Published September 29, 2023 | Version 0.0.1
Dataset Open

FoldingDiff generated structures (n=780, main results) and associated metadata

  • 1. Stanford University
  • 2. Microsoft Research

Description

Backbone structures generated by FoldingDiff spanning lengths [50, 128). Each length has 10 randomly sampled structures for a total of 780 backbone structures. These were used to derive all results in our manuscript's main results section. In addition to structures in .pse format, we provide an excel table with the following sheets:

  • Table containing metadata for each of the aforementioned generated structures. Metadata includes scTM designability scores using ProteinMPNN + OmegaFold and using ProteinMPNN + AlphaFold2, maximum training set TM score (similarity), structure length, and number of sheets/helices present as annotated by P-SEA.
  • Table containing Gauss integral embeddings for each of the 780 backbones generated by FoldingDiff
  • Table containing Gauss integral embeddings for select test set structures between 50 and 128 residues in length. These were used to compare and contextualize structures/embeddings from FoldingDiff.

Files

generated-structures-supplement.zip

Files (13.9 MB)

Name Size Download all
md5:1a6036e243e0b5e11a1a4dd585307642
13.9 MB Preview Download