Simona Cristea

This account is verified. Learn more

@simocristea

multimodal AI for bio & health @TempusAI; ex faculty @DanaFarber & group leader @Harvard; phd cs @eth; views & tweets mine

Boston 🇺🇸 & Zurich🇨🇭 simona.ai Joined January 2016

467 Following 9.5K Followers 4.7K Posts

Latest posts 5

Pinned post

Simona Cristea @simocristea Jul 7, 2025

scRNAseq cell type annotation is notoriously messy. Despite so many algorithms, most researchers still rely on manual annotations using marker genes

In a new preprint accepted at ICML GenAI Bio Workshop, we ask if reasoning LLMs (DeepSeek-R1) can help with cell type annotation🧵

Simona Cristea @simocristea Oct 11

“passes a stage 3 trial” is what gets every single drug into use.

absolutely nobody fully, mathematically understands the mode of action for any drug.

TT must know that drugs get approved via trials, and not by having expert X map out all the biochemical reactions of the drug.

Andrew Curran Verified account @AndrewCurran_ Oct 10

This is a slide from Terence Tao's 'Math 2.0' public lecture yesterday evening at Caltech. I am highly confident that, if the cure actually worked, almost any Stage 4 cancer patient would tell you that human comprehension of its mechanism is completely irrelevant to them.

View quoted post on X

Simona Cristea @simocristea Oct 9

heard similar stuff from maths friends, nobody understands these, nobody can read them, the only way to navigate it is with AI.. and nobody knows what to do

probably parsing out AI proofs into human understandable language is valuable.. but even that seems not possible

Lev Reyzin Verified account @lreyzin Oct 8

Apparently, my coauthor pointed out that the OpenAI theorem drop included something we were working on (with AI help, of course). The OpenAI result has better bounds than we achieved, but it's also unreadable.

Any advice on what I'm supposed to do now?
View quoted post on X

Simona Cristea @simocristea Oct 4

the world remains so mysterious

Simona Cristea @simocristea Oct 3

this is good advice. reminder:

positive control = something I know always to be true. I’m gonna test it to see if I see it as true. if yes, it shows me that my approach is able to discover things that are true. if not, it clearly means that my experiment has limitations in showing as true what is actually true.

negative control = something I know to be false. if I see it as false, it means my approach is able to not highlight things which are not true. if I see it as true, it means that not everything my experiment shows to be true is actually true.

choosing the right controls is a wonderful scientific task and a true testimony to understanding the problem you are trying to solve.

Anshul Kundaje @anshulkundaje Oct 2

AI folks who are new to bio should spend some time with wet lab folks not only to understand how experiments, assays, lab work is done but also to learn the basic concept of CONTROLS. 1/
View quoted post on X