A complete pipeline for analysing directional data &ndash; headings and trajectories on the unit disk. Supply a table of angles directly, or reconstruct headings from movement trajectories imported from tracking tools ('EthoVision', 'DeepLabCut', 'SLEAP', 'TRex', 'ANY-maze', 'TrackMate', 'idtracker.ai', and others) via a unified dialect-based loader system, including multi-bodypart pose estimation with likelihood-weighted centroid. Derives per-trial heading directions using multiple rules (ring-crossing, distal point, pose body-axis, velocity-based, and more). Computes circular statistics including mean direction, resultant length, von Mises and wrapped Cauchy parametric fits, circular correlation (circular-linear and circular-circular), circular-linear regression (Fisher and Lee (1992) &lt;<a href="https://doi.org/10.2307%2F2532334" target="_top">doi:10.2307/2532334</a>&gt;), model selection among orientation models (Fitak and Johnsen (2017) &lt;<a href="https://doi.org/10.1242%2Fjeb.167056" target="_top">doi:10.1242/jeb.167056</a>&gt;), and uniformity and multi-sample hypothesis tests (Rayleigh, Watson-Williams, Hermans-Rasson (Landler, Ruxton and Malkemper (2019) &lt;<a href="https://doi.org/10.1186%2Fs12898-019-0246-8" target="_top">doi:10.1186/s12898-019-0246-8</a>&gt;), and others) with multiple-comparison correction. Path tortuosity follows Benhamou (2004) &lt;<a href="https://doi.org/10.1016%2Fj.jtbi.2004.03.016" target="_top">doi:10.1016/j.jtbi.2004.03.016</a>&gt;. Renders publication-quality 'ggplot2' radial plots with overlaid rose diagrams, parametric density curves, non-parametric kernel density estimates, and circular boxplots (Buttarazzi, Pandolfo and Porzio (2018) &lt;<a href="https://doi.org/10.1111%2Fbiom.12889" target="_top">doi:10.1111/biom.12889</a>&gt;). Includes a browser-based 'shiny' graphical interface requiring no R coding via launch_app().

A complete pipeline for analysing directional data &ndash; headings and trajectories on the unit disk. Supply a table of angles directly, or reconstruct headings from movement trajectories imported from tracking tools ('EthoVision', 'DeepLabCut', 'SLEAP', 'TRex', 'ANY-maze', 'TrackMate', 'idtracker.ai', and others) via a unified dialect-based loader system, including multi-bodypart pose estimation with likelihood-weighted centroid. Derives per-trial heading directions using multiple rules (ring-crossing, distal point, pose body-axis, velocity-based, and more). Computes circular statistics including mean direction, resultant length, von Mises and wrapped Cauchy parametric fits, circular correlation (circular-linear and circular-circular), circular-linear regression (Fisher and Lee (1992) &lt;<a href="https://doi.org/10.2307%2F2532334" target="_top">doi:10.2307/2532334</a>&gt;), model selection among orientation models (Fitak and Johnsen (2017) &lt;<a href="https://doi.org/10.1242%2Fjeb.167056" target="_top">doi:10.1242/jeb.167056</a>&gt;), and uniformity and multi-sample hypothesis tests (Rayleigh, Watson-Williams, Hermans-Rasson (Landler, Ruxton and Malkemper (2019) &lt;<a href="https://doi.org/10.1186%2Fs12898-019-0246-8" target="_top">doi:10.1186/s12898-019-0246-8</a>&gt;), and others) with multiple-comparison correction. Path tortuosity follows Benhamou (2004) &lt;<a href="https://doi.org/10.1016%2Fj.jtbi.2004.03.016" target="_top">doi:10.1016/j.jtbi.2004.03.016</a>&gt;. Renders publication-quality 'ggplot2' radial plots with overlaid rose diagrams, parametric density curves, non-parametric kernel density estimates, and circular boxplots (Buttarazzi, Pandolfo and Porzio (2018) &lt;<a href="https://doi.org/10.1111%2Fbiom.12889" target="_top">doi:10.1111/biom.12889</a>&gt;). Includes a browser-based 'shiny' graphical interface requiring no R coding via launch_app().

DeepLabCut (DLC), a machine learning-based pose estimation tool, has dramatically increased the throughput of behavioral tracking. Yet its implication remains limited by lack of validation data from diverse scenarios and users' unfamiliarity with the machine learning workflow. Here, we provide a practical primer on DLC terminology and workflow and performance evaluation of DLC in a novel scenario where the target is much smaller than those in previous validation data. We assessed model performance in a total of 36 model configurations across three training proportions, six training iterations, and two recording resolutions. Performance was evaluated on training convergence, prediction error, prediction certainty, and generalizability. Additionally, we selected an optimal model based on a balance between model performance and training time and applied it to 20 recordings at each resolution. In the test frames, 15 and 6 out of 18 model configurations met all performance criteria under high- and low-resolution, respectively. Of these, DLC was able to achieve low prediction error, high prediction confidence, and low train-test loss. Prediction error and confidence varied with training proportion, iteration, and resolution. Inference using the optimal models achieved similar high confidence predictions, with low-likelihood predictions predominantly associated with visual obstruction, low light, and reduced contrast between the subject and the background. Notably, the optimal models for high- and low-resolutions performed similarly despite a multi-fold difference in the cost of the recording devices, providing validation data for DLC in low-budget research, teaching and outreach scenarios.

DeepLabCut (DLC), a machine learning-based pose estimation tool, has dramatically increased the throughput of behavioral tracking. Yet its implication remains limited by lack of validation data from diverse scenarios and users' unfamiliarity with the machine learning workflow. Here, we provide a practical primer on DLC terminology and workflow and performance evaluation of DLC in a novel scenario where the target is much smaller than those in previous validation data. We assessed model performance in a total of 36 model configurations across three training proportions, six training iterations, and two recording resolutions. Performance was evaluated on training convergence, prediction error, prediction certainty, and generalizability. Additionally, we selected an optimal model based on a balance between model performance and training time and applied it to 20 recordings at each resolution. In the test frames, 15 and 6 out of 18 model configurations met all performance criteria under high- and low-resolution, respectively. Of these, DLC was able to achieve low prediction error, high prediction confidence, and low train-test loss. Prediction error and confidence varied with training proportion, iteration, and resolution. Inference using the optimal models achieved similar high confidence predictions, with low-likelihood predictions predominantly associated with visual obstruction, low light, and reduced contrast between the subject and the background. Notably, the optimal models for high- and low-resolutions performed similarly despite a multi-fold difference in the cost of the recording devices, providing validation data for DLC in low-budget research, teaching and outreach scenarios.

Behavioural experiments often hinge on moments when animals crowd together. In these situations, existing pose-estimation methods usually fail because detected crops include neighbouring animals, and their body parts are easily confused between individuals. We solved it with DeepLabSAM, a pipeline that segments and tracks every animal from a single text prompt using the SAM 3 foundation model, followed by a mask-gated pretrained pose estimation network from DeepLabCut. Across recordings of multiple interacting mice, the proposed method substantially reduces misassigned keypoints under heavy occlusion. Furthermore, we used DeepLabSAM to automatically construct the Multi-Task Mouse Behaviour (MTMB) dataset, comprising 14,300 frames spanning nine behavioural assays, with each instance annotated for bounding box, mask, identity, and pose. SuperAnimal pretrained pose models transferred unpredictably across this dataset. Conversely, MouseLite, our RF-DETR pose-estimation model trained on the MTMB dataset, matched their in-distribution accuracy, generalized better to unseen assays, and achieved real-time (faster-than-acquisition) inference. Combined, DeepLabSAM, MTMB dataset, and MouseLite provide a scalable framework for robust, cross-assay behavioural pose tracking of freely interacting mice.

Behavioural experiments often hinge on moments when animals crowd together. In these situations, existing pose-estimation methods usually fail because detected crops include neighbouring animals, and their body parts are easily confused between individuals. We solved it with DeepLabSAM, a pipeline that segments and tracks every animal from a single text prompt using the SAM 3 foundation model, followed by a mask-gated pretrained pose estimation network from DeepLabCut. Across recordings of multiple interacting mice, the proposed method substantially reduces misassigned keypoints under heavy occlusion. Furthermore, we used DeepLabSAM to automatically construct the Multi-Task Mouse Behaviour (MTMB) dataset, comprising 14,300 frames spanning nine behavioural assays, with each instance annotated for bounding box, mask, identity, and pose. SuperAnimal pretrained pose models transferred unpredictably across this dataset. Conversely, MouseLite, our RF-DETR pose-estimation model trained on the MTMB dataset, matched their in-distribution accuracy, generalized better to unseen assays, and achieved real-time (faster-than-acquisition) inference. Combined, DeepLabSAM, MTMB dataset, and MouseLite provide a scalable framework for robust, cross-assay behavioural pose tracking of freely interacting mice.

Zebrafish are an important model system in behavioral neuroscience due to their rapid development and suite of distinct, innate behaviors. In this study, we tracked the behavior of 6 d postfertilization larval zebrafish (of indeterminate sex, as larvae at this developmental stage are sexually undifferentiated). Quantifying many of these larval behaviors requires detailed tracking of eye and tail kinematics, which in turn demands imaging at high spatial and temporal resolution, ideally using semi- or fully automated tracking methods for throughput efficiency. However, creating and validating accurate tracking models is time-consuming and labor-intensive, with many research groups duplicating efforts on similar images. With the goal of developing a useful community resource, we trained pose estimation models using a diverse array of video parameters and a 15-keypoint pose model. We deliver an annotated dataset of free-swimming and head-embedded behavioral videos of larval zebrafish, along with four pose estimation networks from DeepLabCut and SLEAP (two variants of each). We also evaluated model performance across varying imaging conditions to guide users in optimizing their imaging setups. This resource will allow other researchers to skip the tedious and laborious training steps for setting up behavioral analyses, guide model selection for specific research needs, and provide ground-truth data for benchmarking new tracking methods.

This study examined how prenatal CBD and {Delta}9-tetrahydrocannabinol (THC) exposure affects maternal caregiving in C57BL/6J mice. We developed the Machine-Automated Scoring of the Pup Retrieval Test (MAS-PRT) to overcome limitations of manual behavioral scoring. MAS-PRT integrates Multi-Animal DeepLabCut for dam and pup tracking with Detectron2 for dynamic nest reconstruction. Validated against 170 manually annotated retrieval trials, the pipeline showed high concordance with manual measurements and enabled reproducible extraction of encounter latency, retrieval latency, and locomotor trajectories. Prenatal exposure did not impair general nest-building or overall home-cage maternal care. However, pups from both CBD and THC groups showed reduced body weight at postnatal day 5. Cox proportional hazards modeling revealed divergent effects by compound: CBD-exposed dams exhibited a weight-dependent increase in probability of encountering and retrieving lighter pups, independent of pup sex. Spatial tracking further showed that dams traversed significantly shorter total trajectories when retrieving female progeny exposed to either compound. Longitudinal trial-by-trial analysis indicated intact task acquisition in controls and CBD dams, whereas THC dams displayed a flattened learning curve driven by lower retrieval latencies on initial trials. Together, these findings indicate that prenatal cannabinoid exposure does not produce generalized disruption of maternal care but instead induces compound-specific alterations in maternal reactivity, retrieval kinematics, and learning dynamics.

This study examined how prenatal CBD and {Delta}9-tetrahydrocannabinol (THC) exposure affects maternal caregiving in C57BL/6J mice. We developed the Machine-Automated Scoring of the Pup Retrieval Test (MAS-PRT) to overcome limitations of manual behavioral scoring. MAS-PRT integrates Multi-Animal DeepLabCut for dam and pup tracking with Detectron2 for dynamic nest reconstruction. Validated against 170 manually annotated retrieval trials, the pipeline showed high concordance with manual measurements and enabled reproducible extraction of encounter latency, retrieval latency, and locomotor trajectories. Prenatal exposure did not impair general nest-building or overall home-cage maternal care. However, pups from both CBD and THC groups showed reduced body weight at postnatal day 5. Cox proportional hazards modeling revealed divergent effects by compound: CBD-exposed dams exhibited a weight-dependent increase in probability of encountering and retrieving lighter pups, independent of pup sex. Spatial tracking further showed that dams traversed significantly shorter total trajectories when retrieving female progeny exposed to either compound. Longitudinal trial-by-trial analysis indicated intact task acquisition in controls and CBD dams, whereas THC dams displayed a flattened learning curve driven by lower retrieval latencies on initial trials. Together, these findings indicate that prenatal cannabinoid exposure does not produce generalized disruption of maternal care but instead induces compound-specific alterations in maternal reactivity, retrieval kinematics, and learning dynamics.