DANDI:001845 · Encore Science

Claims 35

From the paper A Novel Cooperative Behavior Assay in Rats Reveals Distinct Social Coordination Strategies of Increasing Complexity and Social Reciprocity Deficits in a Model of Fragile X Syndrome · claims file · checked 2026-10-21: ready

fig1e-match-rate-wt-higher
WT pairs reach significantly higher match (coordination) rates than Fmr1 pairs under both the 100% and the 50% reward contingency.
WT pairs demonstrated significantly higher coordination rates than both Fmr1 pairs ( Fig. 1 E , I )
Figure 1E and legend; Results, A spatial cooperation task for social coordination in rat pairs
How it will be judged
Per pair and session, the match rate, a match being a simultaneous nose-poke of the two rats at corresponding wells (Methods, Cooperative learning task), counted under rate readings r1 and r2 of the opening comment; WT and Fmr1 pairs, first 21 sessions at 100% and first 35 at 50% (Methods, Behavioral quantification). The genotype effect of the two-way ANOVA under readings A and B of the opening comment. Reproduced when, in both contingencies, the genotype effect has p < 0.05 with the WT mean higher.
fig1f-transition-rate-wt-higher
Fmr1 rats make fewer task-related arm transitions than WT rats (they are slower in the task), under both contingencies.
Fmr1 and mixed pairs were generally slower in the task, as reflected by a reduced number of task-related transitions ( Fig. 1 F , G ; Fig. S1 D )
Figure 1F and legend; Results, A spatial cooperation task for social coordination in rat pairs
How it will be judged
Per rat and session, the number of transitions (visits to a well other than the rat's previous one), under rate readings r1 and r2; rats of the WT and Fmr1 pairs; first 21 and first 35 sessions. The genotype effect of the two-way ANOVA under readings A and B, with rats as units (the df show 20 rats). Reproduced when, in both contingencies, p < 0.05 with the WT mean higher.
fig1g-match-transition-regression
In the 50% condition, match rate rises linearly with transition rate in both genotypes, with a steeper slope and a better fit in WT (beta 0.79, R2 0.74) than in Fmr1 (beta 0.64, R2 0.44).
Linear regression between cooperative match event rates and transition rates for 50% reward condition
Figure 1G legend
How it will be judged
An ordinary least-squares regression of match rate on transition rate per genotype, first 35 sessions at 50%. The legend gives n = 35 sessions, but its p-values imply far more points than 35, and the units of the rates are not given. Readings of the points are (a) one per pair and session, with the pair's summed transitions; (b) one per rat and session, the rat's transitions against its pair's matches; (c) one per session, pairs averaged; each under rate readings r1 and r2. Reproduced when, under one reading, both slopes and both R2 values are within tolerance and both regressions have p < 0.05; partially reproduced when both are significant with WT having the higher slope and R2 but the values are out of tolerance.
fig1h-above-chance-late
Both WT and Fmr1 pairs coordinate no better than chance early in training, and above chance late in training, by a permutation test that circularly shifts one rat's sequence of well visits.
Both WT and Fmr1 rat pairs exhibited performance indistinguishable from chance during early training on both reward contingency tasks; however, by late training, their performance improved significantly, exceeding chance levels ( Fig. 1 H , I )
Figure 1H and legend; Results, A spatial cooperation task for social coordination in rat pairs
How it will be judged
Per pair and session, performance (Eq. 1) against a null from 1,000 circular time shifts of one rat's whole sequence of reward well visits within the session, p being the fraction of shifts whose performance equals or exceeds the observed one (Methods, Permutation testing of coordination performance). Open, not fixed by the paper are which rat is shifted (the readings, the first rat as the dataset orders them, and the second; the result is reported for both) and the range of shifts (a uniform random shift over the session length is the standard choice). The paper shows one pair per genotype. "Early" and "late" can be read as (a) the first session at 100% and the last analysed session at 50%, as the legend's examples; (b) the first and last analysed session of each contingency, as the Results say. The Results state it for "Both WT and Fmr1 rat pairs", that is every pair. The early and late readings name which sessions the paper means, and one of them suffices; the shifted-rat readings are a method detail, and the outcome must hold under both of them, else the claim is blocked as methods_underspecified. Reproduced when every pair of both genotypes has p >= 0.05 in the early session(s) and p < 0.05 in the late session(s); partially reproduced when that holds in at least 3 of 5 pairs per genotype, or when the late sessions are above chance in every pair but the early ones are too.
fig1i-performance-wt-higher
Cooperative performance (matches relative to transitions, Eq. 1) is significantly lower in Fmr1 pairs than in WT pairs under both contingencies, so their impairment is not only due to slower transitions.
Performance metrics for WT and Fmr1 −/y pairs in 100% (left) and 50% (right) reward contingencies shows significant impairment in Fmr1 −/y pairs compared with WT pairs
Figure 1I and legend; Results, A spatial cooperation task for social coordination in rat pairs
How it will be judged
Performance per pair and session as Eq. 1 (opening comment), first 21 and first 35 sessions; the genotype effect of the two-way ANOVA under readings A and B. Reproduced when, in both contingencies, p < 0.05 with the WT mean higher.
fig1j-partner-location-predicts-choice
A multinomial logistic regression predicts a rat's next choice from its partner's current location above chance in both genotypes, and more accurately in WT than in Fmr1 rats, under both contingencies.
This predictive relationship exceeded chance levels across all rat pairs. However, prediction accuracy was significantly higher in WT pairs compared with both Fmr1 and mixed-genotype pairs
Figure 1J and legend; Results, A spatial cooperation task for social coordination in rat pairs
How it will be judged
As in the Methods (Multinomial logistic regression and cross-validation), per rat and contingency, a scikit-learn multinomial logistic regression (default settings, the standard choice) of the rat's next well on the partner's current well, one-hot encoded, scored by accuracy averaged over fivefold stratified cross-validation; the null is 1,000 independent shuffles of the two vectors with the cross-validation repeated, each rat's null value being its mean shuffled accuracy; groups compared with a two-tailed Mann-Whitney U test; 10 WT and 10 Fmr1 rats (U = 100 implies 10 per group). Open, not fixed by the paper is the moment at which the partner's location is read. Readings are the partner's well (a) when the focal rat leaves its current well; (b) at the focal rat's last poke before the choice; (c) at the moment the focal rat arrives at its next well. Also open is whether a rat's sessions of a contingency are pooled (reading p1, as "for each rat individually" suggests) or scored per session and averaged (p2). Reproduced when, under every reading, in both contingencies real accuracy exceeds shuffled in each genotype (p < 0.05) and WT real accuracy exceeds Fmr1 real accuracy (p < 0.05); blocked as methods_underspecified when the readings disagree.
figs1-mixed-pairs-impaired
Mixed WT-Fmr1 pairs also reach lower coordination rates and performance, make fewer transitions and have lower partner-based prediction accuracy than WT pairs.
WT pairs demonstrated significantly higher coordination rates than both Fmr1 pairs ( Fig. 1 E , I ) and mixed pairs ( Fig. S1 A ) under both 100% and 50% reward contingencies ... prediction accuracy was significantly higher in WT pairs compared with both Fmr1 and mixed-genotype pairs
Results, A spatial cooperation task for social coordination in rat pairs; Figure S1A to S1D (not read)
How it will be judged
The tests of Figure S1 are in the supplement, which was not read. The readings are the tests of the matching Figure 1 panels, with mixed pairs in place of Fmr1 pairs, as in fig1e-match-rate-wt-higher, fig1f-transition-rate-wt-higher, fig1i-performance-wt-higher and fig1j-partner-location-predicts-choice, each under the readings named there. Reproduced when, in both contingencies, WT exceeds mixed pairs with p < 0.05 for match rate and for prediction accuracy; partially reproduced when that holds in one contingency. Transitions and performance are reported.
fig2b-opaque-barrier-lowers-performance
Replacing the transparent divider with an opaque one (with white noise) lowers the performance of WT and Fmr1 pairs compared with standard 50% sessions, with no genotype difference in this effect.
Compared with the 50% reward condition, rats showed a significant reduction in performance in the SD sessions ( Fig. 2 B , C )
Figure 2B and legend; Results, Cooperation deficits in Fmr1 rats are not due to sensory or learning deficits
How it will be judged
Performance per session as Eq. 1. Per pair, the mean over its SD sessions and the mean over its transparent 50% sessions, these being the last 9 transparent sessions (legend, right panel); a mixed-design ANOVA (genotype between, condition within, pair as subject; the df F(1,4) fit 6 pairs) and a paired t test per genotype. Open for the left panel is which transparent sessions are compared; readings are (a) the last 9 transparent sessions, (b) the transparent sessions interspersed with the SD sessions. Reproduced when, under one reading, the condition effect has p < 0.05 with SD lower, both paired t tests have p < 0.05 with SD lower, and the genotype x condition interaction of the same ANOVA has p >= 0.05 (the absence part, judged on that test); partially reproduced when the condition effect holds but a paired test does not, or the interaction is significant.
fig2c-opaque-barrier-mixed-pairs
In mixed pairs the opaque barrier also lowers performance (condition effect with WT pairs), though the paired test within mixed pairs is not significant.
Comparison of mean performance of rat pairs between standard and sensory-deprived conditions
Figure 2C and legend
How it will be judged
As fig2b-opaque-barrier-lowers-performance, with the 3 WT pairs and 5 mixed pairs. Reproduced when the condition effect has p < 0.05 with SD lower; the mixed-pair paired t test is reported with its p (the paper's is not significant).
fig2e-visual-cue-learning-similar
Both WT and Fmr1 rats learn the visual cue-reward association task; performance rises over sessions at 100% reward, and there is no significant genotype effect or genotype x session interaction at 100% or 50%. The Results also say Fmr1 rats learned more slowly, which the reported interaction tests do not show.
Both WT and Fmr1 rats learnt to approach the reward-associated arm based on the visual cue ( Fig. 2 E ), demonstrating intact visual cue-reward associative learning
Figure 2E and legend; Results, Cooperation deficits in Fmr1 rats are not due to sensory or learning deficits
How it will be judged
A claim of absence. Percent correct per rat and session; per contingency a linear mixed-effects model with genotype, session (continuous, as the df of 1 show) and their interaction as fixed effects and a random intercept per rat (legend, Figure 2), F tests with Satterthwaite df (the fractional df). Reproduced when the genotype effect has p >= 0.05 in both contingencies and the session effect at 100% has p < 0.05 with performance rising; not reproduced when the genotype effect is significant. The genotype x session terms are reported. The Methods give p = 0.981 for genotype in a model with contingency as a factor; that model (rat random intercept, session, genotype and contingency fixed) is shown as a variant.
fig2f-cue-latency-wt-shorter
In the visual-cue task WT rats respond to the cue with shorter latency than Fmr1 rats over pooled responses (100%, 22.0 against 42.2 s; 50%, 14.0 against 28.4 s); in the session-wise mixed-effects models with rat as random intercept the genotype effect is not significant (p = 0.058 at 100%, 0.064 at 50%).
WT rats respond to the visual LED cue with lower latency relative to Fmr1 −/y rats under both 100% (left) and 50% (right) reward contingencies
Figure 2F and legend
How it will be judged
Latency from cue onset to the correct poke, pooled over responses of all rats of a genotype (the n are responses), two-sided Mann-Whitney U test per contingency. Open is whether latency is counted to the correct poke only (the Methods say "latency to correct choice") or to any poke; the reading is the correct poke, with any poke shown as a variant. Reproduced when both tests have p < 0.05 with WT shorter and the means are within tolerance; partially reproduced when the tests hold but a mean is out of tolerance. The session-wise latency models of the legend (genotype, session and their interaction fixed, random intercept per rat, per contingency) are run and reported; they are not the basis of the outcome.
fig3c-triplets-by-contingency
Under 100% reward rats mostly alternate between two arms, while under 50% they explore all three arms through the optimal 1-2-3 or 1-3-2 triplets, especially WT rats.
Parsing each session's transitions into nonoverlapping triplets revealed that rats in the 100% condition predominantly alternated between two adjacent arms, while those in the 50% condition engaged in comprehensive exploration of all three arms through either the 1-2-3 or 1-3-2 sequences, especially for WT pairs
Figure 3B and 3C; Results, Development of structured spatial and temporal coordination strategies supports cooperative behavior
How it will be judged
The paper shows example rats and gives no test. Per rat and session, the choice sequence (opening comment) cut into consecutive nonoverlapping triplets from its first visit (Methods, Choice triplet counts); the fraction of optimal triplets. The Methods list eight possible triplets, though twelve without a repeated well exist (2-1-3, 2-3-1, 3-1-2 and 3-2-1 are missing), and name 1-2-3 and 1-3-2 as optimal because they visit all three wells. Readings are (o1) every triplet that visits all three wells is optimal, as the rationale says; (o2) only the literal 1-2-3 and 1-3-2. Per rat, over all its triplets of the first 21 (100%) or first 35 (50%) sessions, the fraction of alternation triplets (x-y-x, the same under both readings) and the fraction of optimal triplets. The paper gives no number for "predominantly"; it is read as more than half. Reproduced when, under every reading, the median WT and median Fmr1 alternation fractions at 100% exceed 0.5, the median WT optimal fraction at 50% exceeds the WT median at 100%, and the WT median optimal fraction at 50% exceeds the Fmr1 median at 50%; blocked as methods_underspecified when the readings disagree. The fractions per rat are reported.
fig3d-entropy-wt-lower
The entropy of WT rats' choice sequences is lower than that of Fmr1 rats, a genotype effect under both contingencies. The legend describes a larger decrease with learning in WT rats, but the genotype x session interaction is not significant at either contingency.
Session-wise comparison of entropy of choice sequences of WT and Fmr1 −/y rats during 100% (left) and 50% (right) reward contingencies, showing significantly higher decrease in entropy with learning for WT rats
Figure 3D and legend; Results, Development of structured spatial and temporal coordination strategies supports cooperative behavior
How it will be judged
Per rat and session, the sample entropy of the full choice sequence (Methods). The embedding dimension m and tolerance r are not given; with wells coded 1, 2, 3 any r below 1 means exact matching, the usual r = 0.2 SD falls there, and readings are m = 2 (the standard) and m = 3 (the triplet unit of the Methods), with r = 0.2 times the sequence SD. The genotype effect of the two-way ANOVA under readings A and B, first 21 and first 35 sessions. Reproduced when, under every reading, in both contingencies the genotype effect has p < 0.05 with WT entropy lower; partially reproduced when, under every reading, that holds in the 50% contingency only; blocked as methods_underspecified when the readings disagree. The slope of entropy over sessions per genotype and the genotype x session terms are reported.
fig3e-exploration-efficiency-wt-higher
WT rats have higher exploration efficiency (fraction of optimal choice triplets) than Fmr1 rats, strongly in the 50% condition and modestly in the 100% condition.
Exploration efficiencies of WT and Fmr1 −/y rats during 100% (left) and 50% (right) contingency sessions, showing significantly higher exploration efficiency for WT rats in the 50% condition
Figure 3E and legend; Results, Development of structured spatial and temporal coordination strategies supports cooperative behavior
How it will be judged
Exploration efficiency per rat and session (Methods, Eq. 2), the number of optimal triplets over all nonoverlapping triplets, under the readings o1 and o2 of optimal named in fig3c-triplets-by-contingency; the genotype effect of the two-way ANOVA under readings A and B; first 21 sessions at 100%, and at 50% the first 33 and first 35 sessions. Reproduced when at 50% the genotype effect has p < 0.05 with WT higher under every reading; the 100% result is reported and, if also p < 0.05 with WT higher, the claim is reproduced in full, otherwise partially.
fig3g-virtual-agent-performance-similar
When foraging with a virtual agent under a 50% contingency, WT and Fmr1 rats perform alike.
In this task, both WT and Fmr1 rats achieved similar levels of task performance with comparable learning rates ( Fig. 3 G )
Figure 3G and legend; Results, Development of structured spatial and temporal coordination strategies supports cooperative behavior
How it will be judged
A claim of absence. Percent of correct rewarded responses per rat and session (Methods), 8 sessions; the genotype effect of the two-way ANOVA under reading A, the test the reported df show (as an absence claim it is judged on that test alone; reading B is shown as a variant), and a Mann-Whitney U test on per-rat means. Reproduced when the genotype effect has p >= 0.05; not reproduced when it is significant. The Mann-Whitney result and the group means are reported.
fig3h-virtual-agent-efficiency-similar
With a virtual agent, WT and Fmr1 rats have similar exploration efficiency; both adopt structured transition strategies.
Notably, both genotypes independently demonstrated the capacity to adopt such structured transition strategies when interacting with the virtual agent ( Fig. 3 H )
Figure 3H and legend; Results, Development of structured spatial and temporal coordination strategies supports cooperative behavior
How it will be judged
A claim of absence. Exploration efficiency per rat and session as in fig3e-exploration-efficiency-wt-higher (optimal as reading o1, the Methods' rationale, with o2 shown as a variant); the genotype effect of the two-way ANOVA under reading A, the test the reported df show (reading B shown as a variant). Reproduced when the genotype effect has p >= 0.05; not reproduced when it is significant. The group means are reported.
figs2-jaccard-similarity
Within pairs, choice sequences are similar for both genotypes under 100% reward, but under 50% reward WT pairs have more similar choice sequences than Fmr1 pairs; in the sensory-deprived condition similarity falls in both genotypes alike.
Under the 100% reward condition, Jaccard similarity ( Fig. S2 A ; see Materials and Methods) between the choice sequences of paired rats was high for both genotypes ( Fig. S2 B,C ). In contrast, under the 50% probabilistic condition, WT pairs exhibited significantly higher similarity in their choice sequences compared with Fmr1 pairs ... similarity decreased comparably across genotypes
Results, Development of structured spatial and temporal coordination strategies supports cooperative behavior; Figure S2 (not read)
How it will be judged
The Jaccard similarity of Eq. 3, every session of one rat against every session of its partner (Methods, Cross-rat choice sequence similarity analysis). The test and the summary per pair are in the unread supplement. Open are the sets compared; readings are the sets of (a) distinct ordered transitions (from-well, to-well), (b) distinct nonoverlapping triplets, (c) distinct overlapping triplets. Open is the summary per pair; readings are the mean over (i) all entries of the matrix, (ii) the lower triangle, (iii) the diagonal (same-session pairs). The genotype test is not in the text read; the stand-in is a two-sided Mann-Whitney U test on per-pair values (5 against 5), the test the paper uses for per-rat means in Figures 3G and 5D. It is the reader's choice, and the claim should be revised once Figure S2 is read. Reproduced when, under every reading, the 50% WT similarity exceeds Fmr1 with p < 0.05; blocked as methods_underspecified when the readings disagree. The 100% and SD values are reported.
fig4b-arrival-lag-wt-shorter
During match events WT pairs arrive at the wells with shorter lags between the two rats than Fmr1 pairs, under both contingencies; in WT pairs the lag falls over 100% training.
We observed that, under the 100% reward contingency, the mean arrival lag during match events progressively decreased in WT pairs, whereas it remained consistently elevated in Fmr1 pairs ( Fig. 4 B ). This difference persisted under the 50% reward contingency, with WT pairs exhibiting significantly shorter arrival lags than Fmr1 pairs
Figure 4B and legend; Results, Development of structured spatial and temporal coordination strategies supports cooperative behavior
How it will be judged
The Methods do not define arrival lag (the legend refers to them). Per match, the absolute difference between the two rats' arrival times at the matched wells; readings of arrival are (a) the start of the rat's first poke at that well, (b) the first frame its head is within 40 pixels of the well centre (the radius of the Methods' cross-correlation section). Mean per pair and session; genotype effect of the two-way ANOVA under readings A and B; first 21 and first 35 sessions. Reproduced when, under every reading, the genotype effect has p < 0.05 with WT lower in both contingencies; blocked as methods_underspecified when readings disagree. The slope of the WT lag over the 100% sessions is reported.
fig4c-departure-lag-wt-shorter
After match events WT rats depart from the wells with shorter lags between partners than Fmr1 rats, under both contingencies.
A similar pattern emerged for mean departure lag following match events, with WT rats departing more promptly than Fmr1 rats ( Fig. 4 C )
Figure 4C and legend; Results, Development of structured spatial and temporal coordination strategies supports cooperative behavior
How it will be judged
Departure lag is not defined in the Methods. Per match, the absolute difference between the two rats' departure times from the matched wells; readings of departure are (a) the end of the rat's last poke at that well before it visits another, (b) the last frame its head is within 40 pixels of the well centre. Mean per pair and session; genotype effect of the two-way ANOVA under readings A and B; first 21 sessions at 100%, first 33 and first 35 at 50%. Reproduced when, under every reading, p < 0.05 with WT lower in both contingencies; blocked as methods_underspecified when readings disagree.
fig4d-dwell-time
Fmr1 rats spend longer at reward wells per visit than WT rats (Results), though the legend's genotype effects are not significant (p = 0.25 at 100%, p = 0.063 at 50%).
Additionally, Fmr1 rats exhibited longer dwell times at reward wells compared with WT rats ( Fig. 4 D )
Figure 4D and legend; Results, Development of structured spatial and temporal coordination strategies supports cooperative behavior
How it will be judged
Dwell time per visit, for all visits, match or not (Results); readings are (a) from the first poke to the end of the last poke at the well before the rat visits another, (b) the time the head is within 40 pixels of the well centre. Mean per rat and session; genotype effect of the two-way ANOVA under reading A, the test the reported df show (B as a variant). The Results state a difference that the legend's tests do not support. Reproduced when the Fmr1 mean exceeds the WT mean in both contingencies and the genotype effect has p >= 0.05 in both, as the legend reports; partially reproduced when the Fmr1 mean is higher in both but the genotype effect is significant in one or both (the Results sentence holds, the legend's tests do not); not reproduced when WT dwell times are longer.
fig4eg-cross-correlation-wt-higher
Cross-correlation of the two rats' well-arrival times has peaks above shuffled controls in both genotypes, but correlation is higher in WT pairs and only WT pairs show prominent peaks near zero lag (near-synchronous transitions).
Cross-correlation of arrival times at reward wells revealed significant peaks at multiple lags for both groups, exceeding shuffled controls ... Importantly, only WT rats showed prominent peaks near zero lag ... Fmr1 rats lacked this short-lag synchrony
Figure 4E to 4G and legends; Results, Development of structured spatial and temporal coordination strategies supports cooperative behavior
How it will be judged
As the Methods (Cross-correlation analysis of well arrivals) give it. Head position assigned to the nearest well when within 40 pixels of its centre, else null; downsampled to 1 Hz; an arrival where the well identity differs from the preceding time point; binary arrival series per rat; scipy.signal.correlate in full mode divided by the product of the two series' Euclidean norms; positive lag when rat 2 follows rat 1. The df (40 for lag) fix 41 lags, -20 to +20 s. Open are the well centres (the reading, the median head position during pokes at each well), the downsampling (readings, every 30th frame; the modal well in each 1 s bin), whether a change from a well to null and back to the same well is an arrival (readings, yes as the text says; no), and the shuffle (the reading, a circular shift of one rat's series, as in Figure 1H). Per pair, session and lag, the coefficient; a linear mixed-effects model coefficient ~ genotype x lag (lag categorical) with a random intercept per pair (the df fit one row per pair, session and lag). Reproduced when, in both contingencies, genotype and genotype x lag have p < 0.05 with WT higher, and the mean WT coefficient at lags -1 to +1 s exceeds the Fmr1 value; the near-zero peaks are shown per genotype. The paper gives no test for the peaks exceeding shuffled controls; observed and shuffled coefficients per genotype and lag are reported, not judged.
fig4h-peak-lag-distribution
The distribution of lags at peak cross-correlation differs between WT and Fmr1 pairs in the 50% condition but not in the 100% condition.
Correspondingly, the distribution of lags at peak correlation coefficients differed between WT and Fmr1 rats, particularly under the 50% condition ( Fig. 4 H )
Figure 4H and legend; Results, Development of structured spatial and temporal coordination strategies supports cooperative behavior
How it will be judged
The absolute lag at the maximum of each session's cross-correlation, as in fig4eg-cross-correlation-wt-higher, pooled over sessions and pairs per genotype; a two-sample KS test (scipy.stats.ks_2samp). Open is the range searched for the maximum; readings are (a) -20 to +20 s, the lags of Figure 4G; (b) all lags of the full correlation. Reproduced when, under every reading, the 50% test has p < 0.01 and the 100% test has p >= 0.05; partially reproduced when the 50% test is significant and the 100% test is too.
fig4i-peak-correlation-wt-higher
Session-wise peak cross-correlation coefficients are higher in WT than in Fmr1 pairs under both contingencies.
Session wise peak correlation coefficients for WT and Fmr1 −/y rat pairs, showing greater temporal coordination for WT rat pairs
Figure 4I and legend
How it will be judged
The maximum coefficient of each pair-session cross-correlation (fig4eg), under the two peak ranges of fig4h-peak-lag-distribution; genotype effect of the two-way ANOVA under readings A and B. Reproduced when, under every reading, both contingencies give p < 0.05 with WT higher; blocked as methods_underspecified when readings disagree.
fig5d-leading-near-half
Leadership is not fixed; on average rats of both genotypes lead about half of the match events, with no genotype difference in the mean probability of leading estimated by a state-space model.
On average, both genotypes led approximately half of the match events under both 100% and 50% reward contingencies ( Fig. 5 D , E )
Figure 5C and 5D and legends; Results, Dynamic turn-taking in the initiation of coordinated responses
How it will be judged
Per match the leader is the rat whose poke at its matched well began first (Methods, Assignment of leader and follower). Per rat, the binary leadership series over all its match events of a contingency, sessions in order, fitted with the state-space model of Smith et al. (2004) as the Methods give it (Gaussian random walk, logistic observation, EM with Newton-Raphson). b0 is not given; the reading is that of Smith et al. (2004), b0 = log(p0 / (1 - p0)) with p0 = 0.5, chance for two rats, so b0 = 0. The rat's mean of p_k over events; a two-sided Mann-Whitney U test of WT against Fmr1 rats (10 and 10) per contingency. A claim of absence for the genotype test. Reproduced when both Mann-Whitney tests have p >= 0.05 and every genotype mean is within 0.45 to 0.55; partially reproduced when the tests are not significant but a mean is outside that range; not reproduced when a test is significant. Since every pair is of one genotype and each event has one leader, the raw proportions of the two rats of a pair sum to 1, so a genotype mean near 0.5 follows by construction; the per-rat means are reported to show the spread.
fig5e-leading-difference-50
The within-pair difference in the proportion of leading events does not differ by genotype at 100%, but shows a small significant genotype effect at 50%, present at the start of training and gone later in training (Results).
However, under the 50% reward condition, a small but significant genotype-specific difference emerged, with WT and Fmr1 rats differing significantly in their likelihood of assuming the leader role at the start of training ( Fig. 5 E )
Figure 5E and legend; Results, Dynamic turn-taking in the initiation of coordinated responses
How it will be judged
Per pair and session, the difference between the two rats' proportions of match events led; the sign is not given and rat order is arbitrary, so the reading is the absolute difference, with the signed difference (first rat minus second, as the dataset orders them) as a variant. Genotype effect of the two-way ANOVA under readings A and B; first 21 sessions at 100%, first 33 and first 35 at 50%. The 100% part is a claim of absence, judged on reading A (the test the reported df show) with the absolute difference. Reproduced when the 50% genotype effect has p < 0.05 under every reading and the 100% effect has p >= 0.05; partially reproduced when the 50% effect holds under some readings only. The genotype means of the first five and the last five analysed 50% sessions are reported (the paper does not say which sessions are the start; five is a display choice and decides nothing).
figs3-leadership-history
A rat's chance of leading on a given match event depends on whether it led on the preceding events.
Furthermore, the likelihood of a rat assuming the leader role on a given trial was modulated by its leadership status over the preceding trials
Results, Dynamic turn-taking in the initiation of coordinated responses; Figure S3C and S3D (not read)
How it will be judged
The analysis and its test are in the unread supplement (Figure S3C, S3D); the Methods do not describe it, so no test can be taken from the text read. Until the supplement is read into this file, the claim is blocked as methods_underspecified. Leader per match event as in fig5d-leading-near-half. Shown, not judged, are per rat (a) P(lead | led the previous event) against P(lead | followed it), compared across rats by a Wilcoxon signed-rank test; (b) a logistic regression of leading on lead status in each of the previous 1 to 3 events. These are the reader's choices, not the paper's.
fig6-marl-partner-aware
In a multiagent reinforcement learning model of the task, partner-aware agents learn to coordinate far better than partner-unaware and random agents, reaching about 70% coordination efficiency after the switch to 50% reward.
Partner-aware agent pairs achieved markedly higher performance than both partner-unaware and random agent pairs ( Fig. 6 C )
Figure 6C and legend; Results, Differential reliance on two hierarchical strategies for successful coordination
How it will be judged
A simulation, not a test on recorded data; no code is published. As the Methods give it, tabular Q-learning, two agents on three-location grids, one agent chosen at random to act first each timestep, reward +1 when both choose the same location and it differs from the last rewarded location, epsilon decaying exponentially from 1.0 to 0.01 over 90% of episodes, 100 episodes at 100% then 100 at 50% reward. Not given are the learning rate, discount factor, timesteps per episode, number of runs and the state encoding; these must be named as readings by whoever runs it. Reproduced when partner-aware agents exceed both other types (Wilcoxon p < 0.05 over episodes) and reach 63 to 77% after the switch.
fig7b-fmr1-more-asynchronous-matches
Under 50% reward, a larger proportion of Fmr1 pairs' matches occur in asynchronous trials (rats starting at different wells) than WT pairs' matches; at 100% there is no genotype difference.
whereas Fmr1 pairs showed a higher proportion of matches occurring during asynchronous trials, particularly under the 50% reward condition ( Fig. 7 A , B )
Figure 7A and 7B and legend; Results, Differential reliance on two hierarchical strategies for successful coordination
How it will be judged
A match is synchronous when, before it, the two rats were at corresponding wells and moved together to the matched wells, and asynchronous when they started from different (noncorresponding) wells (Results). Per pair and session, the proportion of matches that are asynchronous. At 100%, a claim of absence, the genotype effect of the two-way ANOVA under reading A, the test the reported df show (B as a variant); at 50%, a linear mixed-effects model with genotype, session and their interaction and a random intercept per pair (the df F(1,280) also fit reading A, which is shown). Reproduced when the 50% genotype effect has p < 0.05 with Fmr1 higher and the 100% effect has p >= 0.05.
fig7c-trial-duration-vs-performance
Session mean trial duration is negatively correlated with cooperative performance in both genotypes (WT R = -0.51, Fmr1 R = -0.40).
Next, we found that session-level mean trial duration was negatively correlated with cooperative performance, indicating that shorter, more efficient trials were associated with higher performance ( Fig. 7 C )
Figure 7C and legend; Results, Differential reliance on two hierarchical strategies for successful coordination
How it will be judged
Pearson correlation of session mean trial duration with performance (Eq. 1), 100% and 50% sessions combined. The trial is not defined for this panel; readings are (a) the asynchronous trial of the gaze analysis, from the rats being at their initial wells to the follower's departure (Methods); (b) the interval between consecutive matches; (c) the interval between consecutive transitions of the pair. The p-values imply a different number of points per genotype; readings of the points are one per pair and session, and one per rat and session. Reproduced when, under one reading, both R are within tolerance and p < 0.05; partially reproduced when both are negative with p < 0.05 but out of tolerance.
fig7e-gaze-final-quartile
Follower-only gaze toward the leader increases from the first to the later quartiles of a trial and is greatest just before departure; mutual gaze shows no quartile effect, and there is no genotype effect.
A mixed-effects ANOVA revealed significant effects of quartile for follower-only gaze
Figure 7D and 7E and legend; Results, Differential reliance on two hierarchical strategies for successful coordination
How it will be judged
Asynchronous trials as the Methods define them (rats at distinct, nonopposing wells; the follower moves while the leader waits; the trial ends when the follower departs). The bearing angle of Eq. 11, from the neck-to-head vector and the head-to-partner-body vector, peer-directed below 40 degrees. Per trial quartile, the fraction of frames with follower-only gaze (follower toward leader, leader not toward follower) and with mutual gaze; mean per rat and quartile; a mixed ANOVA with genotype between and quartile within, rat as subject (the df fit 20 rats), Holm-corrected post hoc tests. Which sessions enter is not given; the reading is all analysed sessions of both contingencies. Reproduced when the follower-only quartile effect has p < 0.001 with quartile 4 highest, the Holm-corrected post hoc tests show quartile 1 below each of quartiles 2, 3 and 4 (p < 0.05), the mutual-gaze effect has p >= 0.05, and genotype and its interactions have p >= 0.05.
fig7g-gaze-100-late
Under 100% reward, follower-only gaze becomes a stronger predictor of matching over training, driven by Fmr1 pairs; in the late block the Fmr1-minus-WT contrast is positive for follower-only gaze (0.34) and mutual gaze (0.60), while early and mid contrasts are not significant.
Under the 100% reward condition, the models revealed that follower-only gaze became an increasingly strong predictor of successful cooperation across training ( Fig. 7 G , left; Table S1 ). Post hoc comparisons indicated that this effect was driven primarily by Fmr1 pairs
Figure 7F and 7G and legends; Results, Differential reliance on two hierarchical strategies for successful coordination; Table S1 (not read)
How it will be judged
Per asynchronous trial at 100% (fig7e), outcome match or not, predictors follower-only and mutual gaze durations normalised by trial duration (Results), genotype and session block; the model of Eq. 12, logit(P(match)) ~ (gF + gM) * Genotype * Block + (1 | animal / session), binomial, logit link (glmmTMB), and between-genotype contrasts of emtrends per block, Tukey adjusted (Methods, Post hoc analyses). Tolerances are the paper's stated spread (the ± values). Open are the blocks (readings, thirds of the 21 sessions as 7, 7 and 7; thirds by the pair's own session count) and the scaling of gF and gM ("normalized"; readings, fraction of trial duration as the Results say; that fraction z-scored). Reproduced when, under one reading, both late contrasts are positive with p < 0.05 and within tolerance and the early and mid follower-only contrasts have p >= 0.05; partially reproduced when the late contrasts are positive and significant but out of tolerance.
fig7h-gaze-50-fmr1-follower
Under 50% reward, Fmr1 pairs depend more than WT pairs on follower-only gaze in every training block (contrasts 0.38, 0.39 and 0.44), while mutual-gaze contrasts are not significant.
Fmr1 pairs remained significantly more dependent on follower-only gaze, i.e., the follower monitoring the leader throughout, compared with WT pairs ( Fig. 7 H , left)
Figure 7H and legend; Results, Differential reliance on two hierarchical strategies for successful coordination; Table S2 (not read)
How it will be judged
As fig7g-gaze-100-late, for the 50% sessions; block readings are thirds of the 35 sessions (12, 12 and 11, or 11, 12 and 12) and thirds by each pair's own session count. Reproduced when, under one reading, all three follower-only contrasts are positive with p < 0.05 and within tolerance and the three mutual-gaze contrasts have p >= 0.05; partially reproduced when the follower-only contrasts are positive and significant but out of tolerance.
fig8b-partner-history-cooperation
After a partner's cooperative choice, WT rats are more likely to cooperate under both contingencies, Fmr1 rats only under 100% reward.
For the 100% reward contingency, both WT and Fmr1 rats were more likely to cooperate after their partner had just done so in the previous trial ( Fig. 8 B ). In contrast, for the 50% reward contingency, only WT rats adjusted their behavior based on recent partner choices
Figure 8B and legend; Results, Partner cooperation history shapes cooperative behavior
How it will be judged
Follower trials as the Methods define them (the leader enters an arm first; the follower cooperates when it chooses the matching arm). For each follower trial, the partner's choice on its own most recent follower trial (Methods, Partner behavior history analysis). Per rat, P(cooperate | partner cooperated) and P(cooperate | partner did not); a Wilcoxon signed-rank test across rats per genotype and contingency (10 per genotype). Open is whether a partner's most recent trial is sought within the session only or across sessions; the reading is within the session, as Δt is a within-session interval, with across sessions as a variant. Reproduced when WT is significant (p < 0.05, higher after cooperation) at 100% and 50%, Fmr1 at 100%, and Fmr1 at 50% is not significant.
fig8d-fmr1-longer-delta-t
The time between the partner's most recent follower choice and the subject's current follower choice is longer in Fmr1 than in WT rats under both contingencies.
Across both contingencies, Fmr1 −/y rats showed significantly longer partner-referenced Δ t intervals than WT rats
Figure 8D and legend
How it will be judged
Δt per follower trial as in fig8b-partner-history-cooperation, pooled over trials of all rats per genotype, a two-sided Mann-Whitney U test per contingency. Reproduced when both tests have p < 0.0001 with the Fmr1 median longer; the U values are reported.
fig8e-glmm-partner-cooperation
In a GLMM of cooperation on each follower trial, the partner's previous cooperation has a larger effect in WT rats at 50% (beta 0.77) than in Fmr1 rats (beta 0.13), a significant genotype interaction (beta 0.64); at 100% both genotypes show an effect (WT 0.36, Fmr1 0.26).
However, under the 50% contingency only WT rats adjusted their choices based on recent partner behavior, while Fmr1 rats did not ( Fig. 8 E )
Figure 8E and legend; Results, Partner cooperation history shapes cooperative behavior
How it will be judged
Per contingency, a binomial GLMM (logit) of cooperation on each follower trial with fixed effects partner's previous choice (binary), Δt (continuous) and their interaction, each interacting with genotype, plus covariates for the subject's own recent choice history, random intercepts for rat and session (Methods, Generalized linear mixed-effects modeling; pymer4 0.8 / lme4). Open are the own-history covariates (readings, the subject's own previous follower choice; its previous two), the scaling of Δt (readings, seconds; z-scored), and how the per-genotype betas are read (readings, simple effects with WT as the reference level and with Fmr1 as reference). The legend gives Fmr1 at 50% beta 0.13 with p < 0.01 while the Results say Fmr1 rats did not adjust; the finding tested is the 50% genotype interaction and the WT effects. Reproduced when, under one reading, the 50% interaction is positive with p < 0.001 (WT larger), the WT effects are positive with p < 0.001, and the betas are within tolerance; partially reproduced when the signs and significance hold but a beta is out of tolerance.

Pipeline

Paper: done (1 of 1)
Claims: done
Claims check: done (ready)
Attempt: todo
Questions: none
Verified: none