And also also shout out to @bkoskella for making RaMP happen and advising on this project!.

Also I'm @ miloj on bsky!.

Eligibility info

Know any recent or upcoming college graduates who are looking for a (full-time&paid!) microbio research experience? Point them towards RaMP! @darian_doakes and I are co-mentors for an MGE project that I think is going to be really cool. Apps+recs by 5/25

(I don't think Sarah and Alena are on here, but I'll tag @skryazhi because I forgot to above).

Major thanks to Alena for leading the experimental work on this, to Sarah for leading the analysis, and to Sarah and Sergey for carefully pulling it all together, writing, etc.!.

All in all, this study gives me more hope that A) there is more low-dimensionality in cellular genotype-phenotype maps than we might think (good news for prediction!), and B) *maybe* someone can figure out mechanistic explanations for these patterns.

Sarah and Sergey point out in the supplement that this trend ^ is also predicted by the statistical model proposed by @Gautam_Reddy_N and @MichaelMDesai, which also can broadly be used to understand all of this data.

A last clue: the λe pivot GR parameter is correlated with the average growth rate, so this environment-specific effect is squeezing the adjusted growth rates towards each other (but note the variation away from the line here is real too):

e.g. these knockouts have a fitness cost strictly related to growth rate or ribosome levels, and also have a fixed benefit (reduction in protein production?) that varies based on the environment. Just a hand-wavey example of a guess, I have no idea!.

I read this as an indication that there are one or a few (possibly measurable) phenotypes underlying the cellular global epistasis we see in our data, and that at least one is correlated with growth rate, and at least one has some environment-specific basis.

These correlations can apply across environments for many mutations and for the mean of the distribution - only one parameter per environment is needed to shift the background growth rates to a universal scale that can predict effects across environments and mutations.

We can see that it isn't a set property of the genotype - there are environments where background fitnesses are largely uncorrelated, and only the background fitness in the assay environment is good at predicting the mean of the distribution of fitness effects (plot color ~ R^2)

But it's not clear that background fitness / growth rate is the right x-axis - maybe there is something correlated to background fitness that is a more proximal predictor of mutn effects, some unseen phenotype or small set of phenotypes.

We've known for a while that low-dimensional (few-parameter) models can explain some variance in the effects of mutations in proteins or whole cells. This "global epistasis" often emerges in lab microbe experiments as negative correlations bwtn background fitness and mutn effects.

A little late to the party, but I'm really excited to see this work out from Sarah Ardell, Alena Martsul, and Sergey Kryazhimskiy over at UCSD, with some help from me from afar. Sergey's thread goes over the main results. Here's why I think they are fun to think about:.

RT @Gautam_Reddy_N: Excited to announce that I've officially joined @PrincetonPhys as an assistant professor. Super grateful to my mentors….

RT @Jess_Gersony: 📢✨ I'm SO thrilled to share that my poetry chapbook "I Could Collect a Lake" is now available through #bottlecappress ! I….

A final version of our paper on the nitty-gritty details of barcode design and sequencing is out!.

RT @loukesio: We have a new preprint with @andydfarr and @RaineyLab. If you are interested in barcoding a microbial species, have a look at….