Advanced microscopes like MOSAIC produce terabytes of data per hour, but researchers take a "couple of year breather" to understand what was recorded in just days. This 100x to 10,000x analysis bottleneck highlights the urgent need for AI-driven interpretation to unlock the discoveries hidden within this data.
To build an AI that can reason about biology, researchers must first create a massive "dictionary" of normal cellular behavior. This involves capturing ~50 petabytes of unperturbed dynamics in organisms like zebra fish. Only after establishing this baseline of "native dynamics" can the model effectively interpret data from perturbed or diseased states.
Nobel laureate Eric Betzig claims most biological research is stuck in a reductionist approach, grinding cells down to their component parts. He compares this to reverse-engineering a car from a pile of parts and argues for a new paradigm of holistic observation, admitting that science has not yet sufficiently observed life as a complete, functioning system.
Human perception is limited to 2D plus time, but cellular data from advanced microscopy is 5-dimensional (3D space, time, and color). The solution isn't to accelerate human analysis but to build AI models that can natively perceive and reason within these higher dimensions, much like AI for self-driving cars handles 3D plus time.
With only 9% of drugs in Phase 1 trials ultimately succeeding, there's massive financial waste. By using advanced microscopes to observe the effects of compounds on genetically similar zebra fish, researchers can cheaply screen for unforeseen toxicity and off-target effects, potentially saving hundreds of millions of dollars and preventing patient harm before human trials begin.