Conformational ensemble engine for
cryptic pocket and allosteric site discovery
Crystal Folding • 2026
"The most druggable pockets are the ones you can't see in the standard structure. You need to watch the protein breathe."
Hidden binding sites that only open in minority conformations. Often the most druggable — deep, hydrophobic cores invisible to single-structure tools.
Bind here, change activity there. Allosteric drugs offer better selectivity and lower toxicity — but require mapping the full conformational landscape.
A pocket "druggable" in one state may be inaccessible in the dominant conformation. You need to know which states are relevant for your target.
Uses Boltz or AlphaFold as structural seed. We build on top of state-of-the-art ML accuracy, not compete with it. Best starting point, guaranteed.
Proprietary physics engine (58K lines) simulates how proteins actually move. Generates 100 diverse, physically realistic conformations per analysis run.
Detects binding pockets across all conformations. Ranks by druggability, flexibility, state support, and chemical accessibility. Cryptic pockets flagged automatically.
Pockets that appear in minority conformations are flagged by high RMSF signature. Often the highest-value targets in difficult disease areas.
Hotspot residues where binding propagates through the structure are identified automatically. Enables allosteric drug design from sequence alone.
100% proprietary code. Zero dependency on Google, Meta, or open-source models for the pocket intelligence layer. Full stack ownership.
Mostly shallow, often hydrophilic. Difficult to achieve selectivity. Miss the protein's breathing modes entirely.
What you see is the dominant conformation. Disease-relevant states, induced-fit pockets, and transient sites are all invisible.
Includes deep hydrophobic pockets, cryptic sites, and allosteric hotspots. Ranked by druggability and state-accessibility.
Know which pockets are open most often, which appear only under specific conditions, and where allosteric opportunities exist.
Paste a protein sequence, fold it in under a minute, and watch the pocket landscape appear across 100 conformations.
Launch Demo →| Capability | AlphaFold 3 | Boltz | Crystal Folding |
|---|---|---|---|
| Structure accuracy | <1Å RMSD | <2Å RMSD | Comparable (Boltz-seeded) |
| Conformational ensemble | Single structure | Single structure | 100 conformations |
| Cryptic pocket detection | No | No | Native — 12/15 validated |
| Allosteric site mapping | No | No | Full hotspot analysis |
| State-specific pocket ranking | No | No | 15/15 on benchmark |
| IP ownership | Google DeepMind | MIT / open | 100% proprietary |
REST + WebSocket API. Submit sequence, get structures + pockets + scores. Integrates into any pipeline in hours.
Conformational ensemble analysis reveals pockets invisible to single-structure methods. More targets, fewer false leads.
Physics engine + proprietary CTT3 algorithm. Not calling AlphaFold API. Genuinely different structural intelligence.
Private session with your science team. Engine internals, pocket detection pipeline, benchmark walkthrough.
Run Crystal Folding on your target protein. Discover pockets. Compare against what existing tools found.
Integrate Crystal Folding into your platform. API access, custom deployment, joint development.
Sergey • Crystal Folding • 2026