prothon compare --ensembles wild_type.dcd mutant.dcd --topology topology.pdb

CBCN (reference: ensemble 0)
  ensemble 1: d = 0.2841 (floor 0.0472) — 34/76 residues differ

Two numbers, and the second is the one that decides whether the first means anything. Split one ensemble in half at random and compare the halves: the answer is not zero, because a finite sample never reproduces a continuous distribution exactly. That self-distance is the noise floor — the smallest difference this much sampling can resolve.

Here 0.2841 clears a floor of 0.0472 by six times. At 0.05 it would not have, and Prothon would say so rather than report a small difference.

Install

conda create -n prothon -c conda-forge prothon
conda activate prothon

prothon info lists what is installed and what version:

prothon info

For source work, clone the repository. GitHub's Download ZIP is an unversioned audit snapshot; installable, versioned artifacts are the PyPI wheel/source distribution and the conda-forge package.

What you can compare

It withholds rather than overstates. Trajectory frames are not independent draws, and a test that assumes they are calls almost every residue different when nothing differs. Prothon estimates the correlation time from your data, keeps the sample in trajectory order, and permutes contiguous blocks of it. Where a trajectory holds too few independent blocks to build a null from, it reports the floor and prints no p-value at all.

The false-positive rate is measured rather than asserted. A test that assumes independent frames calls 99% of residues different when nothing differs; this one calls between 8% and 16% at a nominal 5%, across sample sizes from 500 to 4000 and correlation times from 1 to 50, at 2000 null studies per configuration. That is the honest number and it is on the calibration page with the grid that produced it.

One import, and everything is reachable from it

from prothon import Prothon

prothon = Prothon(["wild_type.dcd", "mutant.dcd"], "topology.pdb", "cbcn",
                  random_state=0)

prothon.compare()                       # where do they differ
prothon.distinguishability()            # differences *between* residues
prothon.coverage_and_fidelity()         # missed states, or invented ones
prothon.rank()                          # several against a reference
prothon.validate("rg", [2.71], [0.08])  # against experiment
prothon.save_config()                   # writes prothon.yml

The order parameter is a property of the study, so it is named once. Any method takes one to override it for a single call.

Three ways to run the same study, and none is a subset of another — flags, file and API are generated from one schema:

Any command line can be written out as a file with --save-config, and any file can be run from the command line.

Documentation

Start here — Install · Your first comparison · Order parameters · Worked examples

Going further — The statistics · Different molecules · Distances · Scoring against experiment · Ranking several ensembles

Measurements — Calibration · Circular parameters · Convergence · Ubiquitin · Performance

Reference — CLI · Configuration · Python API

Citation

Aina, A.; Hsueh, S. C. C.; Plotkin, S. S. PROTHON: A Local Order Parameter-Based Method for Efficient Comparison of Protein Ensembles. J. Chem. Inf. Model. 2023, 63, 3453–3461. DOI: 10.1021/acs.jcim.3c00145

@article{aina2023prothon,
  author  = {Aina, Adekunle and Hsueh, Shawn C. C. and Plotkin, Steven S.},
  title   = {PROTHON: A Local Order Parameter-Based Method for Efficient
             Comparison of Protein Ensembles},
  journal = {Journal of Chemical Information and Modeling},
  volume  = {63},
  number  = {11},
  pages   = {3453--3461},
  year    = {2023},
  doi     = {10.1021/acs.jcim.3c00145},
}

Contributing

Contributions welcome — see CONTRIBUTING.md.

License

MIT. See LICENSE.