How to Do Blind Docking with AutoDock Vina When You Don’t Know the Binding Site

Blind docking means searching a protein’s entire surface for a ligand’s binding site instead of a known pocket. In AutoDock Vina you enclose the whole protein in one large grid box, raise exhaustiveness to compensate for the bigger search space, then cluster the returned poses; the location where several low-energy poses converge is your candidate binding site.

Most docking tutorials assume you already know where the ligand binds, so they hand you the pocket coordinates. Real projects are not that tidy. You have a fresh structure, a candidate molecule, and no annotated site. This guide from the StemSkills Lab team (10+ years in structural bioinformatics, drug design, and multiscale molecular modeling) walks an MSc student through blind (whole-surface) docking in AutoDock Vina, from sizing the box to reading the poses, so you can locate a binding site without prior structural knowledge. For where this sits in the wider workflow, start from our molecular docking pillar guide.

What is blind docking, and when do you need it?

Blind docking is docking a ligand against a receptor without telling the program where the binding site is. Instead of a small box around a known pocket, you define a search space large enough to cover the whole protein and let the search sample the entire surface. The term comes from Hetenyi and van der Spoel, who introduced the approach for finding peptide and drug binding sites without prior knowledge of their location (Hetenyi and van der Spoel, 2002, Protein Science). Their follow-up extended it to drug-sized compounds on proteins with up to a thousand residues.

You need blind docking whenever the pocket is genuinely unknown: a newly solved or predicted structure with no bound ligand, an allosteric site you have not yet mapped, or a case where you want an unbiased check that the “obvious” pocket really is the best one. If a co-crystallised ligand, a documented catalytic site, or a cavity-detection result already tells you where to look, skip blind docking and go straight to a focused run around that pocket. Blind docking is a locator, not a substitute for careful site-specific docking.

Why use AutoDock Vina for blind docking?

AutoDock Vina is a good fit because it is fast, free, and calculates its grid maps internally, so a very large box does not force you to precompute enormous AutoGrid map files by hand. Speed matters here: a whole-protein search is far more expensive than a focused one, and you will often repeat it. In the paper that introduced the program, the authors reported that Vina runs about two orders of magnitude faster than AutoDock4 while “significantly improving the accuracy of the binding mode predictions” (Trott and Olson, 2010, Journal of Computational Chemistry).

Vina is also actively maintained. Version 1.2 added Python bindings, batch docking, and support for multiple ligands, which helps when you want to script a blind screen (Eberhardt et al., 2021, Journal of Chemical Information and Modeling). The official documentation and installation instructions live at autodock-vina.readthedocs.io.

How do you size a grid box to cover the whole protein?

The box is the only thing that changes between focused and blind docking, so get it right. In a Vina configuration file you define the search space with a center (center_x, center_y, center_z) and edge lengths in Angstrom (size_x, size_y, size_z). For blind docking, the center should be the geometric center of the receptor and the edges should span the protein plus a margin.

1. Load the prepared receptor in AutoDockTools (MGLTools), then open Grid > Grid Box.
2. Increase the number of points along each axis until the yellow-green box encloses the entire protein, plus roughly 8 to 10 Angstrom of clearance on every side so surface pockets are not clipped.
3. Read off the center coordinates and the number of points the GUI reports.
4. Convert points to the Angstrom size Vina expects: size = number_of_points x spacing, where the default spacing is 0.375 Angstrom. So 120 points give a 45 Angstrom edge.

For a large protein where 0.375 Angstrom spacing cannot reach across the structure, increase the spacing (for example to 0.5 Angstrom) so the same number of points spans a wider box, or split the surface into two or three overlapping boxes and dock into each. Do not simply shrink the box, because a box that does not enclose the whole protein is no longer a blind search.

How do you run blind docking in AutoDock Vina?

First prepare both partners as PDBQT files, exactly as in focused docking. Use prepare_receptor (or AutoDockTools) for the protein and Meeko or AutoDockTools for the ligand:

1. Prepare the receptor: prepare_receptor -r receptor.pdb -o receptor.pdbqt
2. Prepare the ligand as ligand.pdbqt with its rotatable bonds defined.
3. Write a configuration file, blind.conf, with the whole-protein box and a raised search effort.
4. Run Vina against the config.

A blind configuration file looks like this:

• receptor = receptor.pdbqt
• ligand = ligand.pdbqt
• center_x = 12.5, center_y = 8.0, center_z = -4.2
• size_x = 66, size_y = 60, size_z = 58
• exhaustiveness = 32
• num_modes = 20
• energy_range = 4
• seed = 12345

Then run:

1. vina --config blind.conf --out blind_out.pdbqt

Setting a fixed seed makes the run reproducible so you can compare repeats fairly. The num_modes = 20 line asks Vina to return up to 20 poses instead of the default 9, which you need in blind docking because the interesting signal is where those poses land across the surface, not just the single top score. The defaults for reference are exhaustiveness 8, num_modes 9, and energy_range 3, documented in the Vina manual.

Why do exhaustiveness and box size matter more in blind docking?

In a focused run the search space is small, so even the default exhaustiveness of 8 samples it thoroughly. A whole-protein box can be tens of times larger in volume, which spreads the same search effort thinner and makes it easier to miss the true global minimum. Raising exhaustiveness increases the number of independent search runs, which improves your odds of finding the real site at the cost of more compute time. A common practical range for blind docking is 32 or higher, repeated with a few different seeds.

Box size cuts both ways. Too small and you clip off part of the surface, so the real pocket may never be sampled. Too large and you dilute the search and slow it down. Vina will print a warning when the search space exceeds 27000 cubic Angstrom, which almost always happens in blind docking. The warning is expected, not an error, but it is Vina telling you the search is now large enough that you should raise exhaustiveness to keep the results reliable.

How do you find the real binding site from the poses?

Blind docking does not hand you a labeled site. It hands you a set of poses scattered over the protein, and your job is to find where they agree. The signal is convergence: the location where multiple independent, low-energy poses cluster together is the most likely binding site, more so than any single best-scoring pose sitting alone.

1. Split the output into separate poses (Vina writes them into one multi-model blind_out.pdbqt).
2. Note each pose’s affinity score and its approximate location on the protein.
3. Group poses that occupy the same region. A tight cluster of several poses within a few Angstrom of each other is a candidate pocket.
4. Cross-check the top cluster against any biological knowledge you do have: known catalytic residues, conserved regions, or a cavity a pocket-detection tool such as fpocket flags.

Treat a lone top score with suspicion. A single pose with a good number but no neighbours can be an artifact of a shallow surface groove or a box edge. Convergence plus a plausible pocket is far stronger evidence than score alone. Where binding energy fits into the larger analysis pipeline is covered on our computational biology skills roadmap.

When should you switch to a focused re-dock?

Once a cluster identifies a candidate pocket, blind docking has done its job. Switch to a focused re-dock: build a small box (roughly 20 to 25 Angstrom per side) centered on the candidate site and run again with the default or a moderate exhaustiveness. The focused run gives you tighter, more reliable poses and scores for that pocket because the search effort is now concentrated where it matters. Use the blind run to locate, and the focused run to characterise. Reporting binding poses or scores straight from a blind run, without this refinement step, is a common beginner mistake.

How do the main blind docking approaches compare?

You do not have to do whole-surface docking by brute force. The table compares four practical routes so you can pick by your constraints.

If you want a hands-off result, the CB-Dock2 server automates cavity detection and docking in one step (Liu et al., 2022, Nucleic Acids Research). If you want to understand what the search is actually doing, run the Vina whole-protein route yourself at least once before you trust a server.

Troubleshooting blind docking

The failures below are the ones students hit most often, with the fix for each.

• “WARNING: The search space volume > 27000 Angstrom^3”: Expected in blind docking. It is not an error. Raise exhaustiveness to 32 or more, or split the surface into overlapping boxes.
• No poses returned, or a “could not parse” message: The ligand or receptor PDBQT is malformed. Re-run preparation, and confirm the ligand has valid atom types and defined rotatable bonds.
• All top poses sit on one shallow patch or hug the box edge: The box is off-center or too small, so the search is biased. Recompute the center from the receptor coordinates and add margin so the protein is fully enclosed.
• Poses are scattered with no convergence: The search is undersampled. Increase exhaustiveness and num_modes, run three to five different seeds, and cluster the pooled poses.
• The run is extremely slow: The box is very large. Use a cavity detector (fpocket or CB-Dock2) to shrink the search to a few candidate pockets, then dock into those.

Frequently asked questions

Is blind docking as accurate as focused docking?

No. Blind docking is a locator: use it to find the likely site, then re-dock focused for reliable poses and scores. The larger search space makes single blind poses less trustworthy on their own.

What exhaustiveness should I use for blind docking?

Higher than the default of 8, because the box is much larger. Many groups use 32 or more and repeat the run with several seeds, then judge the result by pose convergence rather than a single score.

Can I do blind docking without installing anything?

Yes. The CB-Dock2 web server automates cavity detection and Vina docking in the browser. It is a good sanity check, though you give up fine control over parameters compared with running Vina yourself.

How large should the grid box be?

Large enough to enclose the entire protein plus about 8 to 10 Angstrom of clearance on each side. For very large structures, increase the grid spacing or split the surface into overlapping boxes rather than clipping the protein.

How do I know I have found the real binding site?

Look for convergence: several low-energy poses clustering at one location, ideally matching known functional residues or a cavity flagged by a pocket-detection tool. A lone high score with no supporting poses is weak evidence.