Andrew Baumgartner

random seed 42 👎👎👎 random seed 69 👍👍👍

This morning, @isbscience.org's Dr. Mary Brunkow was awarded the 2025 Nobel Prize in Physiology or Medicine! Dr. Brunkow shares the award with Fred Ramsdell and Shimon Sakaguchi. Read the press release here: www.nobelprize.org/.../medicine....

What a tremendous honor! Congratulations!

truly wild that a paper in a journal title ALGORITHMS of molecular biology does not have available code

it’s incredible that you can drink 1863268 miller high lifes (lives?) and not get drunk

knee high chelsea boots and soccer shorts. and they seattle isn’t fashionable

people love being like "we need help analyzing this data! it has 223450923707 variables and we have 10 samples"

josie and the pussycats truly is an incredible movie

👼🏻

(extremely over engineered with no comments or documentation)

why do all computer scientists write code Like That™️?

you’re on thin ice here bb

truly insane how sparse the information in ML papers from 5+ years ago is. no code, no comments, no justifications, just vibes

anyway, like i said, this is a step towards that goal, just laying the ground work for interpreting and visualizing transcriptional dynamics using simple tools such as kNN graphs. but with some more time and effort, this bridge can be completed! i'd be happy to work with anyone interested 6/6

however one can take a fully stochastic model and, using the same techniques, turn it into a graph that could be fed into a UMAP, for instance, that could yield be used to accurately visualize transcription dynamics! up to some error that could probably be quantified. 5/

the backward Kolmogorov operator associated to the CRN. here, I take a data driven approach and define a particular SDE that gives rise to the observed data distribution (via the kNN estimate of the sampling density) and use the diffusion map as a basis for clustering. 4/

importantly the weights of the graph define exactly what operator is, and knowing the continuum operator is the most important part for being able interpret the results. we wanted to model the dynamics of chemical reaction networks (CRN) so the correct operator is the Fokker-Planck equation OR 3/

my goal was to bridge commonly used continuum tools of scRNA-seq (diffusion maps, UMAP, etc) with more mechanistic stochastic models. the main idea comes from the insight that markov chains on nearest neighbor graphs are generators for certain operators in the continuum limit. 2/

Fokker-Planck diffusion maps of microglial transcriptomes reveal radial differentiation into substates associated with Alzheimer’s pathology - Communications Biology Combining a novel, biophysically plausible clustering scheme with statistics and interpretable machine learning yields insights into microglia dysfunction in Alzheimer’s disease.

my paper is finally out! i consider this work as a first step in bridging the gap between stochastic and deterministic models of single cell transcriptomics. we model the intermediate N limit via a data driven potential that gives rise to the kNN estimate of the sampling density. let me explain! 1/

me reading ur posts

ive said it once and ill say it again: its incredible that we can learn anything at all using statistics.

people in seattle are paying $16 for a bacon, egg, and cheese

we r cute

missing her (the vince lombardi rest stop)

melville: AND THEYRE KISSING

me [reading moby dick]:

melville [for some reason]: THE AIR IS A GIRL AND THE SEA IS A BOY

i’ll never understand why people chose to stand in the middle of the sidewalk instead of at the actual
bus stop

doi2bib and sci-hub are the unsung heroes of my academic life

i love moby dick because you’ll be reading a chapter about whale’s nose or some shit and all of sudden there’s 3 of the most beautiful sentences ever written in the english language.