { "about": "cue ratio, RPE (CSC)", "test": "one sample t-test against zero", "statistic": "t(99) = -114.74", "p": 4.1e-107, "n": "100 iterations" }, { "about": "cue ratio, RPE (MS)", "test": "one sample t-test against zero", "statistic": "t(99) = -181.32", "p": 1.1e-126, "n": "100 iterations" }, { "about": "cue ratio, ANCCR", "test": "one sample t-test against zero", "statistic": "t(99) = 322.53", "p": 2.1e-151, "n": "100 iterations" }, { "about": "reward ratio against one, RPE (CSC)", "test": "one sample t-test against one", "statistic": "t(99) = 87.67", "p": 1.2e-95, "n": "100 iterations" }, { "about": "reward ratio against one, RPE (MS)", "test": "one sample t-test against one", "statistic": "t(99) = 62.86", "p": 1.3e-81, "n": "100 iterations" }, { "about": "reward ratio against one, ANCCR", "test": "one sample t-test against one", "statistic": "t(99) = 16.78", "p": 1.1e-30, "n": "100 iterations" } ], "kind": "finding", "inferred": "", "outcome": "blocked", "blocked_by": "not_runnable", "notebook": null, "notes": "A simulation of the ANCCR and TDRL models, which uses no recorded data. The paper's simulation code is Matlab (simulation/ and functions/calculateANCCR.m of the ANCCR repository at the commit the claims file cites), and Matlab cannot be installed in the entry's locked Python environment; GNU Octave is not on the runner and cannot be installed from the lock file either. A port of the models to Python was not attempted. Nothing in the dandiset bears on it.", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig6h-simulated-inhibition", "statement": "With dopamine inhibited from CS2 to reward, TDRL RPE predicts a negative CS1 response and no learning, while ANCCR predicts a positive but reduced CS1 response.", "source": "Figure 6H and S12A to S12C; Results, Tests 9-11, paragraph 2", "quote": "With such strong inhibition, TDRL RPE predicted no behavioral learning in this experiment, and a strong negative cue1 dopamine response (Figs 6H, S12). In contrast, ANCCR predicted largely intact learning of cue1, but with slower behavioral learning and reduced cue1 response", "statistics": [ { "about": "CS1 response, inhibited against control, RPE", "test": "two sample t-test", "statistic": "t = -379.13", "p": 1.3e-69, "n": "20 iterations" }, { "about": "CS1 response, inhibited against control, ANCCR", "test": "two sample t-test", "statistic": "t = -10.22", "p": 1.9e-12, "n": "20 iterations" } ], "kind": "finding", "inferred": "", "outcome": "blocked", "blocked_by": "not_runnable", "notebook": null, "notes": "A simulation of the ANCCR and TDRL models, which uses no recorded data. The paper's simulation code is Matlab (simulation/ and functions/calculateANCCR.m of the ANCCR repository at the commit the claims file cites), and Matlab cannot be installed in the entry's locked Python environment; GNU Octave is not on the runner and cannot be installed from the lock file either. A port of the models to Python was not attempted. Nothing in the dandiset bears on it.", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "figs6-learns-causal-structure", "statement": "In simulations where cues arrive at exponential intervals and each is followed by reward at a fixed delay, ANCCR for the cue is significantly positive at every ratio of delay to inter-cue interval, while TDRL RPE is often lower than without an association.", "source": "Figure S6A; Results, A retrospective causal learning algorithm, paragraph 2", "quote": "However, across all ratios shown, contingency learning signals a significantly positive ANCCR for the cue, implying that the algorithm learns the cue-reward association without loss of generality.", "statistics": [], "kind": "finding", "inferred": "", "outcome": "blocked", "blocked_by": "not_runnable", "notebook": null, "notes": "A simulation of the ANCCR and TDRL models, which uses no recorded data. The paper's simulation code is Matlab (simulation/ and functions/calculateANCCR.m of the ANCCR repository at the commit the claims file cites), and Matlab cannot be installed in the entry's locked Python environment; GNU Octave is not on the runner and cannot be installed from the lock file either. A port of the models to Python was not attempted. Nothing in the dandiset bears on it.", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "figs5-timescale-invariance", "statement": "Simulated ANCCR reproduces the timescale invariance of learning (trials to acquisition grow with the cue-reward delay only when the ITI is fixed, not when it is scaled with the delay), and TDRL does not.", "source": "Figure S5C and S5D; Results, A retrospective causal learning algorithm, paragraph 2", "quote": "C. Simulations demonstrating that TDRL cannot explain the timescale invariance, but that retrospective causal learning explains these results (Methods).", "statistics": [], "kind": "finding", "inferred": "", "outcome": "blocked", "blocked_by": "not_runnable", "notebook": null, "notes": "A simulation of the ANCCR and TDRL models, which uses no recorded data. The paper's simulation code is Matlab (simulation/ and functions/calculateANCCR.m of the ANCCR repository at the commit the claims file cites), and Matlab cannot be installed in the entry's locked Python environment; GNU Octave is not on the runner and cannot be installed from the lock file either. A port of the models to Python was not attempted. Nothing in the dandiset bears on it.", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "figs13-ramping", "statement": "Simulated ANCCR reproduces dopamine ramps when a series of cues predicts reward, and the changes of the ramp when the subject is teleported closer to reward or the cue speed changes.", "source": "Figure S13A to S13D; Table S1", "quote": "ANCCR signaling not only ramps up when a series of cues predicts reward, but also captures observed changes in dopamine activity when a subject was teleported to a moment closer to reward (A-B) or the speed of cue presentation changed (C-D).", "statistics": [ { "about": "teleport T1 against standard", "test": "paired t-test", "statistic": "t = 4.36", "p": 3.2e-05, "n": "100 iterations" }, { "about": "teleport T2 against standard", "test": "paired t-test", "statistic": "t = 23.42", "p": 3.7e-42, "n": "100 iterations" }, { "about": "teleport T3 against standard", "test": "paired t-test", "statistic": "t = 29.75", "p": 3.6e-51, "n": "100 iterations" }, { "about": "slow against standard speed", "test": "paired t-test", "statistic": "t = -129.07", "p": 5.6e-105, "n": "100 iterations" }, { "about": "fast against standard speed", "test": "paired t-test", "statistic": "t = 109.14", "p": 3.9e-112, "n": "100 iterations" } ], "kind": "finding", "inferred": "", "outcome": "blocked", "blocked_by": "not_runnable", "notebook": null, "notes": "A simulation of the ANCCR and TDRL models, which uses no recorded data. The paper's simulation code is Matlab (simulation/ and functions/calculateANCCR.m of the ANCCR repository at the commit the claims file cites), and Matlab cannot be installed in the entry's locked Python environment; GNU Octave is not on the runner and cannot be installed from the lock file either. A port of the models to Python was not attempted. Nothing in the dandiset bears on it.", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null } ], "findings": [ { "kind": "metadata_error", "statement": "Event times (acquisition/eventlog/eventtime) are stated in milliseconds but are in seconds. The random reward sessions have a mean inter-reward interval of about 12 in the files' units (Methods, 12 s), and the Fig S8L sessions named 6, 12 and 30 have means near 6, 12 and 30.", "found": null, "status": null }, { "kind": "data_missing", "statement": "HJ-FP-M4 Day38, a start day listed in cumsum_pavlovian.m and cumsum_bgdrw_vs_extinction.m and that animal's first background reward session, is not in the dandiset. Six other integer days inside the day ranges of the paper's analysis code have no session either (HJ-FP-M2 Day23, M3 Day16 and Day31, M4 Day30, F2 Day16, M6 Day49). The paper's mean session counts for Experiments 2, 5 and 4 (13.3, 8.1, 2.6) equal the counts that include all seven days, while the dandiset gives 12.7, 7.9 and 2.4, which suggests that these are missing sessions too.", "found": null, "status": null }, { "kind": "metadata_error", "statement": "In 65 of the 101 Experiment 7 files the raw photometry (acquisition/photometry) has about 12 million samples per stated second and its timestamps end at 0.43 to 4.18, while the sessions last 600 to 4100 s; the timestamps are in thousands of seconds (12 kHz detection).", "found": null, "status": null }, { "kind": "metadata_error", "statement": "Subject HJ-FP-datHT-stGtACR-M7, a DAT-Cre heterozygote by its name and in the paper's code, has genotype wt in its files.", "found": null, "status": null }, { "kind": "metadata_insufficient", "statement": "The experimental condition of each session (Experiments 2 to 5) is not stated; only the first session of a condition carries a suffix in its name (8s, 8s-bgd6s, extinction), and the files of datHT-M6 and datHT-M7 do not name the task. The sessions used for each figure are given only in the analysis code's day lists and in a Google spreadsheet named in the description, which has no fixed version.", "found": null, "status": null }, { "kind": "other", "statement": "In 13 sessions dF/F does not cover the whole event log by more than 1 s; HJ-FP-WT-stGtACR-M2 Day6 has no dF/F for its first 1117 s, and M6 Day22 and M7 Day43 lose their last 175 to 187 s. In Experiments 1 to 6 the first 20 or so dF/F timestamps of each session are NaN.", "found": null, "status": null }, { "kind": "methods_unavailable", "statement": "For the authors. On which scale of dF/F (fraction or percent, and with the AUC in ms) were the bounds of the dopamine changing model in cumsum_pavlovian.m (A -2 to 50, b -10 to 10) set, for the Fig 4I test? On the dandiset's dF/F they bind on either scale. As a fraction, the changing model cannot reach the mean response of 5 of 7 animals and no change is found (t = -0.73, close to the paper's -0.86); in percent it cannot fit at all; without bounds it finds a change in every animal (t = 7.84).", "found": null, "status": null }, { "kind": "methods_unavailable", "statement": "For the authors. Which window (and measure) of the z-scored lick bout onset response were the Fig S8E tests run on? Neither the Methods, the legend nor the code gives it. The non-consummatory response is significantly negative over 0 to 0.5 s (t = -2.56, close to the paper's -2.55) but not over 0 to 1 s or -0.5 to 1 s, and the consummatory t values are about twice the paper's.", "found": null, "status": null } ], "reconciliation": null, "verifications": [ { "date": "2026-10-30", "kind": "agent", "by": "Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook)", "reviewed": "Round 1. The first executed notebook with the conclusion cells and the summary removed, the helper module and the claims file, with the paper. The verifier checked the archive listing and tested helper functions on synthetic inputs. Its report gave its own outcome for each claim; 36 of 38 agree with the record as it was, and it disagreed on figs8e-lick-bout-responses and fig4p-background-faster-decay.", "agreed": 36, "of": 38, "disagreements": [ { "claim": "figs8e-lick-bout-responses", "outcome": "blocked", "reason": "The paired and consummatory tests hold under all three windows. The non-consummatory response is below 0 at p < 0.05 only under reading (c) (t = -2.559, p = 0.038); under (a) p = 0.14 and under (b) p = 0.80. The readings differ, so the claims header's rule makes this blocked (methods_underspecified)." }, { "claim": "fig4p-background-faster-decay", "outcome": "not_reproduced", "reason": "Background is not below extinction at p < 0.05 under any reading, p = 0.053 for (a) and (c), p = 0.097 for (b); the paper has t = -3.51, p = 0.0126. The remains-positive part holds (p = 0.015). Readings (a) and (c) are affine transforms of each other (nc[140] = 1 + 1.8 c), so only two independent readings were tested. Per-animal outputs under alternative session or normalisation choices would show whether the failure comes from such a choice." } ] }, { "date": "2026-10-30", "kind": "agent", "by": "Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook)", "reviewed": "Round 2. The full executed notebook with the conclusions and summary, the helper module, the claims file and the attempt record, with the paper. The verifier agreed with all 38 outcomes, and asked for an explicit assumption on fig4hi-delay-change-dopamine-unchanged, a count of Experiment 7 trials without dF/F, and corrections to sentences the outputs do not support (cells 6, 45, 68, 80, 102 and 119, the record notes of methods-session-counts and the data_missing finding).", "agreed": 38, "of": 38, "disagreements": [] } ] } ], "entries": [ { "name": "000351-jeong-2022", "type": "reproduction", "title": "Mesolimbic dopamine release conveys causal associations (Jeong et al. 2022)", "description": "Tests the findings of Jeong, Taylor, Floeder et al. (2022) on dandiset 000351, which holds fiber photometry of dopamine release in the nucleus accumbens and the event logs of the random reward, Pavlovian, trial-less and sequential conditioning experiments, and the licking of a separate group given random rewards at three intervals. It repeats the paper's tests of how the reward and cue responses change with experience, the comparisons of dopaminergic and behavioural learning, the intermediate cue test and the optogenetic inhibition experiment, with the session lists of the paper's code. The model simulations are not tested.", "datasets": [ "dandi:000351" ], "papers": [ "10.1126/science.abq6740" ], "derived_from": [], "notebooks": [ { "path": "reproduce_jeong_2022.ipynb", "title": "Reproducing Jeong et al. (2022) from dandiset 000351", "executed": "notebooks/000351-jeong-2022/reproduce_jeong_2022.html" } ] } ] }