OPhELIA: New AI maps brain circuits using 95% fewer trials

A team of researchers has developed OPhELIA (Optimal Photostimulation sElection for Iterative Activity maps), a Bayesian framework that slashes the number of experiments needed to map neural circuits. Traditional all-optical two-photon holographic optogenetics can causally connect neurons but is hindered by combinatorial complexity, tissue heating, and photodamage. OPhELIA tackles this by combining a Beta-Bernoulli model for connectivity inference with an ambiguity-based acquisition heuristic that chooses the most informative photostimulation targets. In silico simulations and in vivo experiments in larval zebrafish, OPhELIA with active learning recovered near-exhaustive functional connectomes. Its compressed sensing variant achieved the same goal using only 5% of the total possible trials, mapping neural interactions with dramatically reduced experimental time and biological damage.

This sample-efficient approach could transform causal connectomics, enabling researchers to map brain-wide circuits in a fraction of the time. By learning priors from pre-stimulation neural activity, OPhELIA further reduces uncertainty without extra trials. The method opens the door to systematic brain mapping in organisms where exhaustive connectivity experiments are currently prohibitive. Potential applications include understanding neural computation, brain disorders, and validating computational models of cognition. The code and data are publicly available, allowing the neuroscience community to adopt and extend this technique for faster, less invasive circuit mapping.