Which model did you use to predict Mgnify90 sequences?

[zkskek321]

Hi, I'm microbiome researcher in Korea.

There are many versions of ESM-fold and many versions of pre-trained models depending on the number of parameters.

So, I want to know which ESM-fold version and which pre-trained model did you use to predict Mgnify90_highquality_clust30 protein structure uploaded in github below.
(https://github.com/facebookresearch/esm/blob/main/scripts/atlas/README.md)

Thank you so much.

[tomsercu]

Hi! There are two esmfold models: esmfold_v1 and esmfold_v0, both trained on top of the 3B LM but v1 with more recent data. All predicted structures in the ESM Metagenomic Atlas are predicted using esmfold_v0.

[jgoldford]

adding a related question to this thread -- which pre-trained model (name and size) was used to generate the embeddings for the ESM atlas?

[ebetica]

I'm pretty sure it's the 3B model but i don't have access to the original pipeline anymore unfortunately.