MOSAIC is a reconfigurable multimodal optical microscope that consolidates more than ten imaging techniques into one compact instrument, enabling rapid switching between modes to reduce sample perturbation and preserve biological context.

Adaptive optics and a modular optical switching system let MOSAIC switch between imaging modes in as little as two to five seconds, reducing blur from living tissue and enabling high-throughput data collection.

The platform can cycle through roughly a dozen imaging modes in mere seconds, using a shared set of lasers, mirrors, cameras, and computational hardware to deliver sharp, high-density data with sample-induced blur corrected.

Berkeley Lab’s LDRD-supported efforts, including the PetaKit5D software, enable real-time handling of terabyte-scale data and aim to train AI models that reason over five-dimensional data—three spatial dimensions, time, and molecular identity.

MOSAIC generates up to four terabytes of data per hour, driving the need for advanced computational workflows, real-time processing, and large-scale data analysis to extract meaningful biological insights.

The broader vision envisions self-driving biological laboratories where automated microscopes, sample handling, and perturbation systems accelerate discoveries by integrating experimental context with AI-driven analysis.

Future AI capabilities under development include vision-language models that reason over biology to guide experiments and potentially enable autonomous, self-driving laboratories.

The team emphasizes that the bottleneck has shifted from data acquisition to turning dense, high-dimensional observations into biological understanding, relying on computing, AI, and automated instrumentation.

A related achievement with VIPS used MOSAIC and computational tools to image two mouse olfactory bulbs at nanoscale, generating roughly a petabyte of data in about two weeks and highlighting the processing gap between acquisition and interpretation.

VIPS demonstrated imaging two adult mouse olfactory bulbs at nanoscale, producing about a petabyte of data in two weeks, underscoring the need for advanced analytics to close the processing gap.

The VIPS project showcased Volumetric Imaging via Photochemical Sectioning, imaging two complete adult mouse olfactory bulbs at nanoscale and illustrating the data handling challenges involved.

MOSAIC has been demonstrated across diverse biological contexts, including tracking single molecules in living cells, mapping neuronal architecture in human brain tissue from an Alzheimer’s patient, and imaging neural activity in live mice with adaptive optics increasing detectable calcium events by roughly 2.5 times.