protein-detective

Python package to detect proteins in EM density maps.

It uses

• Uniprot Sparql endpoint to search for proteins and their measured or predicted 3D structures.
• powerfit to fit protein structure in a Electron Microscopy (EM) density map.

An example workflow:

graph LR;
    search{Search UniprotKB} --> |uniprot_accessions|fetchpdbe{Retrieve PDBe}
    search{Search UniprotKB} --> |uniprot_accessions|fetchad{Retrieve AlphaFold}
    fetchpdbe -->|mmcif_files| residuefilter{Filter on nr residues + write chain A}
    fetchad -->|pdb_files| densityfilter{Filter out low confidence}
    residuefilter -->|pdb_files| powerfit
    densityfilter -->|pdb_files| powerfit
    powerfit -->|*/solutions.out| solutions{Best scoring solutions}
    solutions -->|dataframe| fitmodels{Fit models}

Install

pip install protein-detective

Or to use the latest development version:

pip install git+https://github.com/haddocking/protein-detective.git

Usage

The main entry point is the protein-detective command line tool which has multiple subcommands to perform actions.

To use programmaticly, see the notebooks and API documentation.

Search Uniprot for structures

protein-detective search \
    --taxon-id 9606 \
    --reviewed \
    --subcellular-location-uniprot nucleus \
    --subcellular-location-go GO:0005634 \
    --molecular-function-go GO:0003677 \
    --limit 100 \
    ./mysession

(GO:0005634 is "Nucleus" and GO:0003677 is "DNA binding")

In ./mysession directory, you will find session.db file, which is a DuckDB database with search results.

To retrieve a bunch of structures

protein-detective retrieve ./mysession

In ./mysession directory, you will find mmCIF files from PDBe and PDB files and AlphaFold DB.

To filter AlphaFold structures on confidence

Filter AlphaFoldDB structures based on density confidence. Keeps entries with requested number of residues which have a confidence score above the threshold. Also writes pdb files with only those residues.

protein-detective density-filter \
    --confidence-threshold 50 \
    --min-residues 100 \
    --max-residues 1000 \
    ./mysession

To prune PDBe files

Make PDBe files smaller by only keeping first chain of found uniprot entry and renaming to chain A.

protein-detective prune-pdbs \
    --min-residues 100 \
    --max-residues 1000 \
    ./mysession

Powerfit

Generate the powerfit commands for the filtered and pruned structures.

protein-detective powerfit commands ../powerfit-tutorial/ribosome-KsgA.map 13 docs/session1

This will print commands to the terminal, which you can then run in whatever way you prefer. Like just sequentially, or with GNU parallel or as a Slurm array job.

Alternatively, you can use the protein-detective powerfit run ... command to run powerfit commands sequentially, which is useful for small datasets with rough options.

To print top 10 solutions to the terminal, you can use:

```shell
protein-detective powerfit report docs/session1

Outputs something like:

powerfit_run_id,structure,rank,cc,fishz,relz,translation,rotation,pdb_id,pdb_file,uniprot_acc
10,A8MT69_pdb4e45.ent_B2A,1,0.432,0.463,10.091,227.18:242.53:211.83,0.0:1.0:1.0:0.0:0.0:1.0:1.0:0.0:0.0,4E45,docs/session1/single_chain/A8MT69_pdb4e45.ent_B2A.pdb,A8MT69
10,A8MT69_pdb4ne5.ent_B2A,1,0.423,0.452,10.053,227.18:242.53:214.9,0.0:-0.0:-0.0:-0.604:0.797:0.0:0.797:0.604:0.0,4NE5,docs/session1/single_chain/A8MT69_pdb4ne5.ent_B2A.pdb,A8MT69
...

To generate model PDB files rotated/translated to PowerFit solutions, you can use:

protein-detective powerfit fit-models docs/session1

Contributing

For development information and contribution guidelines, please see CONTRIBUTING.md.