Skip to content

Building databases: PDB and AlphaFold

SnpEff supports Protein interactions calculated from PDB or AlphaFold. Adding protein interaction to existing SnpEff databases is rather involved, so it's recommended you use a pre-built database if available (e.g. human).

Requirements

In order to add protein data to a database, you need:

  • A SnpEff database for the organism. The database should have been built or downloaded prior to start adding the protein data
  • Protein data: One *.pdb.gz file for each protein. This can be downloaded from PDB (real data from crystallized proteins) or AlphaFold (predicted protein structures).
  • A "mapping file" that maps protein IDs to transcript IDs

Warning

Adding protein data to a SnpEff database, requires a database to be already built / downloaded

Command line reference

Usage: snpEff pdb [options] genome_version

Options:
    -aaSep <num>              : Minimum amino acid separation within the sequence. Default: 20
    -idMap <file>             : File mapping protein IDs to transcript IDs (tab-separated).
    -maxDist <num>            : Maximum distance in Angstrom for amino acids to be considered 'in contact'. Default: 3.0
    -maxErr <num>             : Maximum amino acid sequence mismatch rate (0.0 to 1.0). Default: 0.05
    -org <name>               : Organism common name (case-insensitive). Default: HUMAN
    -orgScientific <name>     : Organism scientific name (case-insensitive). Default: HOMO SAPIENS
    -pdbDir <path>            : Path to directory containing PDB files (searched recursively).
    -res <num>                : Maximum PDB file resolution in Angstrom. Default: 3.0

Usage example

java -jar snpEff.jar pdb \
    -v \
    -pdbDir /path/to/pdb_files/ \
    -idMap /path/to/id_mapping.txt \
    GRCh38.mane.1.0.ensembl

Output

The command produces an interactions.bin file in the genome's data directory (e.g. data/GRCh38.mane.1.0.ensembl/interactions.bin). This file contains tab-separated records of amino acid pairs that are in close proximity within protein structures, along with their genomic coordinates. Once created, SnpEff will automatically use this file when annotating variants.

Protein structure data

You can download protein structure data from PDB by executing something like: 

rsync -rlpt -v -z --delete --port=33444 rsync.wwpdb.org::ftp_data/structures/divided/pdb/
See PDB's documentation.

For AlphaFold, you can download a tar file with all protein predictions from Ensembl's AlphaFold download page. Please check the documentation on how to cite AlphaFold if you use their databases in a paper.

PDB file filters: By default SnpEff will filter PDB files using the following criteria

  • SnpEff will search PDB formatted file in all subdirectories.
  • The file names must end with any of the extensions (case-sensitive): *.ent, *.ent.gz, *.pdb, or *.pdb.gz (as usual, files ending in .gz are expected to be gzip-compressed)
  • The protein ID must be parsed from the file name. Currently accepted file name formats are "PDB-style" (e.g. pdb7daa.ent.gz, where PDBID is 7daa) or AlphaFold-style (e.g. AF-Q9Y6V0-F9-model_v2.pdb.gz, where Uniprot-ID is Q9Y6V0)
  • The PDB file must have a resolution of at least 3 Angstrom. This value can be changed using the -res command line option

Info

If the PDB file has a missing resolution value, or the resolution is over 99.0, then it is assumed to be 'missing value' and not filtered. For instance, AlphaFold PDB files usually don't have resolution values since they are predictions, not measurements.

Mapping file

Since SnpEff uses transcript ID information, whereas proteins use PDB IDs (or Uniprot IDs in case of AlphaFold), we need a file to map across these different IDs. You should provide a mapping file that maps proteinId to transcriptId:

The format is: tab-separated, one line per entry: 

proteinID   transcriptID