You’ll come to believe it eventually.

Yes. It's all downhill from here. I know. I'm talking from experience.

Please let us know what you think. And much agreed on Bach and Vermeer (maybe I should give Proust another try).

and includes accurate error-bars on all estimates.

We believe RealTrace can dramatically enhance the power of time-lapse microscopy data by allowing identification and quantification of far more subtle features in the dynamics than is possible with simple data smoothing approaches. n/n

Surprisingly, protein production rates adapt to their new steady-states far more quickly and with only a minimal transient.

RealTrace can be applied to essentially any time-lapse fluorescence microscopy data and gives accurate estimates of instantaneous growth and volumic GFP production rates, 10/n

Finally, upon a sudden `downshift' in nutrients, we see very reproducible time-dependent responses across single cells. All cells go almost into growth arrest, then overshoot in growth rate, only to relax to a final growth rate over a time scale of 3 cell cycles. 9/n

minimum early in the cell cycle and a maximum late in the cell cycle. In contrast, protein production rates show more complex patterns that vary across conditions and across promoters, even across constitutive promoters!
As far as we are aware, no current models can explain these patterns. 8/n

time, and in slower growth conditions, both the amplitude and duration of these fluctuations increases.

Second, both growth rate and volumic protein production vary systematically across the cell cycle. Interestingly, the variation in growth rate is the same in all conditions, with a 7/n

We use data from E. coli cells carrying fluorescent reporters of constitutive and ribosomal genes, growing in a microfluidic device in different conditions, to highlight the kind of subtle patterns RealTrace can uncover.
First, instantaneous growth rates vary substantially across cells and 6/n

RealTrace is very generally applicable, which we demonstrate by applications to data from E. coli cells, mouse embryonic stem cell nuclei (data from the lab of @davidsuter.bsky.social), and entire C. elegans larvae (data from the lab of @betowbin.bsky.social). 5/n

are correlated on short time scales, all measurement errors are independent. We implemented this idea into a rigorous Bayesian procedure that uses maximum entropy process priors and recursively approximates the non-linear dynamics over short time intervals. 4/n

We can't fit the data to particular curves because we typically don't know what form the dynamics will take, and local smoothing systematically distorts the true dynamics by replacing it with a time average.

RealTrace solves this challenge using only one assumption: while biological changes 3/n

Tracking the behavior of single cells using fluorescence time-lapse microscopy is becoming increasingly popular but there is a fundamental challenge in analyzing such data: as biological changes are small on short time scales, much of the true dynamics is hidden under measurement noise. 2/n

Are you saying that is bad? I already KNOW I'm interested in your lab.. I just used the LLM to help me express it.