HHMI Janelia launches two big bets: a whole fish brain and AI-in-the-Loop biology
HHMI Janelia is launching a decade-long program: a transparent adult fish is meant to help link brain function to behavior, while AI will suggest the next experiments
On June 15, Janelia Research Campus announced two connected "big bets." The institute wants to build a mechanistic explanation of how a vertebrate brain generates behavior, while also testing AI-in-the-Loop, an approach in which AI helps plan and launch experiments as new data comes in.
Janelia is part of Howard Hughes Medical Institute, a private biomedical institute in the U.S. This campus was built to tackle projects that are hard for a typical university lab to sustain over decades: new microscopes, sensors, brain maps, genetic tools, and large open datasets.
Now Janelia is bringing that model to the tiny fish Danionella. It is only slightly larger than a grain of rice and remains transparent in adulthood. In the standard model, zebrafish larvae, that transparency exists only early in life, when behavior is still limited. Adult Danionella can forage, learn, interact with other fish, and engage in mating behavior, while allowing researchers to see most of the brain of a living vertebrate.
Brain mapping has already reached an engineering bottleneck: NeuroPEEM обещает ускорить съёмку ткани, while Eon проверяет whether neural activity can be predicted from wiring alone. Janelia is adding the live layer: the brain has to be observed together with the body, behavior, and environment while the animal is acting.
The task sounds simple only on the surface: connect molecules, neurons, circuits, physiology, and action. When an animal turns toward food, avoids a threat, or changes its behavior after experience, the brain is integrating vision, smells, internal state, memory, hormones, body movement, and the current goal. Janelia wants to assemble that into a mechanistic picture in which it is possible to point to the specific cells and circuits that make a given action possible.
The second half of the program is called AI-in-the-Loop. In HHMI's version, AI reads the data, builds hypotheses, proposes the next experiment, simulates options in advance, spots unexpected patterns, and updates its predictions after new measurements. In the mature version, part of the experimental cycle will run as a chain of "the model proposed it → the lab tested it → the model changed the plan."
There is a practical reason for this idea. Modern biology is drowning in measurements: microscopy, neural activity, behavior, genetic lines, molecular markers. A human can come up with a strong question, but manually working through thousands of condition combinations quickly breaks the pace of research. Autonomous science begins where the machine chooses the next experiment that most reduces uncertainty.
HHMI has already framed this shift more broadly: the AI@HHMI program is budgeted at $500 mln over 10 years. It includes projects on protein design for new fluorescent tags, AI analysis of cryo-ET, metabolic sensors, RNA structures, robotic work with drosophila, and zebrafish models. Janelia's new bet looks like the place where those pieces are meant to come together into a single research process.
The constraints are visible in the release itself. For Danionella, genetic tools, imaging systems, behavioral assays, and models still need to be built. In the first years, Janelia will refine the methods in drosophila and zebrafish larvae. The institute also writes that scientists still do not know whether AI can deliver reliable mechanistic understanding of biology at this depth.
Janelia's bet comes down to how quickly biology can turn measurements into working models. If AI-in-the-Loop learns to shorten the cycle from hypothesis to experiment from months to days, this kind of infrastructure will spread beyond neuroscience. Any complex biology requires the same path: observe the system, choose an intervention, test the response, update the model, and choose the next experiment again. That cycle is what turns elegant correlations into tests of causality.