1
$\begingroup$

I am looking for a way to download all pdb files that detail a protein-drug complex.

While I could download the entirety of the pdb and search manually (or maybe write a script to discard all proteins that do not have an HETATM entry that details a non-water, non-salt molecule), that sounds effort intensive and error prone.

It seems reasonable that this subset already exist, but I was not able to find it.

Does such a list/database exists?

$\endgroup$
  • 1
    $\begingroup$ Albeit only a small fraction, the SGC has Target Enabling Packages which are few dozen sets of crystal structures each with several hits, i.e. non-natural targets, but not necessarily with nanomolar affinity but generally obeying Lipinski's rules (drug-like). For some times of analysis this wider but shorter set many be a better set than a messier set with only one target each... $\endgroup$ – Matteo Ferla Feb 3 at 9:30
  • 1
    $\begingroup$ Also, if you do end up looking manually, in this blog post is a list of crystallisation ligands, like glycerol, to blacklist. $\endgroup$ – Matteo Ferla Feb 3 at 9:31
  • 1
    $\begingroup$ Above I mention a small but diverse dataset with many ligands each. An alternative is KLIFS, which has only kinases, which is curated, has a lot of members and many ligands, but is not at all diverse. As a result it is a clean dataset often used as a testbed for novel methodologies. However, it represents only one family. $\endgroup$ – Matteo Ferla Feb 6 at 19:04
2
$\begingroup$

Proving a negative is hard, but to my knowledge I do not know a database that has all the structure of drug bound protein. I can give some close matches.

First a few terms to be on the same page. A ligand is something that binds, which can be physiological, such as a cofactor or substrate, or not (especially man-made), such as a drug, i.e. a compound that has a physiological effect, or a hit, a small fragment from a fragment screen that will not bind with nanomolar affinity. Generally, ligands are small molecules, but may include oligopeptides and oligonucleotides. In the case of crystal structures, you also have crystallisation additives, which are not ligands, but are hetatoms in PDBs. Also, the definition of drug may vary. drugbank.ca has the following categories of drugs "approved", "experimental", "withdrawn" for pharmaceuticals, "illicit" for narcotics and "investigational" for probes and counts macromolecules as drugs.

  • RCSB PDB: drug mapping table is a simple searchable table with ligands that are clinical drugs only —experimental compounds (=probes) will be absent.

  • bindingdb.org contains many table, including a '3D' one. This table contains all the PDBs with ligands and their affinities. This contains crystallisation additives and natural ligands. Peter Curran's repo for the HotSpots algorithm (first written by Chris Radoux) has a blacklist, but you are bound to get stuff missed.

  • bindingmoad.org is a database with all the protein-ligand structures from PDB up until 2018, but it includes natural ligands, like cofactors and substrates as above.

  • https://klifs.vu-compmedchem.nl a well curated database of kinases, which has a lot of members and many ligands, but is not at all diverse. As a result it is a clean dataset often used as a testbed for novel methodologies.

  • I mentioned in the comments the SGC Target Enabling Packages as a small subset of proteins with many hits each. But I realise that there is not a handy PDB list and the targets are not drugs, but probes.

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.