lacuna-pockets 0.1.0__tar.gz

lacuna_pockets-0.1.0/.github/workflows/ci.yml

@@ -0,0 +1,33 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ["3.10", "3.11", "3.12"]
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install -e ".[dev]"
+
+      - name: Run tests
+        run: python -m pytest tests/ -v --tb=short
+
+      - name: Check import
+        run: python -c "import lacuna; print('lacuna', lacuna.__version__)"

lacuna_pockets-0.1.0/.github/workflows/publish.yml

@@ -0,0 +1,39 @@
+name: Publish to PyPI
+
+on:
+  release:
+    types: [published]
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+
+      - name: Build
+        run: |
+          python -m pip install build
+          python -m build
+
+      - uses: actions/upload-artifact@v4
+        with:
+          name: dist
+          path: dist/
+
+  publish:
+    needs: build
+    runs-on: ubuntu-latest
+    environment: pypi
+    permissions:
+      id-token: write  # required for OIDC trusted publishing
+    steps:
+      - uses: actions/download-artifact@v4
+        with:
+          name: dist
+          path: dist/
+
+      - uses: pypa/gh-action-pypi-publish@release/v1

lacuna_pockets-0.1.0/.gitignore

@@ -0,0 +1,14 @@
+__pycache__/
+*.pyc
+*.pyo
+.pytest_cache/
+*.egg-info/
+dist/
+build/
+.venv/
+venv/
+*_lacuna/
+*.npz
+*.npy
+*.log
+benchmarks/pdb_cache/

lacuna_pockets-0.1.0/CITATION.cff

@@ -0,0 +1,33 @@
+cff-version: 1.2.0
+message: "If you use Lacuna in your research, please cite it as below."
+type: software
+title: "Lacuna: Cryptic Binding Pocket Discovery via Conformational Ensemble Analysis"
+authors:
+  - family-names: Moore
+    given-names: Clayton W.
+version: 0.1.0
+date-released: 2026-06-02
+url: "https://github.com/mooreneural/lacuna"
+repository-code: "https://github.com/mooreneural/lacuna"
+license: "LicenseRef-Lacuna-NonCommercial"
+abstract: >
+  Lacuna discovers cryptic binding pockets in proteins by generating a
+  conformational ensemble, detecting pockets per conformer, clustering
+  across the ensemble, and ranking by persistence and druggability.
+  Validated on 20 cryptic-pocket targets (14/20, 70%), matching the
+  CryptoSite benchmark. Supports RandomBackend (CPU, ~1–4 s/protein)
+  and Boltz-2 partial diffusion for harder targets.
+keywords:
+  - drug-discovery
+  - cryptic-pockets
+  - protein-structure
+  - conformational-ensemble
+  - computational-biology
+preferred-citation:
+  type: software
+  title: "Lacuna: Cryptic Binding Pocket Discovery via Conformational Ensemble Analysis"
+  authors:
+    - family-names: Moore
+      given-names: Clayton W.
+  year: 2026
+  url: "https://github.com/mooreneural/lacuna"

lacuna_pockets-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Clayton Moore
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

lacuna_pockets-0.1.0/LICENSE_COMMERCIAL

@@ -0,0 +1,42 @@
+Lacuna Commercial License Agreement
+
+Copyright (c) 2026 Clayton Moore. All rights reserved.
+
+This Commercial License Agreement ("Agreement") governs use of the Lacuna
+software ("Software") by for-profit entities and organizations requiring terms
+beyond those provided by the MIT license.
+
+Lacuna is available at no cost under the MIT License for any use. This
+Commercial License provides additional terms — warranties, indemnification,
+support SLAs, and custom integration rights — for organizations that require
+them.
+
+COMMERCIAL LICENSE PROVIDES:
+
+1. INDEMNIFICATION. Licensor will defend and indemnify Licensee against third-
+   party intellectual property claims arising from Licensee's authorized use of
+   the Software, subject to the terms of a separately executed agreement.
+
+2. WARRANTY. Licensor provides a limited warranty that the Software will
+   perform materially as described in the documentation for 90 days from
+   delivery, with remedies defined in the separately executed agreement.
+
+3. SUPPORT. Licensee receives priority bug fixes, direct email/phone support,
+   and guaranteed response times as specified in the separately executed
+   agreement.
+
+4. CUSTOM DEVELOPMENT. Licensor will develop custom features, integrations, or
+   validated workflows for Licensee under a statement of work.
+
+5. DEPLOYMENT RIGHTS. For organizations requiring deployment in air-gapped or
+   restricted environments, Licensor provides deployment assistance and
+   site-specific licensing under this agreement.
+
+TO OBTAIN A COMMERCIAL LICENSE:
+Contact: claytonwaynemoore@gmail.com
+
+Subject line: "Lacuna Commercial License Inquiry"
+
+The MIT License remains the operative license for all use not covered by a
+separately executed commercial agreement. Nothing in this document restricts
+any rights granted by the MIT License.

lacuna_pockets-0.1.0/PKG-INFO

@@ -0,0 +1,341 @@
+Metadata-Version: 2.4
+Name: lacuna-pockets
+Version: 0.1.0
+Summary: Cryptic binding pocket discovery via conformational ensemble analysis
+Project-URL: Homepage, https://github.com/mooreneural/lacuna
+Project-URL: Repository, https://github.com/mooreneural/lacuna
+Project-URL: Issues, https://github.com/mooreneural/lacuna/issues
+Author: Clayton Moore
+License: MIT License
+        
+        Copyright (c) 2026 Clayton Moore
+        
+        Permission is hereby granted, free of charge, to any person obtaining a copy
+        of this software and associated documentation files (the "Software"), to deal
+        in the Software without restriction, including without limitation the rights
+        to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+        copies of the Software, and to permit persons to whom the Software is
+        furnished to do so, subject to the following conditions:
+        
+        The above copyright notice and this permission notice shall be included in all
+        copies or substantial portions of the Software.
+        
+        THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+        IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+        FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+        AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+        LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+        OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+        SOFTWARE.
+License-File: LICENSE
+License-File: LICENSE_COMMERCIAL
+Keywords: cheminformatics,cryptic-pockets,drug-discovery,protein-structure
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Scientific/Engineering :: Bio-Informatics
+Classifier: Topic :: Scientific/Engineering :: Chemistry
+Requires-Python: >=3.10
+Requires-Dist: biopython>=1.81
+Requires-Dist: click>=8.1
+Requires-Dist: numpy>=1.24
+Requires-Dist: rich>=13.0
+Requires-Dist: scipy>=1.10
+Provides-Extra: all
+Requires-Dist: boltz>=0.4; extra == 'all'
+Requires-Dist: openmm>=8.0; extra == 'all'
+Requires-Dist: pdbfixer>=1.9; extra == 'all'
+Provides-Extra: boltz
+Requires-Dist: boltz>=0.4; extra == 'boltz'
+Provides-Extra: dev
+Requires-Dist: mypy; extra == 'dev'
+Requires-Dist: pytest-cov; extra == 'dev'
+Requires-Dist: pytest>=7.0; extra == 'dev'
+Requires-Dist: ruff; extra == 'dev'
+Provides-Extra: openmm
+Requires-Dist: openmm>=8.0; extra == 'openmm'
+Requires-Dist: pdbfixer>=1.9; extra == 'openmm'
+Description-Content-Type: text/markdown
+
+# 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 druggability and persistence.
+
+```
+lacuna discover kras.pdb --conformers 20 --emit-boltz-constraints --emit-vina-boxes
+```
+
+---
+
+## Install
+
+```bash
+pip install lacuna-pockets
+```
+
+**Optional backends** (better conformational sampling):
+```bash
+pip install "lacuna-pockets[openmm]"   # 100ps implicit-solvent MD
+pip install "lacuna-pockets[boltz]"    # Boltz-2 partial diffusion (best quality, GPU recommended)
+pip install "lacuna-pockets[all]"      # everything
+```
+
+Requires Python 3.10+.
+
+---
+
+## Quick start
+
+### CLI
+
+```bash
+# 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
+
+# Use Boltz-2 partial diffusion for highest-quality sampling
+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
+
+```python
+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), OpenMM implicit-solvent MD, or Boltz-2 partial diffusion at varying noise levels
+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)
+5. **Cryptic flagging** - Pockets present in <90% of conformers are marked `cryptic: true`
+
+---
+
+## Outputs
+
+| File | Description |
+|------|-------------|
+| `pocket_report.json` | Ranked pocket metadata: centroid, volume, druggability, 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 | Anisotropic Network Model normal mode analysis - hinge bending, breathing, twist motions |
+| `openmm` | `lacuna[openmm]` | good | ~2s/conf | 100ps Langevin MD, GBn2 implicit solvent |
+| `boltz` | `lacuna[boltz]` | best | ~30s/conf (GPU) | Boltz-2 partial diffusion at varying noise fractions |
+| `random` | built-in | baseline | ~0.04s/conf | Correlated 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 replaces `random` as the zero-dependency default. For large-scale loop rearrangements or very deep cryptic sites, `boltz` remains the best option.
+
+---
+
+## Benchmarks
+
+**14 / 20 cryptic pockets detected (70%, NMABackend, 20 conformers)** - matching the CryptoSite published benchmark rate on a statistically defensible N=20 set.
+
+Success criterion: pocket centroid within 4 Å of the known binding-site centroid (field standard) **or** ≥30% residue overlap in top-5 pockets.
+
+### Cryptic pockets - 14 / 20 (70%)
+
+| Protein | Apo PDB | Drug target | Overlap | Time |
+|---------|---------|-------------|---------|------|
+| ✅ T4L L99A hydrophobic cavity | 1L90 | - | 100% | 0.9s |
+| ✅ K-Ras switch-II pocket | 4OBE | sotorasib / adagrasib | 93% | 0.9s |
+| ✅ MDM2 p53-binding cleft | 1Z1M | nutlin-3 | 95% | 1.1s |
+| ✅ BCL-XL BH3 groove | 1LXL | navitoclax | 91% | 2.9s |
+| ✅ BCL-2 BH3 groove | 1G5M | venetoclax | 73% | 1.2s |
+| ✅ c-ABL myristate pocket | 3CS9 | asciminib | 44% | 1.7s |
+| ✅ PTP1B allosteric helix site | 1A5Y | benzofuran inhibitors | 41% | 2.1s |
+| ✅ p38α DFG-out pocket | 1P38 | BIRB 796 | 38% | 3.0s |
+| ✅ HIV-1 RT NNRTI pocket | 1HMV | nevirapine | 38% | 8.3s |
+| ✅ HCV NS5B thumb-site I | 1NB4 | VXR class | 33% | 4.4s |
+| ✅ PKM2 allosteric activator | 1ZJH | TEPP-46 class | 33% | 4.3s |
+| ✅ PPARγ allosteric AF-2 site | 2PRG | metaglidasen | 35% | 1.7s |
+| ✅ Glucokinase allosteric site | 1V4S | B84 activator | 30% | 3.7s |
+| ✅ MMP-13 S1′ allosteric tunnel | 2OZR | non-zinc inhibitors | 50% | 1.1s |
+| ❌ IL-2 helix-α1 site *(Boltz-2 → 71% ✅)* | 1M47 | - | 21% | 0.7s |
+| ❌ Src myristate pocket | 2SRC | - | 28% | 4.1s |
+| ❌ SHP-2 allosteric tunnel | 2SHP | SHP099 class | 24% | 6.0s |
+| ❌ ERK2 allosteric site | 2ERK | - | 27% | 2.9s |
+| ❌ Caspase-1 dimer interface | 2HBQ | - | 8% | 1.6s |
+| ❌ IDH1 R132H dimer interface | 3MAP | ivosidenib | 0% | 3.7s |
+
+The six misses fall into two mechanistic classes: **near-misses** (IL-2, Src, SHP-2, ERK2 all 21–28%) where a physics-based backend closes the gap, and **dimer-interface pockets** (Caspase-1, IDH1 R132H) where the pocket forms between two chains.
+
+Dimer-interface pockets are now addressable with the `--homodimer` flag, which reads BIOMT symmetry records from the PDB and constructs the full biological assembly before analysis:
+
+```bash
+lacuna discover 2HBQ.pdb --homodimer --conformers 20   # Caspase-1 dimer interface
+lacuna discover 3MAP.pdb --homodimer --conformers 20   # IDH1 R132H dimer interface
+```
+
+IL-2 is confirmed at 71% rank 1 with Boltz-2 partial diffusion.
+
+### Boltz-2 re-evaluation of near-misses
+
+| Protein | RandomBackend | Boltz-2 | Notes |
+|---------|--------------|---------|-------|
+| IL-2 (1M47) | 21% - ❌ | **71% rank 1 - ✅** | Boltz samples the helix-α1 open state |
+| Src myristate (2SRC) | 28% - ❌ | 8% - ❌ | Requires SH2-kinase linker rearrangement; needs MD |
+
+### Conformational and orthosteric controls
+
+| Category | Result | Notable entries |
+|----------|--------|-----------------|
+| Conformational | 1 / 1 (100%) | Adenylate kinase open→closed (35%, rank 1) |
+| Orthosteric | 4 / 6 (67%) | HIV protease 100%, DHFR 100%, HIF-2α 100% |
+| Orthosteric miss | - | Trypsin (residue numbering offset); thrombin (940-residue complex) |
+
+**Overall across all 27 proteins: 19 / 27 (70%).**
+
+### Speed (NMABackend, no GPU)
+
+| Protein size | Time |
+|-------------|------|
+| ~130 residues (lysozyme) | 0.6s |
+| ~170 residues (MDM2) | 1.1s |
+| ~350 residues (K-Ras) | 0.9s |
+| ~530 residues (HIV-1 RT chain A) | 8.3s |
+
+### Head-to-head: Lacuna vs fpocket
+
+fpocket detects pockets on a single static structure. Lacuna generates a conformational ensemble - the critical difference for cryptic sites that are absent in the apo crystal.
+
+| Target | fpocket 4.2 | Lacuna (NMABackend) |
+|--------|------------|----------------------|
+| 1HEL hen lysozyme (orthosteric) | ✅ rank 1 | ✅ 100%, rank 2 |
+| 1L90 T4L L99A **(cryptic)** | ❌ not in top 5 | ✅ 100%, rank 1 |
+| 4OBE K-Ras switch-II **(cryptic)** | ❌ not in top 5 | ✅ 93%, rank 4 |
+| 1HPV HIV-1 protease (orthosteric) | ✅ rank 1 | ✅ 100%, rank 1 |
+| **Score** | **2 / 4** | **4 / 4** |
+
+T4L L99A and K-Ras switch-II are the canonical single-structure benchmark failures: the T4L cavity is physically absent in the apo crystal (<100 ų), and the K-Ras switch-II pocket only opens during nucleotide exchange.
+
+> **Reproduce:**
+> ```bash
+> python benchmarks/cryptic_benchmark.py          # full 27-protein run, NMA backend (~5 min)
+> python benchmarks/cryptic_benchmark.py --quick  # 10 conformers (~2 min)
+> python benchmarks/cryptic_benchmark.py --category cryptic   # cryptic only
+> python benchmarks/boltz_nearmiss.py             # Boltz-2 near-miss eval (GPU)
+> python benchmarks/compare_fpocket.py            # fpocket head-to-head
+> ```
+
+---
+
+## Example: K-Ras switch-II
+
+```bash
+# Download K-Ras apo (from RCSB)
+# Run with Boltz backend for highest-quality switch-II sampling
+lacuna discover 4OBE.pdb \
+    --backend boltz \
+    --conformers 30 \
+    --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`](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.W. (2026). *Lacuna: Cryptic Binding Pocket Discovery via Conformational Ensemble Analysis.* https://github.com/mooreneural/lacuna
+
+**BibTeX:**
+```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 = {0.1.0}
+}
+```
+
+**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
+
+**[MIT License](LICENSE)** — free for any use, including commercial.
+
+A [commercial license](LICENSE_COMMERCIAL) is available for organizations that
+require indemnification, warranty coverage, support SLAs, or custom development
+agreements. Contact claytonwaynemoore@gmail.com.

lacuna_pockets-0.1.0/PREDICTIONS.md

@@ -0,0 +1,144 @@
+# Lacuna Prospective Predictions
+
+Pockets predicted computationally from apo structures, with no prior knowledge
+of small-molecule binding at these sites. Each entry is a testable hypothesis:
+if the pocket is real, a fragment screen or thermal shift assay targeting the
+listed residues should show a hit rate above background.
+
+---
+
+## Prediction 1: Beta-catenin ARM7 cryptic sub-pocket
+
+**Date:** 2026-06-02  
+**Target:** Beta-catenin / CTNNB1 (UniProt P35222)  
+**Structure used:** PDB 2Z6H (apo armadillo repeat domain, residues 149–691)  
+**Backend:** Boltz-2 partial diffusion, 30 conformers  
+**Status:** Prospective — no small molecule known to bind this surface
+
+### Why this matters
+
+Beta-catenin is the central effector of the WNT signalling pathway and is
+constitutively active in ~70% of colorectal cancers, ~30% of hepatocellular
+carcinomas, and many other solid tumours. Despite 40 years of effort, no
+approved small molecule directly inhibits beta-catenin. The primary
+therapeutic interface — the TCF/LEF binding groove — spans ~3000 Ų and
+has been considered too large and flat for drug-like molecules.
+
+### What Lacuna found
+
+Running Boltz-2 on the apo structure surfaces a cryptic sub-pocket that is:
+
+| Property | Value |
+|----------|-------|
+| Rank | 12 of 109 clusters |
+| Druggability score | **0.855** (high; optimal pocket volume ~300 ų) |
+| Persistence | 16% (opens in ~5 of 31 conformers — genuinely cryptic) |
+| Volume | 241 ų (drug-like range: 200–500 ų) |
+| Centroid | (−3.2, −25.5, 10.7) Å |
+
+**Lining residues (ARM7–8 interface):**
+
+| Residue | ARM repeat | Chemical character |
+|---------|------------|--------------------|
+| ARG515 | ARM7 | Positive (also contacts TCF/LEF at surface) |
+| ARG542 | ARM7–8 loop | Positive |
+| ALA543 | ARM8 | Hydrophobic |
+| ARG549 | ARM8 | Positive |
+| ARG550 | ARM8 | Positive |
+| ARG565 | ARM8 | Positive |
+| CYS573 | ARM8 | Nucleophilic — covalent warhead opportunity |
+| ALA576 | ARM8 | Hydrophobic |
+
+**Key observation:** CYS573 is partially buried in the cryptic state and
+becomes solvent-exposed in ~16% of Boltz-2 conformers. This makes it a
+candidate for covalent fragment screening (electrophilic warheads:
+acrylamide, chloroacetamide).
+
+### Structural context
+
+The pocket sits on the concave face of ARM repeats 7–8, adjacent to but
+distinct from the canonical TCF/LEF groove (which centres on ARM1–4 around
+K312, R469, K508). The nearest pharmacological interface is the APC/axin
+binding region (ARM6, ~residues 382–395); the predicted pocket shares no
+lining residues with it.
+
+This region corresponds to the "third face" of the ARM domain described in
+structural biology literature as a potential protein–protein interaction
+surface. No crystal structure shows a small molecule bound here.
+
+### Validation strategy
+
+**Tier 1 — biophysical (2–4 weeks, ~$10K):**
+1. Express and purify CTNNB1 ARM domain (residues 149–691, well-established protocol)
+2. Run a commercial DNA-encoded chemical library (DEL) screen against the ARM domain
+3. Thermal shift assay (TSA/DSF) with 500–1000 fragment library, look for
+   ΔTm ≥ 1°C hits that do NOT compete with a TCF peptide (ruling out
+   canonical groove binders)
+4. SPR or ITC to confirm direct binding of confirmed hits
+
+**Tier 2 — structural (4–8 weeks if hits found):**
+- Co-crystallise or cryo-EM of hit compounds with CTNNB1 ARM domain
+- Confirm binding to ARM7–8 residues predicted by Lacuna
+
+**Tier 3 — cellular (parallel with Tier 2):**
+- STF luciferase reporter assay (WNT pathway readout) in HCT116 or SW480
+  cells (both have CTNNB1-activating mutations)
+- Look for inhibition of TOPFlash signal at non-cytotoxic concentrations
+
+### Why CYS573 is actionable
+
+Covalent fragment libraries targeting cysteines are now commercially available
+(Enamine, Chembridge) and have yielded clinical candidates in KRAS (G12C),
+BTK, and other targets. CYS573 is a non-catalytic cysteine with no known
+function — an ideal covalent handle if the pocket is accessible.
+
+### Docking files
+
+Lacuna output files for this prediction:
+
+```
+ctnnb1_boltz_lacuna/
+  pocket_report.json          — full ranked pocket list
+  pocket_11_site.pdb          — pseudoatom PDB for PyMOL/ChimeraX (rank 12 = index 11)
+  pocket_11_constraint.yaml   — Boltz YAML: add a SMILES and run boltz predict
+  pocket_11_vina.conf         — AutoDock Vina / Gnina box centred on prediction
+```
+
+To dock a fragment into this pocket:
+```yaml
+# pocket_11_constraint.yaml — add your fragment SMILES:
+version: 1
+sequences:
+  - protein:
+      id: A
+      sequence: "..."   # from output file
+  - ligand:
+      id: L
+      smiles: "YOUR_FRAGMENT_SMILES"
+constraints:
+  pocket:
+    - [A, 515]
+    - [A, 542]
+    - [A, 543]
+    - [A, 549]
+    - [A, 573]
+```
+
+---
+
+## Adding predictions
+
+To add a new prospective prediction, run:
+
+```bash
+lacuna discover YOUR_TARGET.pdb \
+    --backend boltz \
+    --conformers 30 \
+    --emit-boltz-constraints \
+    --emit-vina-boxes \
+    --output predictions/YOUR_TARGET/
+```
+
+Then document the novel pockets (not overlapping known interfaces) in this
+file with the residue list, druggability score, persistence, and proposed
+validation assay.