Cofolding-Approach to Fragment Exploration of Allosteric and Cryptic Binding Sites
CAFE is a workflow for probing orthosteric and allosteric pockets with fragment libraries via structure prediction (Boltz) and geometric localization. This repository packages the fragment catalogs, pocket reference structures, and pipeline entry points used in the study.
Private preview release (THGLab).
| Path | Contents |
|---|---|
data/fragments/ |
Paper-style fragment catalogs (kinase, nonkinase, Enamine) |
data/references/sites.csv |
Orthosteric / allosteric site definitions (PDB, chain, ligand, resseq) |
data/references/pdb/ |
Reference structures used for alignment and localization |
scripts/fragmentation/ |
Build / regenerate fragment libraries (BRICS) |
scripts/boltz/ |
Cofolding: prepare inputs → predict → align |
scripts/localization/ |
Score poses vs orthosteric / allosteric references |
A three-arm worked example (no ADP / ADP / Type‑1) is in examples/mapk14_map_a005/.
From the repo root (needs numpy; RDKit only for the optional BRICS demo):
# Score pre-aligned Boltz poses for each ATP-site condition
for arm in no_adp with_adp type1; do
PYTHONPATH=scripts python scripts/localization/score_poses.py \
--target MAPK14 \
--poses examples/mapk14_map_a005/aligned_poses/$arm \
--out examples/mapk14_map_a005/localization_${arm}.csv
doneExpected at 5 Å: no ADP → 10/10 ortho; ADP and Type‑1 → 10/10 allo.
Boltz input YAMLs for the same fragment are under examples/mapk14_map_a005/boltz_yamls/. See that folder’s README for rebuild / predict / align steps.
Human-readable IDs for the libraries used in the paper:
| File | Count | ID pattern | Notes |
|---|---|---|---|
kinase_fragment_catalog.csv |
232 | {KIN}-A### / {KIN}-O### |
116 allosteric + 116 orthosteric BRICS fragments across AKT2, CDK2, CHEK1, CSNK2A1, MAPK14 |
nonkinase_fragment_catalog.csv |
37 | PTP-A### / KRA-A### |
Allosteric-only BRICS fragments for PTP1B (23) and KRAS (14) |
enamine_fragment_catalog.csv |
300 | ENA-### |
Enamine diversity set ordered by selection_rank |
Kinase abbreviations: AKT, CDK, CHK, CSK, MAP.
Example rows:
CHK-A000 allosteric CHEK1 3F9N/38M …
MAP-O012 orthosteric MAPK14 …
PTP-A000 allosteric PTP1B 1T49/892 …
ENA-000 enamine Z1473029920 …
Pocket geometry is defined in data/references/sites.csv:
target,role,pdb_id,chain,ligand,resseq
MAPK14,ortho,3S3I,A,CQ0,1
MAPK14,allo,5N63,A,8OW,401
…
PDB files live under:
data/references/pdb/
kinases/<TARGET>/<PDB>.pdb
nonkinase/<TARGET>/<PDB>.pdb
Kinases — one orthosteric alignment frame per target, plus allosteric localization references:
| Target | Ortho frame | Allo refs |
|---|---|---|
| AKT2 | 2JDO | 8Q61, 9C1W |
| CDK2 | 2UUE | 6Q3F, 6Q49, 6Q4K, 8VQ3, 8VQ4 |
| CHEK1 | 2YEX | 3F9N, 3JVR, 3JVS |
| CSNK2A1 | 3WAR | 5MMF, 5MOD, 5OSU, 6GIH |
| MAPK14 | 3S3I | 3NEW, 5N63, 5N64, 5N67, 5N68, 8X3M, 8YD9 |
Nonkinases
| Target | Ortho / blocker | Allo refs |
|---|---|---|
| PTP1B | 5K9W (OTA) | 1T49, 1T4J, 7GSA, 7GTQ, 8G65, 8G68, 8G69 |
| KRAS | 4OBE (GDP) | 7RPZ, 5V71 |
Structures are RCSB-derived, cleaned to the chains used in the study. Cite the PDB IDs in any downstream work.
fragments ──► Boltz cofolding ──► align to ortho frame ──► localize vs sites.csv
│ │ │
catalogs predicted poses % in pocket (e.g. ≤5 Å)
Typical arms for kinases: empty ATP site (no_adp), ADP-occupied, and Type‑1 inhibitor–occupied (verified CCD ligand). Localization uses heavy-atom minimum distance to reference ligands listed in sites.csv.
Script folders are staged for a public-facing API; wire them to your Boltz install and compute environment as needed.
conda env create -f environment.yml
conda activate cafeRequires a working Boltz installation for cofolding (GPU recommended). RDKit is used for fragmentation; MDAnalysis or PyMOL can be used for alignment helpers.
CAFE/
├── README.md
├── LICENSE
├── environment.yml
├── data/
│ ├── fragments/
│ └── references/
│ ├── sites.csv
│ └── pdb/
├── examples/
│ └── mapk14_map_a005/ # no ADP / ADP / Type-1 walkthrough
└── scripts/
├── lib/ # shared PDB / sites / Kabsch helpers
├── fragmentation/brics_cut.py
├── boltz/prepare_yaml.py
├── boltz/align_to_ortho.py
└── localization/score_poses.py
Manuscript in preparation. Please cite this repository and the RCSB PDB entries for any reference structures you reuse.
Enamine catalog identifiers (Z…) remain the property of Enamine; SMILES are provided for reproducibility of the computational screen.
Code is released under the MIT License (see LICENSE). Reference PDB coordinates remain subject to the RCSB PDB terms; fragment catalog SMILES are provided for research use.