mdciao: Accessible Analysis and Visualization of Molecular Dynamics Simulation Data

mdciao is a Python module that provides quick, "one-shot" command-line tools to analyze molecular simulation data using residue-residue distances. mdciao tries to automate as much as possible for non-experienced users while remaining highly customizable for advanced users, by exposing an API to construct your own analysis workflow.

Under the hood, the module mdtraj is doing most of the computation and handling of molecular information, using BioPython for sequence alignment, pandas for many table and IO related operations, and matplotlib for visualizaton. It tries to automatically use the consensus nomenclature for

• GPCRs, e.g. Ballesteros-Weinstein-Numbering or structure-based schemes by Gloriam et al,
• G-proteins, via Common G-alpha Numbering (CGN), and
• Kinases, via their 85 pocket-residue numbering scheme.

by either using local files or on-the-fly lookups of the GPCRdb and/or KLIFS.

Licenses

• mdciao is licensed under the GNU Lesser General Public License v3.0 or later (LGPL-3.0-or-later, see the LICENSE.txt).
• mdciao uses a modified version of the method mdtraj.compute_contacts of mdtraj. This modified version is published along with mdciao and can be found in contacts/_md_compute_contacts.py. Please see that file for details on the modifications.

• Modules used by mdciao have different licenses. You can check any module's license in your Python environment using pip-licenses:

  >>> pip-licenses | grep module_name

Documentation

Currently, docs are hosted at http://proteinformatics.org/mdciao/, but this can change in the future.

System Requirements

mdciao is developed in GNU/Linux, and CI-tested via github actions for GNU/Linux and MacOs. Tested python versions are:

• GNU/Linux: 3.7, 3.8, 3.9, 3.10, 3.11

* MacOs: 3.7, 3.8, 3.9, 3.10, 3.11. For Python 3.7, four CI-tests involving mdtraj.compute_dssp , are skipped because of a hard to repdroduce, random segmentation fault, which apparently wont fix, see here mdtraj/mdtraj#1574 and here.

So everything should work out of the box in these conditions.

Authors

mdciao is written and maintained by Guillermo Pérez-Hernández (ORCID) currently at the Institute of Medical Physics and Biophysics in the Charité Universitäsmedizin Berlin.

Please cite:

• mdciao: Accessible Analysis and Visualization of Molecular Dynamics Simulation Data

  Guillermo Pérez-Hernández, Peter-Werner Hildebrand
  bioRxiv 2022.07.15.500163
  https://doi.org/10.1101/2022.07.15.500163

Status

mdciao is approaching its first major release, so less changes in the API and CLI calls are expected. For more info on semantic versioning please check the semver page.

TODOs

This is an informal list of known issues and TODOs:

• Adopt this project structure https://github.com/MolSSI/cookiecutter-cms
• keeping vs reporting contacts: a design choice has to be made wrt to the effect of ctc_cutoff_Ang on a ContactGroup: If a given cutoff makes a ContactPair have freq=0, should the CP be kept in the CG, simply not reported? There's now a branch for that: https://github.com/gph82/mdciao/tree/buffer_neighborhood
• overhaul the "printing" system with proper logging and warnings (perhaps use loguru)
• progressbar not very informative for one chunked trajectory or parallel runs
• the affiliation of a residue to a fragment is done as "res@frag" on the string output and res^frag in figures, this implementation is simply using replace("@","^"), could be better
• parallel execution with memory mdtraj.Trajectory objects should be better
• harmonize documentation API cli methods (mdciao.cli) and the CLI scripts (mdc*)
• The interface between API methods and cli scripts could be better, using sth like click
• The API-cli methods (interface, neighborhoods, sites, etc) have very similar flows but a lot of code repetition, I am sure some patterns/boilerplate could be outsourced/refactored even more.
• Most of the tests were written against a very rigid API that mimicked the CLI closely. Now the API is more flexible and many tests could be re-written or deleted , like those needing mock-input or writing to tempdirs because writing figures or files could not be avoided.
• There's some inconsistencies in private vs public attributes of classes. An attribute might've "started" as private and is exceptionally used somewhere else until the number of exceptions is enough for it to make sense to be public, documented and well tested. I'm working on it.
• neighborlists could be computed much more efficiently
• The labelling names should be harmonized (ctc_label, anchor_res...) and the logic of how/where it get's constructed (short_AA vs AA_format) is not obvious sometimes
• There's many other TODOs spread throughout the code
• The way uniprot or PDB codes are transformed to relative and/or absolute filenames to check if they exist locally should be unified across all lookup functions, like GPCR_finder, PDB_finder and/or the different LabelerConsensus objects, possibly by dropping optargs like 'local_path' or 'format'.
• Some closely related methods could/should be integrated into each other by generalising a bit, but sometimes the generalisation is unnecessarily complicated to code (or I simply forget that the closely related method already exists) and re-code (and test!) for a slightly different scenario (though I try to hard to avoid it). E.g. there's several methods for computing, reporting, and saving contact frequencies and contact-matrices, or different methods to assign residue idxs to fragments, depending on particual the goal of the assignment, like find_parent_list, in_what(N)_fragments, or assign_fragments. Still, I opted for more smaller methods, which are individually easier to maintain, but that could simply be a questionable choice.
• The 'dictionary unifying' methods could be replaced with pandas.DataFrame.merge/join
• Writing to files, file manipulation should be done with pathlib