How to do molecular docking, step by step

What docking tells you

Molecular docking predicts how a small molecule sits in a protein's binding site (the pose) and estimates how strongly it binds (the score, in kcal/mol for AutoDock Vina). The pose is often useful; the score is a rough estimate for ranking, not a measured binding affinity. Docking is fast, so it is the usual first step of structure based drug design and virtual screening.

Step 1. Choose the protein structure

• Search the RCSB Protein Data Bank for your target and the right organism.
• Prefer crystal structures with good resolution (about 2.5 Å or better).
• Prefer a structure with a ligand already bound in the site you care about; that ligand defines the site and lets you validate the setup.
• Check for missing residues near the site; if a loop lining the pocket is missing, choose another structure.

Step 2. Prepare the protein

• Remove waters, unless one clearly bridges the ligand and the protein.
• Remove buffers, salts and additives; keep cofactors and metal ions that belong in the site.
• Add missing atoms and hydrogens at physiological pH (about 7.4).
• Write the receptor as PDBQT with polar hydrogens and Gasteiger charges.

On ChemServer, tick Auto Fix Protein and Clean Receptors in Batch Docking, or use Protein Prep and Repair for full control. Choose AutoDock Tools (MGLTools) or Meeko for the PDBQT step, matching the papers you compare with.

Step 3. Prepare the ligands

• Start from SMILES or from 3D files.
• Use the protonation state and tautomer expected at pH 7.4: carboxylic acids are usually charged, amines usually protonated.
• Generate a reasonable 3D structure and write it as PDBQT with rotatable bonds defined.

ChemServer builds 3D structures from SMILES and prepares PDBQT files automatically. The format converter does the same if you only need the files.

Step 4. Define the binding site (grid box)

• Known site: centre the box on the co crystal ligand and leave about 4 to 8 Å around it. Boxes of 20 to 25 Å per side suit most drug sized ligands.
• Unknown site: run blind docking over the whole protein, or find pockets first with pocket detection, then dock again in the best pocket.
• Keep boxes at 30 Å per side or less for Vina-GPU.

Step 5. Choose the docking settings

Step 6. Validate by redocking

Dock the co crystal ligand back into its own structure. If the best pose lies within about 2 Å RMSD of the crystal pose, the protein preparation, box and settings work. If not, fix the setup before docking anything new.

Step 7. Read the results

• More negative scores mean stronger predicted binding. Differences under about 1 kcal/mol are within the noise.
• Compare with a reference: the co crystal ligand or a known active docked the same way.
• Look at every top pose: hydrogen bonds, hydrophobic contacts and whether it fills the pocket sensibly. A good score with a strange pose is a warning sign.
• Rescore with GNINA's CNN for a second opinion on pose ranking (a higher CNN score is better).

Step 8. Go further

• Check poses with pose quality and interaction fingerprints.
• Make figures with the protein viewer.
• Test the stability of the best complexes with molecular dynamics.

Common mistakes

• Docking into a structure with a missing loop in the pocket.
• Leaving buffers or crystallisation additives in the site.
• A box that misses part of the pocket, or one so large that the search gets lost.
• Treating the score as an experimental binding energy.
• Reporting results without redocking validation.

Do it on ChemServer

1. Create a free account and open Batch Docking.
2. Type the PDB ID (for example 1HSG) and press Fetch & Scan All.
3. Paste your SMILES, one per line.
4. Choose Reference Ligand (Auto) and pick the co crystal ligand.
5. Press Start and download the scores, poses and Excel report when the run finishes.

Try it with your own molecules

Create a free account