1/ The biggest blocker to effective AI-in-bio is lack of quality data, but @vevo_ai is open-sourcing their gigantic #Tahoe100M single cell dataset (on @arcinstitute's new Virtual Cell Atlas)!
I talk w/Vevo's founders and Arc CEO/CTO about "Virtual Cell Models" and significance https://t.co/ldR9S2DLGj
2/ Or watch @nalidoust @iamjohnnyyu @pdhsu @davey_burke @genophoria on @NoPriorsPod YouTube: https://t.co/mMGpb0KN1S or listen wherever you get your podcasts
3/ Access the dataset here https://t.co/IiMECoNuTs
4/ Some takeaways:
- this is an "ImageNet" moment for biology
- predicting protein structure (Evo, AlphaFold, Chai) was a big AI/biology breakthrough, but we need to model entire human cells—how they behave in disease, or how they respond to new therapeutics
- the same way LLMs needed enormous text corpora to become truly powerful, Virtual Cell Models need massive, high-quality cellular datasets to train on
- today’s single-cell data (how drugs affect individual cells) is too small / narrow to fully unlock AI’s potential in drug discovery. most data out there is observational (just looking at normal cells) rather than perturbational ((systematically measuring how cells change when exposed to drugs, genetic edits, disease contexts, etc.)
- the new Tahoe-100 dataset is effectively 50X larger than the public single-cell datasets, the first big enough for AI (generated with @vevo_ai's Mosaic platform)
- @arcinstitute is aggregating a broader “Virtual Cell Atlas,” which also includes their own AI-processed single-cell data (scBaseCamp)
5/ Congrats to the teams!
It’s morning in biology ☀️
6/ More on scBaseCamp from Arc. Exciting that the first "AI co-scientists" are eating into the grind of data collection and curation
https://t.co/bpQPy0c5D0
Quoted context
These are the quoted author’s words.
At Arc we are building AI models of cell state from the ground up, rethinking every step, from data generation to biologically relevant evaluation
Today we launch scBaseCamp, the largest public repository of single cell RNAseq data, uniformly processed from raw sequencing reads. https://t.co/kiqpe3zSqh
yusufroohani ↗
Today, we're launching the Arc Virtual Cell Atlas, a growing resource for computation-ready single-cell measurements. As the initial contributions, @vevo_ai has open sourced Tahoe-100M, the world's largest single-cell dataset, mapping 60,000 drug-cell interactions, and we’re announcing scBaseCamp, the first RNA sequencing data repository curated using AI agents. Combined, the release includes data from over 300 million cells.
https://t.co/OrjRT6YwWS
arcinstitute ↗