Curated ProteinMPNN training dataset

The multi-chain training data for ProteinMPNN

Quickstart Usage

Install HuggingFace Datasets package

Each subset can be loaded into python using the Huggingface datasets library. First, from the command line install the datasets library

$ pip install datasets

Optionally set the cache directory, e.g.

$ HF_HOME=${HOME}/.cache/huggingface/
$ export HF_HOME

then, from within python load the datasets library

>>> import datasets

Load model datasets

To load one of the group_mpnn model datasets, use datasets.load_dataset(...):

>>> dataset_tag = "train"
>>> dataset_models = datasets.load_dataset(
  path = "leebecca/group_mpnn",
  name = f"{dataset_tag}_models",
  data_dir = f"{dataset_tag}")['train']

and the dataset is loaded as a datasets.arrow_dataset.Dataset

>>> dataset_models
Dataset({
    features: ['id', 'pdb', 'Filter_Stage2_aBefore', 'Filter_Stage2_bQuarter', 'Filter_Stage2_cHalf', 'Filter_Stage2_dEnd', 'clashes_bb', 'clashes_total', 'score', 'silent_score', 'time'],
    num_rows: 211069
})

which is a column oriented format that can be accessed directly, converted in to a pandas.DataFrame, or parquet format, e.g.

>>> dataset_models.data.column('pdb')
>>> dataset_models.to_pandas()
>>> dataset_models.to_parquet("dataset.parquet")

Dataset Description

This dataset contains metadata and per-chain sequence and tensorized coorindates for the multi-chain training data for ProteinMPNN

Overview of Dataset

The original pdb folder has been curated such that the entire dataset has been split into train,test and validate folders.

Original dataset for ProteinMPNN curated by Ivan Anishchanko.

Each PDB entry is represented as a collection of .pt files: PDBID_CHAINID.pt - contains CHAINID chain from PDBID PDBID.pt - metadata and information on biological assemblies

PDBID_CHAINID.pt has the following fields: seq - amino acid sequence (string) xyz - atomic coordinates [L,14,3] mask - boolean mask [L,14] bfac - temperature factors [L,14] occ - occupancy [L,14] (is 1 for most atoms, <1 if alternative conformations are present)

PDBID.pt: method - experimental method (str) date - deposition date (str) resolution - resolution (float) chains - list of CHAINIDs (there is a corresponding PDBID_CHAINID.pt file for each of these) tm - pairwise similarity between chains (TM-score,seq.id.,rmsd from TM-align) [num_chains,num_chains,3] asmb_ids - biounit IDs as in the PDB (list of str) asmb_details - how the assembly was identified: author, or software, or smth else (list of str) asmb_method - PISA or smth else (list of str)

asmb_chains    - list of chains which each biounit is composed of (list of str, each str contains comma separated CHAINIDs)
asmb_xformIDX  - (one per biounit) xforms to be applied to chains from asmb_chains[IDX], [n,4,4]
                 [n,:3,:3] - rotation matrices
                 [n,3,:3] - translation vectors

list_with_splits.csv: CHAINID - chain label, PDBID_CHAINID DEPOSITION - deposition date RESOLUTION - structure resolution HASH - unique 6-digit hash for the sequence CLUSTER - sequence cluster the chain belongs to (clusters were generated at seqID=30%) SEQUENCE - reference amino acid sequence SPLIT - split which each pdb id belongs to

Train, test, and validate splits are available for download as .tar.gz files

• Acknowledgements: We kindly acknowledge the ProteinMPNN team, RosettaCommons, and the following institutions: University of California, Los Angeles; University of Maryland; University of Oregon; University of Michigan; University of Pennsylvania; and the Wistar Institute.

• License: rosetta-license-1.0

Dataset Sources

• Repository: https://github.com/dauparas/ProteinMPNN/tree/main/training
• Paper: Dauparas, J., Anishchenko, I., Bennett, N., Bai, H., Ragotte, R. J., Milles, L. F., … Baker, D. (2022). Robust deep learning-based protein sequence design using ProteinMPNN. Science (New York, N.Y.), 378(6615), 49–56. doi:10.1126/science.add2187

Uses

Exploration of sequence-structure relationships, not limited to inverse folding models.

Out-of-Scope Use

This dataset has been curated with restraints imposed by the ProteinMPNN team. Thus caution much be used when using it as training data for protein structure prediction.

Source Data

A set of 19,700 high-resolution single-chain structures from the Protein Data Bank (PDB) were split into train, validation, and test sets (80/10/10) based on the CATH protein classification database. This set contains protein assemblies in the PDB (as of 2 August 2021) determined by x-ray crystallography or cryo–electron microscopy (cryo-EM) to better than 3.5-Å resolution and with fewer than 10,000 residues.

Citation

@article{Dauparas2022,
title = {Robust deep learning–based protein sequence design using ProteinMPNN},
volume = {378},
ISSN = {1095-9203},
url = {http://dx.doi.org/10.1126/science.add2187},
DOI = {10.1126/science.add2187},
number = {6615},
journal = {Science},
publisher = {American Association for the Advancement of Science (AAAS)},
author = {Dauparas,  J. and Anishchenko,  I. and Bennett,  N. and Bai,  H. and Ragotte,  R. J. and Milles,  L. F. and Wicky,  B. I. M. and Courbet,  A. and de Haas,  R. J. and Bethel,  N. and Leung,  P. J. Y. and Huddy,  T. F. and Pellock,  S. and Tischer,  D. and Chan,  F. and Koepnick,  B. and Nguyen,  H. and Kang,  A. and Sankaran,  B. and Bera,  A. K. and King,  N. P. and Baker,  D.},
year = {2022},
month = oct,
pages = {49–56}
}

Dataset Card Authors

Miranda Simpson (miranda13nicoles@gmail.com), Becca Lee (beccalee5@g.ucla.edu), Nathaniel Felbinger (nfelbing@umd.edu), Pratyush Dhal (pdhal@umich.edu), Colby Agostino (colby.agostino@pennmedicine.upenn.edu)