Social behavior is a complex and dynamic process shaped by movement, coordination and physical touch. A new study in Cell introduces s-DANNCE, a machine-learning system that can map the fine-scale movements of freely interacting rats in three dimensions.

Simons Fellows-to-Faculty grantee and lead author Ugne Klibaite says that this method can “analyze amounts of data that would take humans years and years to scroll through.” Co-author and fellow Autism Rat Consortium member Bence P. Ölveczky adds that with this new tool, “we can replace the subjective human observer with a very rigorous and reproducible method for behavioral quantification and identification of particular gestures or even interaction motifs.”

By applying s-DANNCE to seven genetic rat models of autism, researchers have uncovered distinct social phenotypes, offering new insights into the diversity of autism-related behaviors.

Mapping the Social Landscape

Traditional approaches to studying rodent social behavior rely on coarse measurements made by human observers, such as how much time the animals spend in close proximity or how often they groom or touch each other. While useful, these methods overlook the fine-grained body movements that structure social interactions.

To address this, Klibaite’s team developed s-DANNCE, an advanced 3D tracking system that can monitor animals’ full-body postures, coordinated movements and patterns of physical contact. Unlike previous methods, which struggle to track animals when they gather closely and partially block each other, s-DANNCE uses graph neural networks and semi-supervised learning to accurately assign body parts—even in crowded environments.

Using this approach, the team mapped over 140 million rat and mouse 3D pose samples in lone and social contexts, identifying distinct behavioral motifs that define different types of social engagement. Furthermore, researchers also validated the use of this tool in mice, extending its potential applications in future research.

A New Lens on Autism

One application of s-DANNCE with potential for considerable impact is in autism research. The team applied their system to rat lines with mutations in seven autism risk genes: Arid1b, Chd8, Cntnap2, Grin2b, Fmr1, Nrxn1 and Scn2a. Datasets from these lines are shared online for researchers to access.

Like in humans, where mutations in these genes produce an array of conditions, and do not always result in an autism diagnosis, autism-related mutations in rats do not produce a single behavioral phenotype. Instead, each model showed differences in social behavior, ranging from reduced social engagement and touch avoidance to increased, uncoordinated social interactions.

For example, rats with mutations in Arid1b and Grin2B displayed less physical contact and fewer close interactions, whereas those with mutations in Nrxn1 and Scn2a were more touch-seeking and interactive than controls. These findings mirror the variability observed in individuals on the autism spectrum and with related neurodevelopmental conditions, where some people experience social withdrawal while others exhibit heightened social interest.

Uncovering Drug Effects on Social Behavior

To further validate their approach, the researchers tested whether amphetamine, a drug known to affect social behavior, altered interaction patterns in rats. Their analysis revealed unexpected shifts in social coordination, with drug-treated animals showing disruptions in behavioral synchrony and changes in the distribution of

These results suggest that s-DANNCE could serve as a powerful tool for testing the effects of drugs on social behavior, offering a new way to evaluate potential interventions.

The researchers next plan to integrate neural recordings and vocalization tracking into s-DANNCE to gain a deeper understanding of how the brain controls social interactions. They have also made the tool freely available to the larger research community.

By enabling high-resolution mapping of social behavior, this AI-driven approach offers new possibilities for studying autism and related neurodevelopmental disorders as well as identifying more precise interventions.