Zeming Lin

I'll be at NeurIPS for the workshops if folks want to meet up!

Super excited to preprint our work on developing a Biomolecular Emulator (BioEmu): Scalable emulation of protein equilibrium ensembles with generative deep learning from @msftresearch.bsky.social ch AI for Science.

www.biorxiv.org/content/10.1...

Evolution is non-linear, the next step in the evolutionary lineage of the ESM series of models is a sequence only protein language model that sets a new state of the art in representation learning.

Check it out! The 300M and 600M models are open weights, 300M available for commercial usage.

#CASP16 results are in! Template-based VFold seems to be lead method for nucleic acid structure prediction! AlphaFold2 and 3 still seem to be best methods for protein monomer and complex prediction.

Have you ever wondered how to train an autoregressive generative transformer on text and raw pixels, without a pretrained visual tokenizer (e.g. VQ-VAE)?

We have been pondering this during summer and developed a new model: JetFormer 🌊🤖

arxiv.org/abs/2411.19722

A thread 👇

1/

Better VQ-VAEs with this one weird rotation trick!

I missed this when it came out, but I love papers like this: a simple change to an already powerful technique, that significantly improves results without introducing complexity or hyperparameters.

Diffusion is starting to look much more comprehensible and less like dragons these days 😊

I'm pleased to announce a new blog post from
@ruiqigao.bsky.social and some other folks on my team, showing that diffusion == Gaussian flow matching.
diffusionflow.github.io

Anne Gagneux, Ségolène Martin, @quentinbertrand.bsky.social Remi Emonet and I wrote a tutorial blog post on flow matching: dl.heeere.com/conditional-... with lots of illustrations and intuition!

We got this idea after their cool work on improving Plug and Play with FM: arxiv.org/abs/2410.02423

Plot it as a CDF and it'll look like a sigmoid curve.

I think the #1 way you can tell apart AI images is when they don't seem to tell a story at all. A lot of these seem like psychedelic fractals rather than art with some kind of underlying theme.

✋✋✋

Arxiv sharing reminder

pdf ❌
abs ✅

Update: I clicked follow on entire starter packs and now regret it. I wish I just skimmed them and were a bit more selective.

Self nominating as the 4th person on Pytorch and author on the ESM series of protein LMs.

Nice!

😵😵😵

Hey folks! I got a tattoo of my beloved PIEZO ion channel on my arm! Now I can show how they activate (flatten) under pressure by flexing—Will add a bit of performance art to my science talks 🧪🧠.

Evo, a 7-billion-parameter genomic foundation model, learns biological complexity from individual nucleotides to whole genomes.

Pretraining a genomic foundation model across prokaryotic life.

Fine-tuning on CRISPR-Cas sequences enables generative design of protein-RNA complexes.

Fig. 4. Fine-tuning on IS200/IS605 sequences enables generative design of transposable biological systems.

Evo: A genomic language model of prokaryote genomes generates functional cas9 proteins and transposons.

@brianhiestand.bsky.social

www.science.org/doi/10.1126/...

www.astralcodexten.com/p/how-did-yo...

Two full starter packs for your bioml needs!

We're about to release a pretty massive multiome atlas (includes chromatin) of several adult tissues via ENCODE4 + a significant expansion of the bulk DNase and ATAC datasets (already out). Many more regions than ENCODE3.

What did it? The starter packs??

Complicated configs to me look like programs so I don't know why you would do that in Python, so whatever you pick it should just be in the language you use :)

GitHub - google/fiddle Contribute to google/fiddle development by creating an account on GitHub.

github.com/google/fiddle ?

Hi Bluesky! I made it on Kevin's starter pack 😌😌😌 here to discuss all things bio and all things ML!

In a gratuitous attempt to acquire more followers myself 😁, I've made a start on a "starter pack". Hopefully as more people from 🐦 make it over to 🦋, we can extend this a bit. Suggestions welcome!

I've noticed not all accounts seem to be eligible to be added, anyone know what's up with that? 🤔

Two BioML starter packs now:

Pack 1: go.bsky.app/2VWBcCd
Pack 2: go.bsky.app/Bw84Hmc

DM if you want to be included (or nominate people who should be!)

I don't really understand how this app works. How do I follow all the ML people? 🙃