Alexey Sergushichev

Assistant professor, Wash U in St. Louis, @wusm_pathology

[Alexey Sergushichev @assaron]

I'm happy to present our new web-tool CORESH https://t.co/r1bswFuZXo for searching public gene expression datasets using gene signature as a query, developed by @vd_sukhov, which was just published in @NAR_open https://t.co/PdDMgQz56a 1/n

[Alexey Sergushichev @assaron]

@nomad421 @dougcj37 @onefiftyfivemm I look at it as a reverse of the famous saying that biology is chemistry, chemistry is physics, and physics is mathematics. Going each step back raises the complexity a lot. So mathematics is "solved", physics is next, chemistry will be harder, and biology (in general) even more.

[Alexey Sergushichev @assaron]

@zelepukiny Name the journal

[Mikhail Dvorkin @mikhail_dvorkin]

Смотрите, какая актуальная игра по нашим-то временам!
(На самом деле там не напрямую про цензуру, а просто хорошее интеллектуальное упражнение.)

[Maxim Artyomov @maxim_artyomov]

Out today in Immunity! Our pan-atlas of human immune aging: 12.4 million PBMCs from 2,609 people across 8 global cohorts (including new 188 people cohort from UK).
https://t.co/hYbe2k3XEi
We map shared aging patterns onto 59 key immune populations and link the balance of GZMK+ and GZMB+ CD8 effector memory T cells to health trajectories; show differences in RTEs and Th2s with sex etc; ethnicity associated features.

https://t.co/AHmFVQVfM6

How it came to be:
in 2023 we published first hierarchically annotatied atlas of human immune aging in PBMC - 166 people, 55 populations, key changes revealed - CD8 naives, KLRC2/NKG2C+ CD8 memory going down, GZMK+ CD8 T cells and Th2 and Tc2 type 2 memory cells going up! (Terekhova et al, Immunity 2023)

in 2024 we added major phenotypes within naive cells - SOX4+ RTE cells and their CD38 marker going down. (Bohacova et al, Immunity 2024)

In 2025-26 more atlases published with their own annotations making impossible to leverage power of data..

Now we provide unique in its power and consistency resource - 2600 people across ages, sexes, races, continents etc. All annotated in sync and analyzed for what aging looks like globally!

[Alexey Sergushichev @assaron]

@zygohistoprepro @ChShersh It might help to install a plugin to run Doom first: https://t.co/gzq6PsNbmO

[alex rubinsteyn @iskander]

FYI I wrapped a lot of immuno-informatics predictors so you don't have to:

https://t.co/FWqNqgkIXG

(eg NetMHCpan, MHCflurry, Pepsickle, DeepTAP, ERAMER, BigMHC, PRIME, &c)

[Alexey Sergushichev @assaron]

@matt_barrie @hqmank https://t.co/7gbOjv7rG3

[Mykyta Artomov @nikart89]

New preprint led by @dmitrij22981 & @ivan_molotkov_ :
Genetic data improves prediction beyond clinical history in 96/150 diseases in @uk_biobank & @FinnGen_FI and is most valuable for people at low clinical risk
Preprint: https://t.co/5VJTixbrHJ
Models: https://t.co/AaiF27kmQb🧵

[Dmitry Penzar @dmitrypenzar]

🧬 EPIC is open: the Eukaryotic Promoter and transcription Initiation prediction Challenge.
The goal: predict where, and how strongly, RNA Pol II starts transcribing, from nothing but DNA sequence. At single-nucleotide, strand-specific resolution.
https://t.co/wS43RvhuY2
🧵 1/7

[karlicoss @karlicoss]

What if fucking web developers started using input html tags with proper type= at least for email/password fields so browser autocompete didn't require AI 😭

[Joe Bebel, PhD @joeintheory]

Knuth is a legend, but if P = NP, then we already know explicit polynomial time algorithms that can solve NP-complete problems! Here is a simple (partial) example: 1/7

[Alexey Sergushichev @assaron]

@JulesJacobs5 @joeintheory @_Felipe I see your point, never thought about it like this. It's still not practically constructive, but mathematically it looks like it is.

[Alexey Sergushichev @assaron]

@joeintheory @JulesJacobs5 @_Felipe Why is this true? I can imagine a proof that says there is a polynomial algorithm, without presenting one.

[Alexey Sergushichev @assaron]

@dovepony Эта аналогия призвана продемонстировать, что ваш аргумент "это просто ..." — несостоятелен, по крайней мере в такой формулировке. В мире довольно много примером эмерджентного поведения, когда из более простых частей получаются сложные объекты, включая нас с вами.

[Alexey Sergushichev @assaron]

@dovepony Клетки вашего тела тоже ничего не "решают", но сгруппировавшись вполне успешно пишут треды в твиттере

[Sasha Malysheva @aimalysheva]

meet @mostik_ai!

what happens when you put 12 PhDs in one room for four months? first place on the ARC-AGI leaderboard, which I can't say much about while the competition is still running. and this, which I can.

everyone's arguing about whether open models will catch up to frontier models. we think it's the wrong question. here's the one we pose: why does a frontier model have to generate your answer at all, when the only thing you need from it is the reasoning?

we do this by enabling models to communicate in latent space. through our protocol, hidden states pass straight from a frontier model into a small one running on your infrastructure -- no text between them, and neither model is fine-tuned. two models from different families, sharing reasoning, both left untouched.

how do we know it works? we tested it on a setup where a 753B model reads the problem, and a 4B edge-class model writes the answer. with this approach, we get results 80% as accurate as the frontier model, but at 20x faster performance.

we're committed to preventing frontier model lock-in and are already partnering with inference providers to accelerate open-weight adoption. we've done this between 15 of us, in four months, 12 PhDs and a Fields medalist, backed by @generalcatalyst

WIRED has the first external account of the company and the work: https://t.co/tP8nItCsDl
full writeup, the setup, and all the numbers: https://t.co/C9NZ5vtV1V

[Alexey Sergushichev @assaron]

Link for the paper https://t.co/f5TeWgod5F Source code is at https://t.co/cTegs8BaBC Documentation is at https://t.co/XmLHkDhwsq 5/5

[Alexey Sergushichev @assaron]

New application note from the lab: "GAMclust: identification of regulated metabolic modules in bulk, single cell and spatial gene expression data" 1/5

[Alexey Sergushichev @assaron]

But it can work with spatial Vizium data too: as a case study we analyzed a glioma dataset: https://t.co/mmfghOj7wM There, one of the modules is glycolysis, which localizes nicely to the tumor core 4/5