Alison G. TeboGroup Leader · Janelia Research Campus · HHMI
Fluorescent biosensors turn the hidden chemistry of living cells into light — and my lab is building a systematic way to engineer them.
The idea
A biosensor is a window. It lets us watch — in real time, inside living systems — when and where the molecules of life interact.
Yet building one is still an art: idiosyncratic, stochastic, one sensor at a time. I want to make it a science.
Why it matters
Calcium spikes. Metabolic flux. Enzyme activity. A biosensor converts these invisible events into fluorescence we can measure — directly, in a living cell.
Give each a different color, and we can watch many at once: a multiplexed, real-time portrait of cellular state.
Green and red-shifted sensors, side by side — the colors that make multiplexing possible.
The approach
An optimized sensor often differs from where it started by only a handful of mutations — yet finding them can take years of trial and error, and the lessons rarely transfer. We change that by combining three things.
Explore sequence space at scale and by design — sampling architectures and variants far beyond what hand-picked guesses can reach.
Understand the allostery and photophysics that make a sensor work, so improvements come from mechanism rather than luck.
Learn the rules that connect sequence to function — and let models guide which sensors are worth building next.
From stochastic search toward model-guided, mechanism-informed design.
The work so far
Red calcium indicator
A high-performance red calcium sensor built on mScarlet, engineered through a new topology and directed evolution to deliver the bright, clean red signals that multiplexed imaging has been missing.
Tested in vivo across mouse, zebrafish, fly & worm
Population-split directed evolution yields HYlight2.
Glycolysis · fructose-1,6-bisphosphate
A brighter sensor for the sentinel metabolite of glycolysis, evolved by exploring four regions of sequence space in parallel — letting us resolve metabolic dynamics that were previously out of reach.
2.5× the sensitivity of HYlight1Beneath both sits a quieter result: machine-learning and biophysical models that turn noisy screening data into interpretable, transferable design rules.
What's next
Couple large-scale assays with machine learning, so each round of experiments tells the model what to build next.
Map how sequence sets the allostery between domains and the photophysics of the fluorophore, turning empirical screening into design rules that generalize across sensor classes.
Build sensors for the hard targets — protease activity, molecular ratios that reflect pathway flux — using new topologies that current heuristics can't reach.
A biologist who imagines a new measurement should be able to hold a tested sensor within months.
That is the lab we are building.