Cryptic and allosteric pockets in docking

Open a docking program, load a crystal structure, define the binding site, and dock. The unspoken assumption is that the pocket you are docking into actually exists in the structure in front of you. For a large fraction of disease-relevant proteins, it does not. The pocket only appears when a ligand pushes the protein into a conformation no apo crystal ever captured. Miss that, and your rigid-receptor run confidently reports that nothing binds to a target that is, in fact, druggable.

Cryptic versus allosteric: two different ideas

The terms get used loosely, so it is worth separating them.

• An allosteric site is defined by location: it is a pocket distinct from the orthosteric (substrate or ATP) site, where a ligand can modulate activity from a distance. Allosteric sites can be perfectly visible in the resting structure.
• A cryptic site is defined by visibility: it is a pocket that is absent or too shallow to detect in the ground-state structure and only forms through a conformational change, often induced by the ligand itself. A cryptic site may be orthosteric or allosteric.

The hard case for docking is the cryptic site, because the geometry you need simply is not in the input coordinates.

Why this matters: targets that hide their pockets

Cimermancic et al. (2016), building the CryptoSite predictor, estimated that accounting for cryptic sites raises the fraction of the disease-associated human proteome considered druggable from roughly 40% to nearly 80%. The PocketMiner work (Meller et al., 2023) went further, arguing from simulation that over half of proteins that look pocketless in available structures likely harbor cryptic pockets. These are not edge cases.

Two clinically validated examples make the point. The KRAS switch-II pocket that every covalent G12C inhibitor exploits is essentially invisible in early KRAS structures; it opens only in specific states, which is part of why KRAS was called undruggable for thirty years. And asciminib, the BCR-ABL1 inhibitor, works by binding the myristoyl pocket, an allosteric site that the kinase normally uses for autoregulation rather than the ATP site every prior TKI targeted (Wylie et al., 2017). Dock asciminib into an ATP-site-only model and you learn nothing.

Why rigid docking fails here

Standard docking holds the receptor fixed and samples ligand poses against it. If the input is an apo or closed-state structure, the cryptic pocket is collapsed: there is no cavity to dock into, sidechains occlude the space, and the scoring function rewards poses that sit on the wrong surface. The result is not a useful negative; it is an artifact of using the wrong conformation. The protein never got the chance to open.

What to do instead

• Dock against the right conformation. If a ligand-bound (holo) structure exists where the pocket is open, use it. The pocket geometry from a co-crystal with any ligand is usually a far better receptor than the apo form.
• Ensemble docking. Generate multiple receptor conformations and dock against all of them, keeping the best score per ligand. Conformations can come from multiple crystal structures, from molecular dynamics snapshots, or from enhanced-sampling simulations designed to open transient pockets.
• Induced-fit and flexible-receptor docking. Let binding-site sidechains, or in aggressive protocols the backbone, relax in response to the ligand. This recovers small pocket openings that rigid docking cannot.
• Predict where the pockets are first. Tools like CryptoSite and PocketMiner flag residues likely to participate in a cryptic pocket, telling you whether to invest in conformational sampling before you waste a rigid-docking campaign.
• Mind AlphaFold. A predicted structure typically gives you a single, usually ground-state, conformation. It is a starting point for sampling, not a guarantee that a cryptic or allosteric pocket will be present and open.

The practical takeaway

Before you trust a docking result, ask whether the pocket you docked into is the pocket the drug would actually use. For well-behaved orthosteric sites the answer is usually yes. For allosteric programs, for cryptic-pocket targets, and for anything historically called undruggable, a single rigid structure is the wrong tool, and the fix is conformational sampling rather than a better scoring function.

Try the docking yourself

The cleanest way to feel the conformation problem is to dock the same ligand against more than one receptor state. Open Studio and use the ensemble docking option to run a ligand against multiple conformations of a target at once, then compare the per-conformation scores. A compound that scores poorly against a closed state and well against an open one is showing you exactly the cryptic-pocket effect this post is about. Liganx is molecular docking online: free, browser-based, and set up so you can test conformational dependence without standing up an MD pipeline. If you want to try molecular docking against multiple receptor states without a local install, that is the fastest path.

Primary sources

• Cimermancic P, et al. CryptoSite: Expanding the Druggable Proteome by Characterization and Prediction of Cryptic Binding Sites. J Mol Biol 428, 709-719 (2016). doi:10.1016/j.jmb.2016.01.029
• Meller A, et al. Predicting locations of cryptic pockets from single protein structures using the PocketMiner graph neural network. Nat Commun 14, 1177 (2023). doi:10.1038/s41467-023-36699-3
• Wylie AA, et al. The allosteric inhibitor ABL001 enables dual targeting of BCR-ABL1. Nature 543, 733-737 (2017). doi:10.1038/nature21702