ekernf01.bsky.social
Statistician and computational biologist; uw alum; jhu student. He/him. http://ekernf01.github.io
266 posts
1,213 followers
391 following
Regular Contributor
Active Commenter
comment in response to
post
This is awesome! Little chaos gremlin promoters for when you need to be extra random!
comment in response to
post
(h/t @neurostats.org )
comment in response to
post
Both seem sensible. Thanks for answering!
comment in response to
post
I personally seem to lack a gap gap -- perhap, a gap gap gap?
comment in response to
post
They have an impossible job. Intersection has a fast two-lane one-way separating a daycare from a rec center and a church, all heavily attended. And Baltimore drivers are the most vicious and unrepentant I have seen in any US city. So I'm rooting for the DOT.
comment in response to
post
So is the main function of it the foamy, squishy structure, and it managed to co-opt the right set of genes to achieve that without making your ears droop in times of famine?
comment in response to
post
Oh the lipocartilage!! People were passing this around in the Cahan lab a while back but I never gave it a close read!!
comment in response to
post
I admit I don't know compressed sensing. But you're right, there is a difference between the factorial design community testing a small number of factors and worrying about interactions versus the compressed sensing community testing way more and worrying about too many main effects.
comment in response to
post
There are, right now, very deliberate and carefully planned philanthropy projects aiming to unlock new discoveries across many fields of science.
www.gap-map.org
comment in response to
post
There are, right now, very deliberate and carefully planned philanthropy projects aiming to unlock new discoveries across many fields of science.
www.gap-map.org
comment in response to
post
One set of assumptions leads to a really cute optimality result. If you screen N perturbations and expect K of them to each individually be lethal, then allocate each treatment to 1 out of every K+2 samples.
I thought that was nice. Hope you enjoyed it too.
(n/n)
comment in response to
post
How do we balance the need for cost savings (stack more perts per sample) against the chance of killing the samples? I wrote up various assumptions here.
ekernf01.github.io/multiplexed_... (5/n)
comment in response to
post
This can kill the samples. For example, CRISPR with more than 12 target sites can produce >99% growth inhibition. pmc.ncbi.nlm.nih.gov/articles/PMC...
comment in response to
post
To scale up screens with a readout like Cell Painting or RNA-seq, people are starting to just stack up multiple perturbations on top of each other. (3/n)
comment in response to
post
The usual screening approach for e.g. CRISPR screens involves sorting out the phenotype of interest, which only works for yes-no phenotypes. (2/n)
comment in response to
post
"... is all you need" is all you need.
comment in response to
post
"Complete" statistics contain nothing but the truth, but they do not contain the full truth. They would be better called "parsimonious" statistics. End rant.
comment in response to
post
If you want to knock me for buying a second copy, or for including a second set of liner notes, and you don't care that **I only bought the first chapter**, "parsimonious" is what you want.
comment in response to
post
Buying the Complete Guide to Playing the Lute as a two-volume set should not render your lute library incomplete.
comment in response to
post
A similar funny thing happens when you distribute a complete statistic over two pieces: let W = [U + X, U - X] for unit normal X; set v(W)=W[1]+W[2].
comment in response to
post
This is inconsistent with the usual meaning of "complete": buying a second copy of the Complete Guide to Playing the Lute should not render your lute library incomplete.
comment in response to
post
Let W = [U,U]. If you set v(W) = W[1] - W[2], then you find that E[v(W)]=0 for nonzero v.
comment in response to
post
Second, a funny thing happens when you copy a "complete" statistic: it becomes incomplete (?).
comment in response to
post
What's wrong with "complete"? First, "complete" statistics do not contain all information in a sample. ¯\_(ツ)_/¯
stats.stackexchange.com/questions/18...
comment in response to
post
Why care? Completeness is at the very foundation of statistics; it is assumed by a theorem (Lehmann-Scheffe) that says "this estimation method is the most efficient among all unbiased methods".
en.wikipedia.org/wiki/Complet...