MDSynthesis: a persistence engine for molecular dynamics data¶
Although the raw data for any study involving molecular dynamics simulations are the full trajectories themselves, often we are most interested in lower-dimensional measures of what is happening. These measures may be as simple as the distance between two specific atoms, or as complex as the percentage of contacts relative to some native structure. Some measures may even be comparisons of two or more trajectories against each other. In any case, it may be time-consuming to obtain these lower-dimensional intermediate data, and so it is useful to store them.
MDSynthesis is designed to perform the logistics of medium-to-large-scale analysis of many trajectories, individually or as entire groups. It is intended to allow the scientist to operate at a high level when working with the data, while letting MDSynthesis handle the details of storing and recalling this data.
In other words, MDSynthesis lets the computer do the boring work of keeping track of where things are and how they are stored.
Efficiently store intermediate data from individual simulations for easy recall¶
For a given simulation trajectory, MDSynthesis gives an interface (the Sim object) to the simulation data itself through MDAnalysis. Data structures generated from raw trajectories (pandas objects, numpy arrays, or any pure python structure) can then be stored and easily recalled later. Under the hood, datasets are stored in the efficient HDF5 format when possible.
Collect aggregated data and keep track of it, too¶
Query for simulation results instead of manually hunting for them¶
This feature is planned, but not yet present in the codebase.
Sim and Group objects persistently store their data to disk automatically, but it can be tedious to navigate around the filesystem to recall them later. The Coordinator object gives a single interface for querying all Sim and Group objects it is made aware of, allowing retrieval of specific datasets with a single line of code.
We have yet to make an official release, but you can get the current state of the codebase from the master branch on GitHub.
See the installation instructions to set it up.
- MDAnalysis: 0.9.1 or higher
- pandas: 0.16.1 or higher
- PyTables: 3.2.0 or higher
- h5py: 2.5.0 or higher
- scandir: 1.0 or higher
This project is still under heavy development, and there are certainly rough edges and bugs. Issues and pull requests welcome!