Lacuna

Lacuna finds cryptic binding pockets: sites that are closed or absent in a single static structure and only open when the protein moves. Standard cavity detectors analyse the coordinates you give them, so a pocket that is shut in that snapshot is invisible to them.

Lacuna generates a conformational ensemble from your input, runs pocket detection on every conformer, then clusters the detections so one site can be tracked across the ensemble. Sites are then scored by how far they open relative to the input structure and how many conformers they appear in, not just by their size in one frame.

Install

pip install lacuna-pockets

Base install is numpy, scipy, biopython, click and rich. No compiled toolkits, no GPU, Python 3.10+.

Usage

lacuna discover protein.pdb --conformers 20 --output results/

Accepts PDB and mmCIF. A 456-residue dimer takes about 3 seconds on a laptop CPU.

Output

A ranked table and a pocket_report.json reporting, per site: volume dynamics across the ensemble, druggability, persistence, crypticity, and lining residues. It will also emit AutoDock Vina search boxes, Boltz pocket constraints, and pocket pseudoatom PDBs for downstream docking and visualisation:

lacuna discover protein.pdb --emit-vina-boxes --emit-boltz-constraints

Ensemble backends

Normal mode analysis is the default and needs nothing beyond the base install. OpenMM implicit-solvent MD and Boltz-2 diffusion sampling are optional extras:

pip install "lacuna-pockets[openmm]"
pip install "lacuna-pockets[boltz]"

Known limitation

I benchmarked six detector configurations on CryptoBench and the bottleneck is ranking, not detection. Pooling every method, a qualifying candidate exists for 92.2% of targets, but conversion into a top-five hit varies by 36.7 points across methods. Lacuna is under the same ceiling. If a site matters to you, look further down the ranked list than five.

Links

Code: https://github.com/mooreneural/lacuna Docs and preprints: see the README License: MIT

Also runnable in a browser with no install, on Tamarind Bio and Neurosnap, both linked from the README.

Feedback and bug reports welcome, especially cases where it ranks a known site poorly.