externelly/alphafill repository overview

Overview based on from: https://github.com/eporetsky/PlantDiffusion/tree/main/docker/AlphaFill⁠
Some additional thoughts about AlphaFill: https://eporetsky.github.io/bioinformatics/protein_ligand⁠

This Docker container runs AlphaFill⁠, a method for transplanting missing cofactors and molecules from experimental PDB structures to predicted structures.

Build the Docker container:

docker build --rm -f Dockerfile -t externelly/alphafill:latest .

# Needs the PDB-REDO files in the folder to create index
alphafill create-index

# You can run AlphaFill on all mmCIF files in a folder
mkdir filled_cif
for filename_path in cifs/*
do
    filename=$(basename "$filename_path")
    alphafill process cifs/${filename} filled_cif/${filename}
done

Run the docker container with a shared PDB mount:

docker run --name=alphafill -v ${PWD}/:/mount --rm -it externelly/alphafill:latest```

Once you start running the Docker container, your commandline should be redirected to the Docker terminal. From there you can run the different packages mentioned below. The current directory is mounted with the container /input folder to have persistant access with the Docker environment. When using the Docker commandline, you can read and write all files in the /input folder.

⁠Some overview of the AlphaFill workflow

• The AlphaFill installation does not automatically download the PDB-REDO database that AlphaFill uses to extract information from. If you want the complete database, I recommend downloading it seperately since it might take a couple of hours and I read that it weighs over 1TB. Use the following command to download the PDB-REDO⁠ database rsync -av --exclude=attic rsync://rsync.pdb-redo.eu/pdb-redo/ pdb-redo/
• If you want to make additional edits to the config file, you can find it at /usr/local/etc/alphafill.conf
• AlphaFill seems to work only with CIF formatted files for both input and output.
• A more managable solution is to build a custom database of select PDB-REDO files. I will write a longer post about it but the basic idea is if you are dealing with a specific family of proteins, you can extract a subet of PDB IDs from UniProt that share PFAM/InterProScan domains and use these PDBs to make a custom mini-database.

NOTE: To build a custom AlphaFill database, you need to save the PDB-REDO files into a folder formatted as pdb-redo/pdb_id[1:3]/pdb_id/pdb_id_final.cif:

⁠Creating the proprely formatted directory that works with the 'alphafill create-index' command

for pdb_id in pdb_list:
    os.makedirs('pdb-redo/00/{}'.format(pdb_id), exist_ok=True)
    os.system("wget https://pdb-redo.eu/db/{}/{}_final.cif --directory-prefix pdb-redo/{}/{}".format(pdb_id, pdb_id, pdb_id[1:3], pdb_id))

Finally, after all the PDB-REDO files have been downloaded and you have a pdb-redo folder, you can generate the fasta index and run the AlphaFill pipeline with the following commands:

⁠Assumes a /input/pdb-redo folder with PDB-REDO files exists

alphafill create-index

⁠The AlphaFill output CIF will be found at: input/filled/

alphafill process input/cifs/af2_struct.cif input/filled/af2_struct.cif

Enjoy transplanting!

⁠Cite the original work

Hekkelman, M.L., de Vries, I., Joosten, R.P. and Perrakis, A., 2023. AlphaFill: enriching AlphaFold models with ligands and cofactors. Nature Methods, 20(2), pp.205-213.