An international team of researchers has built the largest three-dimensional database of ant anatomy ever assembled, scanning roughly 2,000 specimens in a single week by combining synchrotron imaging, robotics, and artificial intelligence. The effort, described in a Nature Methods article, covers hundreds of species and genera, offering scientists a new way to study how body structure relates to colony behavior and evolutionary history across one of the most diverse insect families on Earth.
A Week of Scanning, Thousands of Specimens
The project, called Antscan, is a joint initiative between the Karlsruhe Institute of Technology in Germany and the Okinawa Institute of Science and Technology in Onna, Japan. Researchers used synchrotron micro-CT, a technique that fires intense X-ray beams through tiny specimens to produce high-resolution three-dimensional images of internal and external anatomy. The IMAGE beamline at the KIT Light Source and the TopoTomo station at the former ANKA synchrotron facility were among the instruments used during the scanning campaigns.
What sets this work apart from earlier morphological studies is raw speed. Traditional micro-CT scanning of museum specimens can take hours per individual ant. The Antscan team, by contrast, scanned roughly 2,000 specimens in a week, according to the University of Maryland Department of Entomology. That throughput was possible because the workflow combined robotic specimen handling with automated image processing, drastically cutting the human labor that normally bottlenecks large-scale imaging projects.
Specimens came from existing museum and research collections, meaning the project could tap into decades of fieldwork without removing more ants from the wild. Each individual was mounted in a standardized holder, loaded into an automated carousel, and passed through the X-ray beam in rapid succession. The resulting image stacks capture not just overall body size, but also fine structures such as mandible teeth, antenna segments, and the intricate joints that allow ants to climb, dig, and fight.
How AI Turns Raw Scans into Usable Data
Collecting thousands of three-dimensional image stacks is only half the challenge. Each scan must be segmented, meaning individual body parts like the head capsule, mandibles, legs, and gaster need to be digitally separated so researchers can measure and compare them. Doing that by hand for thousands of ants would take years. The Antscan team instead relied on deep-learning segmentation through Biomedisa software, an open-source online platform originally developed for biomedical image analysis and published in Nature Communications.
Biomedisa allowed the team to train neural networks on a subset of manually labeled scans and then apply those models across the full library. Once trained, the algorithm could identify and outline key structures in new scans with minimal user input, drastically accelerating the conversion from raw X-ray slices to clean, labeled three-dimensional meshes. Researchers could then export standard measurements such as head width, leg length, and total cuticle volume, or derive more complex shape descriptors for evolutionary analyses.
The finished scans and associated metadata are stored in KIT’s open-access RADAR repository, making the entire dataset available for reuse by other research groups. Each record links a specimen’s 3D volume with information such as species name, collection locality, and caste, creating a bridge between digital morphology and traditional natural history data. Because the pipeline is documented and reproducible, other teams can extend the database with compatible scans from additional collections.
Pairing Shape with Genome
The database does not exist in isolation. A parallel effort under the Global Ant Genomics Alliance, known as GAGA, has been assembling high-quality genome sequences for many of the same ant lineages. Those genomic resources are publicly archived under the GAGA BioProject at the National Center for Biotechnology Information. A related peer-reviewed paper on ant adaptive radiation, published in Cell, provides the genomic backbone that overlaps with many of Antscan’s scanned taxa.
This convergence of phenomic and genomic data is what gives the project its analytical power. Researchers can now ask whether specific genetic changes correlate with measurable shifts in body proportion, armor thickness, or limb geometry, all without destroying irreplaceable museum specimens. The ability to pair a three-dimensional model of an ant’s exoskeleton with its sequenced genome opens a direct line between form, function, and evolutionary history that was previously difficult to study at scale.
For example, scientists can examine whether genes involved in cuticle formation or muscle development show repeated changes in lineages that independently evolved similar foraging strategies. They can also test how shifts in social organization, such as the emergence of highly specialized worker castes, map onto both genomic variation and subtle shape differences that only become obvious in standardized 3D measurements.
Body Size, Colony Size, and an Evolutionary Trade-Off
Early analysis of the Antscan dataset has already produced a notable finding. Researchers found a strong negative correlation between cuticle volume and colony size, according to the UMD computer science department. In plain terms, ant species that form very large colonies tend to have smaller individual workers, suggesting that evolution has pushed these lineages to prioritize the quantity of workers over the size of each one.
That pattern is not entirely surprising to myrmecologists, who have long suspected such a trade-off. But quantifying it across hundreds of species with standardized three-dimensional measurements is new. Previous studies relied on two-dimensional photographs or manual caliper measurements of a handful of specimens, making broad comparisons unreliable. The Antscan library, because it uses the same imaging protocol and segmentation pipeline for every specimen, controls for measurement inconsistency in a way that smaller studies cannot.
The finding also carries ecological weight. Colony size influences how ant species forage, defend territory, and respond to habitat disturbance. If the body-size trade-off is strong and predictable, it could help ecologists model how ant communities will reorganize as climate shifts alter food availability and nesting conditions. With roughly 14,000 ant species known worldwide, according to a joint release from KIT and OIST, even partial coverage of that diversity through standardized scans represents a significant step toward testable predictions.
What Most Coverage Misses
Much of the early attention on Antscan has focused on its sheer scale and the eye-catching visualizations that come from animating detailed 3D models of ants. But several aspects of the work are easy to overlook. One is that the project doubles as a testbed for high-throughput imaging in natural history collections more broadly. The same combination of synchrotron micro-CT, robotics, and AI-based segmentation could be applied to other small organisms, from beetles to marine plankton, turning static museum drawers into dynamic digital archives.
Another underappreciated element is accessibility. The Nature Methods study is available through a publisher access portal that links directly to the technical description of the Antscan pipeline, including details on imaging parameters, reconstruction algorithms, and segmentation training. By documenting the workflow in this way and pairing it with open repositories for both images and code, the team lowers the barrier for other labs that may not have extensive imaging expertise but do have valuable specimens to contribute.
The project also highlights a cultural shift in how systematists and evolutionary biologists work with data. Instead of single-author monographs based on a few dozen specimens, Antscan embodies large, collaborative teams pooling engineering, computer science, and taxonomy. That model can be harder to coordinate, but it enables questions, such as global patterns linking morphology, social behavior, and genome evolution, that would be impossible for any one lab to tackle alone.
Looking ahead, the researchers involved in Antscan and GAGA suggest that the current dataset is a starting point rather than an endpoint. As more species are scanned and sequenced, and as AI tools improve, the resolution of questions that can be asked will sharpen. Future work may move beyond simple measurements of size and volume to analyses of mechanical performance, such as bite force inferred from mandible shape, or locomotion efficiency derived from leg geometry. In doing so, the project could turn ants (from leafcutters to army ants) into some of the best-understood animals on the planet in terms of the links between genes, bodies, and societies.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
