Fine action features from body-point tracking, 3D reconstruction, behavior decomposition, and clustering — combined with prediction models trained on autism modeling data for single-sample classification and multi-modal visual reports. Load project → predict → view report in three steps.
// prediction stream
[load] Explorer.ba3e · 24 samples · 40 action classes
40
Action classes
3 steps
Workflow
10+
Report modules
Research Background
Autism spectrum disorder (ASD) animal models often show social deficits, repetitive stereotypy, and abnormal activity patterns — yet conventional tests (open field, three-chamber social, etc.) cover only limited dimensions and miss subtle spontaneous phenotypes. High-dimensional 3D skeleton tracking and unsupervised action clustering decompose continuous behavior into dozens of fine actions for richer cross-model and cross-group profiles.
The Autism Model Automated Behavior Assessment System builds on the BehaviorAtlas pipeline with prediction models trained on autism modeling cohorts: import a pre-processed project (Explorer.ba3e), predict per sample in one click, and the report panel shows multi-view video, 3D skeleton, action spectra, low-dimensional clustering, KNN classification, transition chord diagrams, kinematics, and 10+ visualization modules — for Shank3, Scn2a, and pharmacological intervention phenotyping.
Unsupervised clustering splits spontaneous behavior into 40 fine action segments — non-locomotor and locomotor features.
Trained on autism modeling data — per-sample class prediction with confidence scores.
10+ linked charts: action spectra, UMAP clustering, transition matrices, word clouds, and kinematics distributions.
Capture multi-channel dual-tier multi-view acquisition · Explorer project prediction and multi-modal reports — full capture-to-analysis pipeline.
Multi-channel · dual-tier multi-view sync capture
Multi-channel capture software for autism model mouse spontaneous behavior. Multi-view cameras on upper and lower cage tiers — synchronized multi-stream recording with millisecond alignment; sample ID, group metadata, and long-duration unattended acquisition for high-quality raw data for 3D tracking and action clustering.
Project import · single-sample prediction · multi-modal reports
Built for autism model behavioral big-data analysis. Load projects via Explorer.ba3e config with completed 3D tracking and action clustering; sample list shows predictions, model panel runs inference, and the report area covers multi-view video, 3D skeleton, action heatmaps, low-dimensional clustering, KNN classification, action transitions, and kinematics.
Capture multi-channel dual-tier acquisition and Explorer project prediction and reporting — full pipeline from data capture to visual analysis.
3Dskeleton (3D skeleton csv/c3d), BeAOutputs (h5 skeleton/kinematics/action labels), Figures (clustering & skeleton plots), VideoSeg (40 action-class clips), and Explorer.ba3e config.
Menu File → Import Data, select Explorer.ba3e config — auto-loads sample list and pre-processed results.
Select sample and model, click Predict — sample list updates with predicted class instantly.
Main report panel: multi-view video & 3D skeleton, action density & spectra, UMAP clustering, KNN classification, and group-level dimensionality reduction.
Multi-view raw video with 3D skeleton overlay — selectable body points and motion trajectories synced to playback.
Export analysis charts, action-segment videos, and structured parameter tables for group statistics, publication figures, and downstream analysis.
Load project → single-sample prediction → view report
After launch, click File in the menu bar → Import Data, navigate to the project folder, and open Explorer.ba3e config.
Select a sample in the list, choose a model in the prediction panel, and click Predict — success when results appear in the sample list.
Select a sample with prediction results — the report panel loads the full analysis with 10+ chart and video-linked modules.
After tracking, reconstruction, decomposition, and clustering — prediction models drive multi-dimensional visual reports.
Project metadata, sample grouping, and prediction model configuration.
Multi-view raw video with synchronized 3D skeleton — selectable body-point motion trajectories.
Representative autism action density and sample action spectrum — scrubber linked to video for current action name.
UMAP1/UMAP2 for non-locomotor features; Norm-velocity axis for locomotion — all sample actions in low-dimensional space.
KNN on combined kinematics and action data; 3D group-level dimensionality reduction of action fractions and kinematics.
Transition chord diagrams (arc width = frequency) and action-name transition frequency heatmaps for sample and F/S/W groups.
Action fraction word clouds per sample and group — larger text = higher action proportion.
Velocity trajectory, position distribution, and body overlay heatmaps; speed/intensity/body-size distributions and body-angle polar histograms.
Autism gene model phenotyping, pharmacological intervention assessment, and psychiatric translational research.
High-dimensional spontaneous behavior phenotyping and predictive classification for Shank3, Scn2a, and other ASD gene models.
Pre/post intervention action spectra, transition patterns, and kinematics for objective efficacy assessment.
Group-level action distribution and kinematics comparison — sex-difference phenotypes in ASD models.
Bridge from animal spontaneous behavior to human symptom dimensions — objective metrics under the RDoC framework.
Contact us to learn more about high-dimensional autism model phenotyping and predictive behavioral analysis.