Building the Computational Mind for the “Swiss Army Knife” of Microscopes
Lawrence Berkeley National Laboratory researchers unveiled MOSAIC, a reconfigurable microscope integrating over ten imaging techniques with adaptive optics to track biological processes across scales in real time.
Video
Video available
Biology operates across multiple scales, from molecular interactions to tissue-level changes, yet conventional microscopes force researchers to analyze samples separately, often damaging biological context and delaying results. MOSAIC, developed by Berkeley Lab and collaborators, consolidates more than ten imaging methods into a single instrument, enabling simultaneous observation of processes across scales within the same sample. The microscope uses adaptive optics—originally developed for astronomy—to correct distortions caused by living tissue, improving signal and resolution without invasive adjustments.
MOSAIC generates up to four terabytes of data per hour, far exceeding the capacity of conventional processing workflows or manual analysis. To address this, Berkeley Lab’s computational tools, including the PetaKit5D software toolkit and allocations on the Perlmutter supercomputer at NERSC, process data in real time and reduce costs by over an order of magnitude compared to prior methods. The instrument’s ability to image with minimal invasiveness has already enabled experiments such as tracking single molecules in living cells and mapping neuronal architecture in Alzheimer’s-affected brain tissue.
The microscope’s adaptive optics correction revealed roughly 2.5 times more detectable neural calcium events in live mouse brains than imaging without it, suggesting conventional microscopy may undercount brain activity. MOSAIC also powered a 2025 study that imaged two adult mouse olfactory bulbs at nanoscale resolution, generating approximately a petabyte of data over two weeks—a volume that took two years to analyze, highlighting the gap between data acquisition and processing capabilities.
Berkeley Lab is now using MOSAIC to capture five-dimensional data—spanning three spatial dimensions, time, and molecular identity—intended to train advanced AI models for biological research. The lab envisions a future where connected automated systems could enable self-driving laboratories, accelerating discovery by integrating imaging, sample handling, and experimental context to identify meaningful observations and guide subsequent experiments.