The coordinates submodule contains code to read coordinates, either single frames (e.g. the PDB module) or trajectories (such as the DCD reader). All readers are supposed to expose a Reader object that presents a common Trajectory API to other code.
The Universe contains the API entry point attribute
In order to write coordinates, a factory function is provided (MDAnalysis.coordinates.core.writer()) which is made available as MDAnalysis.Writer()) that returns a Writer appropriate for the desired file format (as indicated by the filename suffix). Furthermore, a trajectory Reader can also have a method Writer() that returns an appropriate Writer for the file format of the trajectory.
In analogy to MDAnalysis.coordinates.core.writer(), there is also a MDAnalysis.coordinates.core.reader() function available to return a trajectory Reader instance although this is often not needed because the Universe class can choose an appropriate reader automatically.
The table below lists the coordinate file formats understood by MDAnalysis. The emphasis is on formats that are used in popular molecular dynamics codes. MDAnalysis figures out formats by looking at the extension. Thus, a DCD file always has to end with ”.dcd” to be recognized as such. (A number of files are also recognized when they are compressed with gzip or bzip2 such as ”.xyz.bz2”.)
|CHARMM, NAMD, LAMMPS||dcd||r/w||standard CHARMM binary trajectory; endianness is autodetected. Fixed atoms may not be handled correctly (requires testing). Module MDAnalysis.coordinates.DCD|
|Gromacs||xtc||r/w||Compressed (lossy) xtc trajectory format. Module MDAnalysis.coordinates.XTC|
|Gromacs||trr||r/w||Full precision trr trajectory. Only coordinates are processed at the moment. Module MDAnalysis.coordinates.TRR|
|XYZ||xyz||r||Generic white-space separate XYZ format; can be compressed. Module MDAnalysis.coordinates.XYZ|
|Amber||trj, mdcrd||r||formatted (ASCII) trajectories; the presence of a periodic box is autodetected (experimental). Module MDAnalysis.coordinates.TRJ|
|Brookhaven ||pdb||r/w||a simplified PDB format (as used in MD simulations) is read by default; the full format can be read by supplying the permissive=False flag to MDAnalysis.Universe. Module MDAnalysis.coordinates.PDB|
|PDBQT ||pdbqt||r/w||file format used by AutoDock with atom types t and partial charges q. Module: MDAnalysis.coordinates.PDBQT|
|PQR ||pqr||r||PDB-like but whitespace-separated files with charge and radius information. Module MDAnalysis.coordinates.PQR|
|GROMOS96 ||gro||r/w||basic GROMOS96 format (without velocities). Module MDAnalysis.coordinates.GRO|
|CHARMM ||crd||r/w||“CARD” coordinate output from CHARMM; deals with either standard or EXTended format. Module MDAnalysis.coordinates.CRD|
|||(1, 2, 3, 4, 5) This format can also be used to provide basic topology information (i.e. the list of atoms); it is possible to create a full Universe by simply providing a file of this format: u = Universe(filename)|
The Trajectory API defines how classes have to be structured that allow reading and writing of coordinate files. By following the API it is possible to seamlessly enhance the I/O capabilities of MDAnalysis. The actual underlying I/O code can be written in C or python or a mixture thereof.
Typically, each format resides in its own module, named by the format specifier (and using upper case by convention).
Reader and Writer classes are derived from base classes in MDAnalysis.coordinates.base.
In various places, MDAnalysis tries to automatically select appropriate formats (e.g. by looking at file extensions). In order to allow it to choose the correct format, all I/O classes must be registered in one of three dictionaries with the format (typically the file extension in upper case):
A Timestep instance holds data for the current frame. It is updated whenever a new frame of the trajectory is read.
Timestep classes are derived from MDAnalysis.coordinates.base.Timestep, which is the primary implementation example (and used directly for the DCDReader).
depending on the type of arg, Timestep instances are created in different ways:
- empty Timestep for arg atoms (allocate arrays etc)
- makes a deep copy of the arg
- update the Timesteps positions array with the contents of arg
Anything else raises an exception; in particular it is not possible to create a “empty” Timestep instance.
- position(s) of atoms; can be a slice or numpy array and then returns coordinate array
- number of coordinates (atoms) in the frame
- iterator over all coordinates
- deep copy of the instance
- number of atoms in the frame
- current frame number
- system box dimensions (x, y, z, alpha, beta, gamma) (typically implemented as a property because it needs to translate whatever is in the underlying Timestep._unitcell attribute)
- system box volume (derived as the determinant of the box vectors of dimensions)
These attributes are set directly by the underlying trajectory readers. Normally the user should not have to directly access those (although in some cases it is convenient to directly use Timestep._pos).
- raw coordinates, a numpy.float32 array; X = _pos[:,0], Y = _pos[:,1], Z = _pos[:,2]
- unit cell dimensions and angles; the format depends on the underlying trajectory format. A user should use Timestep.dimensions to access the data in a standard format.
Trajectory readers are derived from MDAnalysis.coordinates.base.Reader. Typically, many methods and attributes are overriden.
The MDAnalysis.coordinates.DCD.DCDReader class is the primary implementation example.
The following methods must be implemented in a Reader class.
- __init__(filename, **kwargs)
open filename; other kwargs are processed as needed and the Reader is free to ignore them. Typically, MDAnalysis supplies as much information as possible to the Reader in kwargs; at the moment the following data are supplied in keywords when a trajectory is loaded from within MDAnalysis.Universe:
- numatoms: the number of atoms (known from the topology)
allow iteration from beginning to end:for ts in trajectory: print ts.frame
- close the file and cease I/O
- ensure that the trajectory is closed
- advance to next time step or raise IOError when moving past the last frame
- reposition to first frame
Not all trajectory formats support the following methods, either because the data are not available or the methods have not been implemented. Code should deal with missing methods gracefully.
- number of frames in trajectory
advance to time step arg = frame and return Timestep; or if arg is a slice, then return an iterator over that part of the trajectory.
The first functionality allows one to randomly access frames in the trajectory:universe.trajectory
would load frame 314 into the current Timestep.
Using slices allows iteration over parts of a trajectoryfor ts in universe.trajectory[1000:2000]: process_frame(ts) # do something
or skipping framesfor ts in universe.trajectory[1000::100]: process_frame(ts) # do something
The last example starts reading the trajectory at frame 1000 and reads every 100th frame until the end.
The performance of the __getitem__() method depends on the underlying trajectory reader and if it can implement random access to frames. In many cases this is not easily (or reliably) implementable and thus one is restricted to sequential iteration.
- Writer(filename, **kwargs)
returns a Writer which is set up with the same parameters as the trajectory that is being read (e.g. time step, length etc), which facilitates copying and simple on-the-fly manipulation.
If no Writer is defined then a NotImplementedError is raised.
The kwargs can be used to customize the Writer as they are typically passed through to the init method of the Writer, with sensible defaults filled in; the actual keyword arguments depend on the Writer.
- timeseries(atomGroup, [start[,stop[,skip[,format]]]])
- returns a subset of coordinate data
- populate a MDAnalysis.core.Timeseries.TimeseriesCollection object with observable timeseries computed from the trajectory
- filename of the trajectory
- number of atoms (coordinate sets) in a frame (constant)
- total number of frames (if known) – None if not known
- bool, saying if there are fixed atoms (e.g. dcds)
- step size for iterating through the trajectory 
- number of integrator steps between frames + 1 (i.e. the stride at which the MD simulation was sampled)
- integrator time step (in native units); hence the “length” of a trajctory frame is skip_timestep*delta time units
- contains box information for periodic boundary conditions
- the Timestep object; typically customized for each trajectory format and derived from base.Timestep.
- dictionary with keys time and length and the appropriate unit (e.g. ‘AKMA’ and ‘Angstrom’ for Charmm dcds, ‘ps’ and ‘nm’ for Gromacs trajectories, None and ‘Angstrom’ for PDB).
- string that identifies the file format, e.g. “DCD”, “PDB”, “CRD”, “XTC”, “TRR”; this is typically the file extension in upper case.
- time between frames in ps; a managed attribute (read only) that computes on the fly skip_timestep * delta and converts to the MDAnalysis base unit for time (pico seconds by default)
- total length of the trajectory = numframes * dt
- time of the current time step, in MDAnalysis time units (ps)
- frame number of the current time step
- string that identifies the compression (e.g. “gz” or “bz2”) or None.
Trajectory readers are derived from MDAnalysis.coordinates.base.Writer. They are use to write multiple frames to a trajectory file. Every time the write() method is called, another frame is appended to the trajectory.
Typically, many methods and attributes are overriden.
W = TrajectoryWriter(filename,numatoms,**kwargs) W.write_next_timestep(Timestep)
W.write(AtomGroup) # write a selection W.write(Universe) # write a whole universe W.write(Timestep) # same as write_next_timestep()
- opens filename and writes header if required by format
- write Timestep data in obj
- write data in timestep to trajectory file
- take the dimensions from the timestep and convert to the native unitcell representation of the format
- close file and finish I/O
- ensures that close() is called
- name of the trajectory file
- start, stop, step
- first and last frame and step
- dictionary with keys time and length and the appropriate unit (e.g. ‘AKMA’ and ‘Angstrom’ for CHARMM dcds, ‘ps’ and ‘nm’ for Gromacs trajectories, None and ‘Angstrom’ for PDB)
- string that identifies the file format, e.g. “DCD”, “PDB”, “CRD”, “XTC”, “TRR”
- Timestep instance
A single frame writer is a special case of a trajectory writer in that it writes only a single coordinate frame to a file, for instance, a pdb or gro file. Unlike trajectory formats, which only contains coordinates, single frame formats contain much more information (e.g. atom and residue names and numbers) and hence it is possible to write selections of atoms in a meaningful way.
W = FrameWriter(filename, **kwargs) W.write(AtomGroup) W.write(Universe)
The blanket kwargs is required so that one can pass the same kind of arguments (filename and numatoms) as for the Trajectory writers. In this way, the simple writer() factory function can be used for all writers.
Trajectory and Frame writers can be used in almost exactly the same manner with the one difference that Frame writers cannot deal with raw Timestep objects.
The following data structures connect reader/writer classes to their format identifiers. They are documented for programmers who want to enhance MDAnalysis; the casual user is unlikely to access them directly.
standard trajectory readers (dict with identifier as key and reader class as value)
readers of files that contain both topology/atom data and coordinates (currently only the keys are used)
formats of readers that can also handle gzip or bzip2 compressed files
frame writers: export to single frame formats such as PDB, gro, crd Signature:
W = FrameWriter(filename) W.write(AtomGroup)
trajectory writers: export frames, typically only saving coordinates Signature:
W = TrajectoryWriter(filename,numatoms,**kwargs) W.write_next_timestep(TimeStep) W.write(Timestep) W.write(AtomGroup) W.write(Universe)