From 57da016e5d740a9ac5bcf62c3689a42e88584bc6 Mon Sep 17 00:00:00 2001 From: Tom Sercu Date: Tue, 6 Dec 2022 18:31:01 -0500 Subject: [PATCH] document esmfold version (#398) --- scripts/atlas/README.md | 1 + 1 file changed, 1 insertion(+) diff --git a/scripts/atlas/README.md b/scripts/atlas/README.md index 1dd5a0eb..07249eb9 100644 --- a/scripts/atlas/README.md +++ b/scripts/atlas/README.md @@ -4,6 +4,7 @@ The [ESM Metagenomic Atlas](https://esmatlas.com) is a repository of over databa Bulk download instructions are available here, as well as foldseek databases available for download. +All structures in the ESM Metagenomic Atlas were predicted with the ESMFold model released as `esm.pretrained.esmfold_v0()`. We find that protein structures with predicted LDDT > 0.7 and predict TM > 0.7 to be both reasonably well structured and interesting. Therefore, we provide both the small set of "high quality" metagenomic structures, as well as the full set. The small set of structures is built from taking a 30% sequence identity clustering of MGnify90, and using the best structure from each cluster.