Ligand extraction

Ligand extraction

What it does

Takes an experimental structure that has something bound in it — a drug, a cofactor, an inhibitor — and gives you that molecule back as a separate SDF file, in its crystallographic position.

It is the step between "I have a protein–drug complex from the PDB" and being able to do anything with the drug half of it. Structure cleanup removes heterogens; Contact mapper (chemical) reports what a ligand touches but never hands the ligand back. This is the tool that gives you the file.

Why you would want it

Two things need it, and neither works without it:

• A docking box around the real site. Small-molecule docking's reference-ligand box draws the search space around a file you give it. The bound ligand is the most reliable thing to point it at — it is, by definition, exactly where a molecule binds.
• A reference pose. Pose RMSD measures docked poses against a position already known to be correct. The crystal ligand is that position.

Together those two make redocking validation possible: strip the ligand out, dock a fresh copy back in, and see whether the calculation reproduces an answer nature already gave you.

What to give it

The template matters more than it looks

A ligand in a PDB or mmCIF file is stored as coordinates and element symbols, and nothing else. There are no bond orders in the file at all. Which bonds are single, double or aromatic has to be worked out, and working it out from atom positions alone is a guess that fails quietly — aromatic rings come back as alternating single and double bonds, or as neither.

Supplying a template fixes this exactly: the bond orders come from a molecule that genuinely has them, and only the coordinates come from the entry.

Without a template the tool still works, and warns you. For a docking box that is usually fine — a box only needs to know where the atoms are. For anything chemical, supply the template.

If the template is a different compound, the run fails rather than producing a molecule with the wrong bonds. That is deliberate: nothing downstream could detect the mistake.

Reading the output

One SDF, plus a summary that tells you what you actually got:

• SMILES — check it. This is the fastest way to confirm you extracted what you meant to.
• Heavy atoms — the count, hydrogens excluded (crystal structures rarely resolve hydrogens).
• Chain and copies in structure — if the entry has several copies of the ligand, the first is used and the tool says so. Name a chain to pick another.
• Bond orders — from template or perceived from geometry.

Limitations

• One copy per run. A structure with three copies of the ligand gives you one of them, named. Run it again with a different chain for another.
• First model only. In an NMR ensemble, later models are the same molecule in other conformations; combining them would produce nonsense.
• Hydrogens are not invented. What comes out is what the entry contains, which for almost every crystal structure means heavy atoms only. Tools that need hydrogens add their own.
• It does not know what "the ligand" means. You name the residue. A structure often contains glycerol, buffer ions and hundreds of waters alongside the thing you care about; if you give a code that is not there, the error lists what is.

Typical workflow

1. Collect an experimental complex (RCSB PDB).
2. Collect the compound (PubChem) — this is your template.
3. Ligand extraction → the crystal pose.
4. Structure cleanup on the complex, Remove heterogens → the apo receptor.
5. Small-molecule docking into that receptor, box drawn around the extracted ligand.
6. Pose RMSD, docked poses against the extracted ligand.