localpdb

localpdb provides a simple framework to store the local mirror of the protein structures available in the PDB database and other related resources.

The underlying data can be conveniently browsed and queried with the pandas.DataFrame structures. Update mechanism allows to follow the weekly PDB releases while retaining the possiblity to access previous data versions.

You may find localpdb particularly useful if you:

• already use Biopython Bio.PDB.PDBList or similar modules and tools like CCPDB,
• build custom protein datasets based on multiple criteria, e.g. for machine learning purposes,
• create pipelines based on the multiple or all available protein structures,
• are a fan of pandas DataFrames.

Overview

To find more about the package and its functionalities please follow the docs.

Installation

pip install localpdb

Setup the database and sync protein structures in the mmCIF format:

localpdb_setup -db_path /path/to/localpdb --fetch_cif

More information on the setup options are available via docs.

Examples

Find number of entries added to the PDB every year

from localpdb import PDB

lpdb = PDB(db_path='/path/to/your/localpdb')
lpdb.entries = lpdb.entries.query('deposition_date.dt.year >= 2015 & deposition_date.dt.year <= 2020')

df = lpdb.entries.groupby(by=['method', lpdb.entries.deposition_date.dt.year])['mmCIF_fn'].count().reset_index()

sns.barplot(data=df, x='deposition_date', y='mmCIF_fn', hue='method')

Create a custom dataset of protein chains

Select:

• human SAM-dependent methyltransferases,
• solved with X-ray diffraction,
• with resolution below 2.5 Angstrom
• deposited after 2010.
• remove the sequence redundancy at 90%,

# Install plugins providing additional data
localpdb_setup -db_path /path/to/your/localpdb -plugins SIFTS ECOD PDBClustering

from localpdb import PDB
import gzip

lpdb.entries = lpdb.entries.query('type == "prot"') # Protein structures
lpdb.entries = lpdb.entries.query('method == "diffraction"') # solved with X-ray diffraction
lpdb.entries = lpdb.entries.query('resolution <= 2.5') # with resolution below 2.5A
lpdb.entries = lpdb.entries.query('deposition_date.dt.year >= 2010') # added after 2010
lpdb.chains = lpdb.chains.query('ncbi_taxid == "9606"') # human proteins
lpdb.chains = lpdb.chains.query('50 < sequence.str.len() < 1000') # with defined length range
lpdb.ecod = lpdb.ecod.query('t_name == "S-adenosyl-L-methionine-dependent methyltransferases"') # SAM dependent methyltransferases

# Remove redundancy (select only representative structure from each sequence cluster)
lpdb.load_clustering_data(redundancy=90)
lpdb.chains = lpdb.chains[lpdb.chains['clust-90'].notnull()]

representative = lpdb.chains.groupby(by='clust-90')['resolution'].idxmin()
lpdb.chains = lpdb.chains.loc[representative]

lpdb.chains.to_csv('dataset.csv') # Save dataset

Advanced examples

• Dynamics of the HIV protease inferred from the PDB structures.

• Amino acid preferences among the viral coiled-coil domains.

• Creating a dataset for machine learning - example of building DeepCoil training and test sets.

Troubleshooting

In case of any troubles free to contact us or open an issue.

Acknowledgments

This work was supported by the National Science Centre grant 2017/27/N/NZ1/00716.