CREVICE

Cavity and Residue Environment Visualisation, Interaction and Connectivity Evaluation

CREVICE is a Python package and command-line tool for measuring pores, channels, tunnels and cavities in protein structures and molecular dynamics (MD) trajectories, and for describing the residues and water that surround them. It combines ideas from HOLE, HOLLOW, CAVER and 3V in one workflow and writes numerical tables, figures and matching PyMOL, VMD and ChimeraX scenes.

With CREVICE you can:

• measure through-channel radius profiles with continuous atom-clearance checks and an automatically proposed or user-supplied axis;
• fill pockets, cavities and channels with atom-clear probe spheres and export the measured binary map and a smoothed display surface;
• search for enclosed cavities and widest-path tunnels on clearance grids;
• identify the residues that form a measured boundary, their nonlocal contact partners and residue interaction networks;
• follow a cavity through an MD trajectory: aligned per-frame volumes, sectional widths, changing boundary residues and contacts, with approximate uncertainty on means;
• measure hydration: explicit water contacts, solvent-accessible area, water density, typed interactions and water inside the instantaneous cavity;
• define, version and compare named regions over identical frames.

Every analysis runs from the crevice command or from Python.

Installation

CREVICE needs Python 3.10 or newer.

python -m pip install crevice

NumPy, SciPy, MDAnalysis, Matplotlib and scikit-image are installed automatically. Optional extras add libraries used by specific features:

For example: python -m pip install "crevice[science]".

PyMOL, VMD and ChimeraX are separate programs and are not installed by pip. CREVICE writes scene scripts for them.

Quick start

Structures can be given as a local file (PDB, mmCIF or GRO), an RCSB accession or an AlphaFold DB identifier (for example AF-P02920-F1); remote entries are downloaded once and cached with a provenance record.

# Radius profile of the gramicidin A channel (table also written as 1GRM_profile.csv)
crevice profile 1GRM --axis auto -o 1GRM_profile.json --png 1GRM_profile_radius.png

# 3D cast of the dominant interior cavity of GLUT1, with viewer scenes
crevice cast 4PYP --out-dir 4PYP_cast

# Everything for one structure: profile, cast, residues, cavities, tunnels,
# network, hydration, figures and PyMOL/VMD/ChimeraX scenes
crevice publish 1GRM --out-dir results/1GRM --prefix 1GRM

The same profile from Python:

from crevice import annotate_residues, load_structure, pore_profile

frame = load_structure(".crevice/pdb/1GRM.cif")  # cached by the command above, or any local file
profile = pore_profile(frame, axis="auto", samples=81)
contacts = annotate_residues(frame, profile.points)

print(profile.min_radius, profile.bottleneck.nearest_residue)
for contact in contacts[:5]:
    print(contact.residue, contact.role, round(contact.min_distance, 2))

Python functions return result objects and do not write files; writers such as crevice.write_profile_json and plots such as crevice.plot_profile_radius are called separately.

Tools

Run crevice <command> --help for every option.

A trajectory example (GRO/XTC or any format MDAnalysis reads):

crevice cavity-trajectory system.gro run.xtc \
  --reference-volume-dx pocket_volume.dx --geometry-mode reference-region \
  --probe-radius 1.4 --region-margin 2 --spacing 0.2 --axis 0,0,1 \
  --max-frames 1100 --workers 4 --out-dir trajectory-results --prefix system

The reference map must be a measured binary map (*_volume.dx from cast or publish) in the first frame's coordinates. --max-frames is a guard against loading too many frames by accident, not a stride.

Outputs

Workflow commands write a directory with a *_manifest.json that lists every file, the input provenance and the parser that ran. Typical outputs are:

• CSV tables for every dataset you would analyse (profiles, cast regions, residues, cavities, tunnels, networks, per-frame trajectory and hydration series, statistics), with units in the column names (radius_A, volume_A3, time_ps); JSON only for manifests, provenance, full structured records read by other commands and viewer scene definitions;
• a measured binary OpenDX map (*_volume.dx), from which every volume, radius and residue attribution is computed, and separate smoothed display maps and meshes (--smooth X, default 0.4 Å, changes the picture only);
• figures (radius profiles, residue contacts with a contact-role legend, network summaries and chords, trajectory distributions; 200 dpi by default, --dpi 400 for print), clean by default (--annotate adds titles and labels);
• PyMOL (.pml), VMD (.vmd/.tcl) and ChimeraX (.cxc) scenes that share one viewer PDB, with crevice_render / *_render.cxc for high-resolution images. Keep the whole output directory together. No HTML is written.

Unresolved, unavailable and zero are kept distinct: a method that cannot give a defensible answer records its status and reason instead of a number.

Scope and interpretation

CREVICE measures geometry. Please read its results with these limits in mind:

• Geometric output is not biological validation. A cavity, path, volume, radius or residue ranking is a measurement under stated settings (grid spacing, probe radii, atom selection, alignment). None of the bundled benchmark entries is curated, and CREVICE does not decide which space is biologically relevant.
• Results depend on parameters. Check sensitivity to grid spacing, probe radius and region definition before interpreting a number, and report the settings with it. Requested grids are never coarsened silently.
• Unresolved is not closed. If no unique two-ended channel is found, the profile is reported as unresolved; that does not show the protein is closed. Missing water in a static structure is not evidence of a dry pocket.
• Contacts and correlations are associations. Residue networks, motion correlations and hydration statistics identify candidates for follow-up, not causal pathways or residues that control transport.
• Trajectory uncertainty is approximate. Mean intervals assume stationary, representative sampling and exclude force-field, grid and probe errors. Frame-range bands describe fluctuations, not confidence in the mean.
• Viewer scenes should be inspected. A generated scene is not checked just because CREVICE wrote it; open it in the viewer before relying on it.

Each command's documentation page states its method, assumptions, parameters, outputs, testing status and known limitations.

Documentation

The full documentation (installation, quick start, one page per tool, method notes, CLI and Python API reference) is at https://crevice.readthedocs.io. Its sources are in docs/; to build them locally:

python -m pip install . -r docs/requirements.txt
python -m sphinx -b html docs docs/_build/html

Contributing

Contributions, bug reports and careful reports of results that look wrong are welcome. See CONTRIBUTING.md and the code of conduct. To run the tests:

python -m pip install -e ".[test]"
python -m pytest -q -rs tests

Citing CREVICE

A paper is in preparation:

Lambden, E., Findlay, H. and Booth, P. J. CREVICE: Analysis and Visualisation of Protein Cavities, Channels, and Residue Interaction Networks. In preparation.

Until it is published, please cite the software using CITATION.cff and give the version you used. Please also cite the methods and tools your analysis relies on (for example HOLE, HOLLOW, CAVER, 3V and MDAnalysis).

Licence

CREVICE is released under the MIT licence.