Zeming Lin (@ebetica) / X

Zeming Lin

2,417 posts

Zeming Lin

@ebetica

Co-founder at EvolutionaryScale, now Biohub. ESM3 / ESMFold / PyTorch. Unsupervised learner but sometimes still gets a few rewards. Views are my own.

New York, NY

Joined January 2013

Pinned
Zeming Lin
@ebetica
May 27
I'm so excited to show the world what we've been working on the for the past months!! I'm going to highlight some of the fun results from this paper that I find particularly exciting.
Alex Rives
@alexrives
May 27
Today we're announcing ESMFold2, an open scientific engine to power prediction, design, and discovery across protein biology. The new model delivers state of the art performance on protein interactions, especially antibodies, a critical modality for therapeutics. We have
GIF
73K
Zeming Lin
@ebetica
May 11, 2020
@ylecun: "NaNs are bad, except in Indian restaurants." 🤣🤣🤣 Too bad we're in quarantine and the only NaNs I have are in my code 😢
Zeming Lin
@ebetica
Aug 8, 2017
We're releasing a dataset of 65k StarCraft: Brood War games, 1.5b frames, 500m actions, 400GB of data. Check it out!
GitHub - TorchCraft/StarData: Starcraft AI Research Dataset
From github.com
Zeming Lin
@ebetica
Dec 21, 2023
I defended my thesis! Big thanks to my committee @ylecun @rob_fergus @kchonyc @ZhongingAlong @RichBonneauNYC! Thanks to @alexrives and the entire Meta AI protein team as well, @proteinrosh @BrianHie @TomSercu and too many others to name. Now on to more fun science!
94K
Zeming Lin
@ebetica
Jul 2, 2024
biorxiv.org/content/10.110… We're live!
biorxiv.org
Simulating 500 million years of evolution with a language model
More than three billion years of evolution have produced an image of biology encoded into the space of natural proteins. Here we show that language models trained on tokens generated by evolution can...
75K
Zeming Lin
@ebetica
Jun 25, 2024
🧵 Check out what I've been working on at @EvoscaleAI 🧵 We trained a bidirectional masked generative model on protein sequence, structure and function! From 1.4B to 98B parameters, our models are so good that beating ESMFold was just a footnote 🤯 👇👇👇
Alex Rives
@alexrives
Jun 25, 2024
We have trained ESM3 and we're excited to introduce EvolutionaryScale. ESM3 is a generative language model for programming biology. In experiments, we found ESM3 can simulate 500M years of evolution to generate new fluorescent proteins. Read more: evolutionaryscale.ai/blog/esm3-rele…
00:00
71K
Zeming Lin
@ebetica
Mar 23, 2020
Replying to @KodyJones5 and @ScottGottliebMD
I think it's probably safe to say the New York numbers are waaaaay higher than we're admitting though.
Zeming Lin
@ebetica
Mar 16, 2023
Our paper on protein folding with a language model is out in Science!
Science Magazine
@ScienceMagazine
Mar 16, 2023
In a Science study, @metaai researchers show the power of a large language model, #ESMFold, to enable protein structure prediction and analysis. Using ESMFold, they generated a database—the ESM Metagenomic Atlas—of over 600 million metagenomic proteins. scim.ag/1Uv
50K
Zeming Lin
@ebetica
Jul 18, 2024
The ESM3 package was updated on pip. We fixed an issue with iterative structure prediction. Expect to see major improvements after doing: pip install -U esm
13K
Zeming Lin
@ebetica
Jul 21, 2022
Excited to present our work on single-sequence protein folding from a language model! By stacking a simple folding trunk and Alphafold2's structure module on top of the language model, we get accurate structure prediction in a fraction of the runtime.
Alex Rives
@alexrives
Jul 21, 2022
We have trained ESMFold to predict full atomic protein structure directly from language model representations of a single sequence. Accuracy is competitive with AlphaFold on most proteins with order of magnitude faster inference. By @metaai Protein Team. biorxiv.org/content/10.110…
00:00
Zeming Lin
@ebetica
Nov 1, 2022
Introducing ESMAtlas, a repository of 600M+ protein structures we folded in the few weeks after ESMFold. It's an unprecedented view into the "dark matter" of the protein universe. Check it out! esmatlas.com
AI at Meta
@AIatMeta
Nov 1, 2022
Announcing the ESM Metagenomic Atlas — the first comprehensive view of the ‘dark matter’ of the protein universe. Made possible by ESMFold, a new breakthrough model for protein folding from Meta AI. More in our new blog ➡️ bit.ly/3Nm7Pxn 1/3
00:00
esmatlas.com
ESM Metagenomic Atlas | Meta AI
An open atlas of 617 million predicted metagenomic protein structures
Zeming Lin
@ebetica
Jul 13, 2023
Replying to @LukeGessler
Yeah but the out of domain datasets are literally languages the models haven't been trained on. Not surprising at all. The bigger takeaway is that it beats SentenceBERT on in-domain datasets...
25K
Zeming Lin
@ebetica
May 16, 2018
Replying to @fermatslibrary
(pi * 10^7 seconds) is often a great mathematical approximation for 1 year, and can greatly simply some equations, especially in physics.
Zeming Lin
@ebetica
Aug 8, 2017
We're releasing a dataset of 65k StarCraft: Brood War games, 1.5b frames, 500m actions, 400GB of data. Check it out!
arxiv.org
STARDATA: A StarCraft AI Research Dataset
We release a dataset of 65646 StarCraft replays that contains 1535 million frames and 496 million player actions. We provide full game state data along with the original replays that can be viewed...