Lacuna

Cryptic binding pocket discovery via conformational ensemble analysis.

Most protein structure predictors (AlphaFold, Boltz, Chai) give you one static structure. But ~70% of disease-relevant proteins are considered "undruggable" not because they're biologically intractable - it's because no pocket is visible in their ground state. K-Ras was "undruggable" for 30 years until a transient cryptic pocket was found in its switch-II region. That pocket now backs sotorasib and adagrasib.

Lacuna finds those pockets. It generates a conformational ensemble from any input structure, detects pockets per conformer, and clusters them across the ensemble to surface sites that only appear transiently - ranked by a continuous crypticity score.

lacuna discover kras.pdb --conformers 20 --emit-boltz-constraints --emit-vina-boxes

Install

pip install lacuna-pockets

Optional backends (better conformational sampling):

pip install "lacuna-pockets[openmm]"   # 100ps implicit-solvent MD
pip install "lacuna-pockets[boltz]"    # Boltz-2 diffusion sampling (experimental, GPU)
pip install "lacuna-pockets[all]"      # everything

Requires Python 3.10+.

---

Quick start

CLI

# Discover pockets with defaults (NMA backend - physically grounded, no GPU needed)
lacuna discover protein.pdb --conformers 20

# Filter and limit output
lacuna discover protein.pdb --min-druggability 0.5 --min-persistence 0.3 --top 5

# Analyze a homodimer - detects pockets at the dimer interface (e.g. Caspase-1, IDH1)
# Reads BIOMT records from PDB; for best results use the biological assembly download from RCSB
lacuna discover protein.pdb --homodimer --conformers 20

# Optional Boltz-2 backend (experimental - see the Backends note)
lacuna discover protein.pdb --backend boltz --conformers 30

# Emit all docking file formats
lacuna discover protein.pdb --emit-boltz-constraints --emit-vina-boxes --emit-pocket-pdbs

# Generate docking files from a previous report
lacuna dock-prep kras_lacuna/pocket_report.json kras.pdb --format all

Python API

from lacuna import load_structure, detect_pockets, cluster_pockets
from lacuna.ensemble.nma_backend import NMABackend
from lacuna.io.structure import coords_array
from lacuna.io.writers import write_report, write_boltz_constraint

structure = load_structure("protein.pdb")
backend = NMABackend(seed=42)
coord_sets = backend.generate("protein.pdb", n_conformers=20)

all_coords = [coords_array(structure)] + coord_sets
pocket_lists = []
for ci, coords in enumerate(all_coords):
    pockets = detect_pockets(coords, structure)
    for p in pockets:
        p.conformer_idx = ci
    pocket_lists.append(pockets)

clusters = cluster_pockets(pocket_lists, n_conformers=len(all_coords))
for c in clusters[:5]:
    print(f"Rank {c.rank}  druggability={c.druggability:.3f}  "
          f"persistence={c.persistence:.0%}  cryptic={c.cryptic}")
    print(f"  Residues: {', '.join(c.lining_residues[:5])}")

---

How it works

1. Ensemble generation - Generate N conformers via elastic network model normal mode analysis (built-in, default), OpenMM implicit-solvent MD, or experimental Boltz-2 diffusion sampling
2. Pocket detection - Grid-based alpha-point analysis per conformer: compute distance transform, find local maxima within the 1.4-5.5 Å interaction zone, cluster nearby alpha-points into pocket candidates
3. Cross-ensemble clustering - Greedy centroid merging clusters corresponding pockets across all conformers
4. Druggability scoring - Gaussian volume reward centered at 300 ų + enclosure + hydrophobicity + aromaticity (Halgren 2009), scored in each conformer
5. Crypticity scoring & ranking - Each site gets a continuous crypticity score (how much it opens relative to the apo state × druggability when open) and is flagged cryptic: true if present in <90% of conformers. Pockets are ranked by crypticity by default; --rank-by druggability is available for always-open / orthosteric sites

---

Outputs

File	Description
pocket_report.json	Ranked pocket metadata: centroid, volume + apo→open range, druggability, crypticity, persistence, lining residues
pocket_N_site.pdb	Pseudoatom PDB for PyMOL/ChimeraX visualization
pocket_N_constraint.yaml	Boltz YAML - add a SMILES and run boltz predict to dock into this site
pocket_N_vina.conf	AutoDock Vina / Gnina / QuickVina box config

---

Backends

Backend	Install	Quality	Speed	Notes
nma	built-in	good	~0.1s/conf	Elastic-network normal modes (default)
openmm	lacuna[openmm]	good	~2s/conf	Implicit-solvent MD, 100ps
boltz	lacuna[boltz]	experimental	~100s/protein (GPU)	Diffusion sampling, noisy (see note below)
random	built-in	baseline	~0.04s/conf	Gaussian backbone perturbation

Auto-selection order: boltz → openmm → nma → random. On a plain pip install lacuna, the NMA backend runs automatically.

The nma backend samples physically meaningful collective motions - the same hinge-bending and breathing modes that open cryptic pockets in nature - without requiring a GPU or force field. It is the zero-dependency default.

Boltz backend status (honest note). The boltz backend runs Boltz-2 diffusion sampling on a GPU, but it currently predicts each conformer de novo from sequence (not partial diffusion from the input structure), which yields structurally divergent, noisy ensembles (150-300+ pocket clusters vs NMA's ~35). In GPU benchmarking it did not reliably improve cryptic detection over NMA. A proper apo-templated integration with sequence-based residue mapping is planned; until then, NMA is the recommended backend.

---

Benchmarks

7 / 22 cryptic pockets localized (32%, size-robust criterion; NMA backend, crypticity ranking, 20 conformers).

This curated result is cross-validated on two further independent datasets - PocketMiner 31% and CryptoBench 18% (the largest and hardest) - see Independent validation below.

Size-robust success criterion (top-5 pockets): a pocket whose lining residues reach a Jaccard overlap ≥ 0.25 with the known ligand-contact site (Jaccard = |found ∩ known| / |found ∪ known|), or whose center is within 4 Å of the site centroid. Lining residues use a true atomic-contact definition (any residue with an atom within 5 Å of the detected cavity). Reproduce with python benchmarks/cryptic_benchmark.py --category cryptic.

Why the number is lower than you may have seen before - please read. Earlier releases led with plain recall (|found ∩ known| / |known| ≥ 30%), which gave 13/22 (59%) but is size-gameable: a large pocket engulfs a small known site and scores high recall while sitting nowhere near it. A learned re-ranker confirmed this by reaching 84% on recall purely by ranking pockets on raw volume. The headline now leads with size-robust Jaccard (or a ≤4 Å centroid hit) instead, which roughly halves the numbers but is the one we can defend on held-out data; only 2/22 pass the strict centroid test alone. Both criteria print side by side in every benchmark script.

Cryptic pockets - 7 / 22 (32%)

The remaining gap is mostly sampling, not ranking: at top-20 the size-robust score only rises to 10/22, so 12 of the misses are never localized at all rather than found-but-mis-ranked. The hard cases split into oligomeric-interface pockets (invisible to single-chain analysis, partly addressable with --homodimer) and large-rearrangement sites that need sampling beyond elastic-network modes.

Full 22-target breakdown, per-target Jaccard/recall/rank →

Independent validation - three benchmarks

Measured on three independent datasets (NMA + crypticity, top-5). Both criteria are reported: the size-robust headline (Jaccard ≥ 0.25 or ≤ 4 Å centroid) and the legacy recall number (≥ 30% recall or ≤ 4 Å centroid) that earlier releases led with.

Benchmark	N	Size-robust	Legacy recall	Notes
Curated apo/holo set (this repo)	22	32%	59%	literature cryptic pairs
PocketMiner	45	31%	60%	per-residue cryptic labels
CryptoBench (test fold)	180	18%	49%	largest & most diverse
CryptoBench (train, generalization)	749	13%	50%	held out from all tuning

Datasets: PocketMiner (Meller et al. 2023, Nat. Commun.); CryptoBench (Vavra et al. 2024, Bioinformatics).

The two curated/field-standard sets converge at ~31-32% under the size-robust metric; CryptoBench - the field's largest cryptic set (1107 structures; 180 of its 222-structure held-out test fold evaluated here) - is harder at 18%. The legacy recall column roughly doubles every number: that gap is the size-gaming headroom the recall metric leaves open (a large pocket covers a small known site without being localized on it), which is exactly why the size-robust number is the one we lead with.

Generalization. To check that these numbers are not an artifact of the specific test fold, we scored all 749 CryptoBench train-fold structures, which were never used in any tuning: 13% size-robust (95% CI 10-16%) and 50% legacy. Both are statistically consistent with the test fold (overlapping confidence intervals), so the honest headline holds up on genuinely unseen pockets. Reproduce (each script prints both criteria):

python benchmarks/pocketminer_benchmark.py    # PocketMiner (auto-downloads)
python benchmarks/cryptobench_benchmark.py    # CryptoBench test fold (auto-downloads, ~10 min)

Limitations and scaling (this ceiling is a compute problem)

The honest ceiling above (about 32% on the curated set, 13 to 18% on CryptoBench under the size-robust metric) is set by conformational sampling, not by ranking or pocket detection. At a top-20 cutoff the numbers rise only slightly, which means the pocket is usually not found-but-mis-ranked; it is simply never sampled in an open state.

The remaining misses concentrate in the large-collective-motion classes, hinge and oligomeric-interface openings. The default NMA backend is harmonic and cannot generate those motions. Molecular dynamics can in principle, but a cryptic opening is a rare event: in our tests, short trajectories (0.5 to 3 ns) essentially never caught one, and enhanced-temperature MD, metadynamics along an apo-derived collective variable, and SWISH scaled-water MD were all null at the sampling a single workstation affords (see benchmarks/experiments/).

Raising this ceiling is a compute problem, not a missing algorithm. Reliably observing rare openings needs orders of magnitude more MD sampling: tens to hundreds of nanoseconds per trajectory across dozens of independent replicas, aggregating microseconds per target, the scale used by the successful literature (for example PocketMiner's ~940,000 simulation windows and Folding@home-style datasets). As an anchor, the development GPU runs a small protein at roughly 300 ns/day; sampling rare openings across the 22 to 885 benchmark targets, with the frontier proteins several times slower, is tens to hundreds of GPU-days. That is cluster or cloud GPU scale. With that budget, the same pipeline could be driven by long multi-replica MD (or cosolvent MD) to attack the hinge and interface classes that are out of reach on a single machine.

Crypticity ranking (the default) intentionally de-prioritizes always-open sites; for orthosteric / general pocket finding use --rank-by druggability (orthosteric controls: 3/6 →, a known relative weakness of this tight-contact pipeline).

Crypticity score

Every reported pocket carries a continuous crypticity score in [0, 1] - the conformational-selection signature of a cryptic site, defined as how much the pocket opens relative to the apo/input structure × how druggable it is once open:

opening    = (max_volume − apo_volume) / max_volume        # 1.0 if absent in the apo state
crypticity = opening × peak_open_state_druggability

A constitutive pocket already formed in the input structure scores ≈ 0; a pocket absent in the apo structure that opens into a druggable cavity scores near 1. Ranking by crypticity is the default and recovers the most cryptic targets. The JSON report also includes per-pocket volume dynamics (apo_volume_A3, volume_range_A3) and max_druggability.

--rank-by selects how pockets are ordered: crypticity (default, most cryptic sites first, 12/20 cryptic pass), druggability, balanced, or persistence. NMA runtime is sub-second to ~8s per protein on a laptop CPU, no GPU required. Full ranking-strategy ablation and per-size timing →

Head-to-head: Lacuna vs fpocket

fpocket detects pockets on a single static structure. Lacuna generates a conformational ensemble to expose sites that only become visible once the protein moves. Run side by side on the same structures under the same size-robust criterion (top-5, Jaccard ≥ 0.25 or centroid ≤ 4 Å), the two tools catch largely different pockets:

Set	fpocket	Lacuna	Combined (either)
CryptoBench test fold (n=180)	28% (51/180)	16% (29/180)	38% (68/180)
Curated cryptic set (n=22)	18% (4/22)	18% (4/22)	36% (8/22)

On CryptoBench, Lacuna independently recovers 17 pockets that fpocket misses entirely: sites invisible to single-structure geometric detection that only open once the ensemble samples them. On the curated set, the hit lists don't overlap at all: fpocket catches T4 lysozyme's buried cavity and PTP1B's allosteric site, while Lacuna catches the BCL-2/BCL-XL BH3 grooves, MDM2's p53-binding cleft, and IL-2's helix pocket, sites that open through conformational change rather than being present in one fixed geometry. Running both and taking the union beats either tool alone on both benchmarks.

Reproduce:

python benchmarks/compare_fpocket.py                            # 22 curated cryptic targets vs fpocket
python benchmarks/compare_fpocket_cryptobench.py --folds test   # CryptoBench test fold vs fpocket (~7 min)
python benchmarks/cryptic_benchmark.py --category cryptic       # 22 cryptic targets, NMA (~4 min)
python benchmarks/cryptic_benchmark.py --quick                  # 10 conformers, faster
python benchmarks/cryptic_benchmark.py --category cryptic --rank-by druggability  # ablation
python benchmarks/cryptic_benchmark.py --category cryptic --top-n 20              # detection ceiling

---

Example: K-Ras switch-II

# Download K-Ras apo (from RCSB); NMA backend (default) recovers switch-II at rank 3
lacuna discover 4OBE.pdb \
    --conformers 20 \
    --emit-boltz-constraints \
    --output kras_pockets/

# pocket_0_constraint.yaml is ready - add your SMILES:
#   - ligand:
#       id: L
#       smiles: YOUR_SMILES_HERE
boltz predict kras_pockets/pocket_0_constraint.yaml

See examples/kras_cryptic.py for a full annotated Python workflow.

---

Input formats

Accepts PDB or mmCIF from any predictor or database:

• AlphaFold 2 / AlphaFold 3
• Boltz-1 / Boltz-2
• Chai-1
• RCSB PDB
• ESMFold, RoseTTAFold, OpenFold, etc.

---

Citation

If you use Lacuna in published research, please cite:

Moore, C. (2026). Lacuna: Cryptic Binding Pocket Discovery via Conformational Ensemble Analysis. https://github.com/mooreneural/lacuna

BibTeX:

@software{moore2026lacuna,
  author  = {Moore, Clayton W.},
  title   = {Lacuna: Cryptic Binding Pocket Discovery
             via Conformational Ensemble Analysis},
  year    = {2026},
  url     = {https://github.com/mooreneural/lacuna},
  version = {0.3.1}
}

Methodology papers Lacuna builds on:

• Atilgan et al. (2001) Biophys. J. 80(1):505-515 - Anisotropic Network Model (NMA backend)
• Halgren (2009) J. Chem. Inf. Model. 49(2):377-389 - SiteMap druggability scoring
• Le Guilloux et al. (2009) BMC Bioinformatics 10:168 - fpocket alpha-sphere approach
• Schmidtke & Barril (2010) J. Med. Chem. 53(15):5858-5867 - enclosure scoring

---

License

GNU AGPL-3.0-or-later - free to use, study, modify, and share. The AGPL's copyleft requires that if you distribute a modified version, or run a modified version as a network/hosted service, you make the complete corresponding source available under the same license.

A separate commercial license removes the AGPL copyleft obligation (for embedding Lacuna in closed-source products or hosted services without releasing your own source) and adds warranty, indemnification, support SLAs, and custom development.

Versions ≤ 0.1.2 were released under the MIT License and remain available under those terms. AGPL-3.0 applies from version 0.2.0 onward.