| MRCFILE(1) | mrcfile | MRCFILE(1) |
mrcfile - mrcfile Documentation
Get started by reading the Overview of mrcfile.py and the Usage Guide.
You can also look at the Source documentation.
mrcfile is a Python implementation of the MRC2014 file format, which is used in structural biology to store image and volume data.
It allows MRC files to be created and opened easily using a very simple API, which exposes the file's header and data as numpy arrays. The code runs in Python 2 and 3 and is fully unit-tested.
This library aims to allow users and developers to read and write standard-compliant MRC files in Python as easily as possible, and with no dependencies on any compiled libraries except numpy. You can use it interactively to inspect files, correct headers and so on, or in scripts and larger software packages to provide basic MRC file I/O functions.
The mrcfile library is available from the Python package index:
pip install mrcfile
Or from conda-forge:
conda install --channel conda-forge mrcfile
It is also included in the ccpem-python environment in the CCP-EM software suite.
The source code (including the full test suite) can be found on GitHub.
The easiest way to open a file is with the mrcfile.open and mrcfile.new functions. These return an MrcFile object which represents an MRC file on disk.
To open an MRC file and read a slice of data:
>>> import mrcfile
>>> with mrcfile.open('tests/test_data/EMD-3197.map') as mrc:
... mrc.data[10,10]
...
array([ 2.58179283, 3.1406002 , 3.64495397, 3.63812137, 3.61837363,
4.0115056 , 3.66981959, 2.07317996, 0.1251585 , -0.87975615,
0.12517013, 2.07319379, 3.66982722, 4.0115037 , 3.61837196,
3.6381247 , 3.64495087, 3.14059472, 2.58178973, 1.92690361], dtype=float32)
To create a new file with a 2D data array, and change some values:
>>> array = np.zeros((5, 5), dtype=np.int8)
>>> with mrcfile.new('tmp.mrc') as mrc:
... mrc.set_data(array)
... mrc.data[1:4,1:4] = 10
... mrc.data
...
array([[ 0, 0, 0, 0, 0],
[ 0, 10, 10, 10, 0],
[ 0, 10, 10, 10, 0],
[ 0, 10, 10, 10, 0],
[ 0, 0, 0, 0, 0]], dtype=int8)
The data will be saved to disk when the file is closed, either automatically at the end of the with block (like a normal Python file object) or manually by calling close(). You can also call flush() to write any changes to disk and keep the file open.
To validate an MRC file:
>>> mrcfile.validate('tests/test_data/EMD-3197.map')
File does not declare MRC format version 20140 or 20141: nversion = 0
False
>>> mrcfile.validate('tmp.mrc')
True
Full documentation is available on Read the Docs.
If you find mrcfile useful in your work, please cite:
Burnley T, Palmer C & Winn M (2017) Recent developments in the CCP-EM software suite. Acta Cryst. D73:469--477. doi: 10.1107/S2059798317007859
Please use the GitHub issue tracker for bug reports and feature requests, or email CCP-EM.
Code contributions are also welcome, please submit pull requests to the GitHub repository.
To run the test suite, go to the top-level project directory (which contains the mrcfile and tests packages) and run python -m unittest tests. (Or, if you have tox installed, run tox.)
The project is released under the BSD licence.
This is a detailed guide to using the mrcfile Python library. For a brief introduction, see the overview.
MRC files can be opened using the mrcfile.new() or mrcfile.open() functions. These return an instance of the MrcFile class, which represents an MRC file on disk and makes the file's header, extended header and data available for read and write access as numpy arrays. :
>>> # First, create a simple dataset
>>> import numpy as np
>>> example_data = np.arange(12, dtype=np.int8).reshape(3, 4)
>>> # Make a new MRC file and write the data to it:
>>> import mrcfile
>>> with mrcfile.new('tmp.mrc') as mrc:
... mrc.set_data(example_data)
...
>>> # The file is now saved on disk. Open it again and check the data:
>>> with mrcfile.open('tmp.mrc') as mrc:
... mrc.data
...
array([[ 0, 1, 2, 3],
[ 4, 5, 6, 7],
[ 8, 9, 10, 11]], dtype=int8)
Alternatively, for even quicker access to MRC data but with minimal control of the file header, you can use the read() and write() functions. These do not return MrcFile objects but instead work directly with numpy arrays:
>>> # First, create a simple dataset
>>> import numpy as np
>>> example_data_2 = np.arange(6, dtype=np.int8).reshape(3, 2)
>>> # Write the data to a new MRC file:
>>> mrcfile.write('tmp2.mrc', example_data_2)
>>> # Read it back:
>>> mrcfile.read('tmp2.mrc')
array([[0, 1],
[2, 3],
[4, 5]], dtype=int8)
All of the functions shown above can also handle gzip- or bzip2-compressed files very easily:
>>> # Make a new gzipped MRC file:
>>> with mrcfile.new('tmp.mrc.gz', compression='gzip') as mrc:
... mrc.set_data(example_data * 2)
...
>>> # Open it again with the normal open function:
>>> with mrcfile.open('tmp.mrc.gz') as mrc:
... mrc.data
...
array([[ 0, 2, 4, 6],
[ 8, 10, 12, 14],
[16, 18, 20, 22]], dtype=int8)
>>> # Same again for bzip2:
>>> with mrcfile.new('tmp.mrc.bz2', compression='bzip2') as mrc:
... mrc.set_data(example_data * 3)
...
>>> # Open it again with the normal read function:
>>> mrcfile.read('tmp.mrc.bz2')
array([[ 0, 3, 6, 9],
[12, 15, 18, 21],
[24, 27, 30, 33]], dtype=int8)
>>> # The write function applies compression automatically based on the file name
>>> mrcfile.write('tmp2.mrc.gz', example_data * 4)
>>> # The new file is opened as a GzipMrcFile object:
>>> with mrcfile.open('tmp2.mrc.gz') as mrc:
... print(mrc)
...
GzipMrcFile('tmp2.mrc.gz', mode='r')
MrcFile objects should be closed when they are finished with, to ensure any changes are flushed to disk and the underlying file object is closed:
>>> mrc = mrcfile.open('tmp.mrc', mode='r+')
>>> # do things...
>>> mrc.close()
As we saw in the examples above, MrcFile objects support Python's with statement, which will ensure the file is closed properly after use (like a normal Python file object). It's generally a good idea to use with if possible, but sometimes when running Python interactively (as in some of these examples), it's more convenient to open a file and keep using it without having to work in an indented block. If you do this, remember to close the file at the end!
There's also a flush() method that writes the MRC data to disk but leaves the file open:
>>> mrc = mrcfile.open('tmp.mrc', mode='r+')
>>> # do things...
>>> mrc.flush() # make sure changes are written to disk
>>> # continue using the file...
>>> mrc.close() # close the file when finished
For most purposes, the top-level functions in mrcfile should be all you need to open MRC files, but it is also possible to directly instantiate MrcFile and its subclasses, GzipMrcFile, Bzip2MrcFile and MrcMemmap:
>>> with mrcfile.mrcfile.MrcFile('tmp.mrc') as mrc:
... mrc
...
MrcFile('tmp.mrc', mode='r')
>>> with mrcfile.gzipmrcfile.GzipMrcFile('tmp.mrc.gz') as mrc:
... mrc
...
GzipMrcFile('tmp.mrc.gz', mode='r')
>>> with mrcfile.bzip2mrcfile.Bzip2MrcFile('tmp.mrc.bz2') as mrc:
... mrc
...
Bzip2MrcFile('tmp.mrc.bz2', mode='r')
>>> with mrcfile.mrcmemmap.MrcMemmap('tmp.mrc') as mrc:
... mrc
...
MrcMemmap('tmp.mrc', mode='r')
MrcFile objects can be opened in three modes: r, r+ and w+. These correspond to the standard Python file modes, so r opens a file in read-only mode:
>>> # The default mode is 'r', for read-only access:
>>> mrc = mrcfile.open('tmp.mrc')
>>> mrc
MrcFile('tmp.mrc', mode='r')
>>> mrc.set_data(example_data)
Traceback (most recent call last):
...
ValueError: MRC object is read-only
>>> mrc.close()
r+ opens it for reading and writing:
>>> # Using mode 'r+' allows read and write access:
>>> mrc = mrcfile.open('tmp.mrc', mode='r+')
>>> mrc
MrcFile('tmp.mrc', mode='r+')
>>> mrc.set_data(example_data)
>>> mrc.data
array([[ 0, 1, 2, 3],
[ 4, 5, 6, 7],
[ 8, 9, 10, 11]], dtype=int8)
>>> mrc.close()
and w+ opens a new, empty file (also for both reading and writing):
>>> # Mode 'w+' creates a new empty file:
>>> mrc = mrcfile.open('empty.mrc', mode='w+')
>>> mrc
MrcFile('empty.mrc', mode='w+')
>>> mrc.data
array([], dtype=int8)
>>> mrc.close()
The new() function is effectively shorthand for open(name, mode='w+'):
>>> # Make a new file
>>> mrc = mrcfile.new('empty.mrc')
Traceback (most recent call last):
...
ValueError: File 'empty.mrc' already exists; set overwrite=True to overwrite it
>>> # Ooops, we've already got a file with that name!
>>> # If we're sure we want to overwrite it, we can try again:
>>> mrc = mrcfile.new('empty.mrc', overwrite=True)
>>> mrc
MrcFile('empty.mrc', mode='w+')
>>> mrc.close()
Normally, if an MRC file is badly invalid, an exception is raised when the file is opened. This can be a problem if we want to, say, open a file and fix a header problem. To deal with this situation, open() and mmap() provide an optional permissive argument. If this is set to True, problems with the file will cause warnings to be issued (using Python's warnings module) instead of raising exceptions, and the file will continue to be interpreted as far as possible.
Let's see an example. First we'll deliberately make an invalid file:
>>> # Make a new file and deliberately make a mistake in the header
>>> with mrcfile.new('invalid.mrc') as mrc:
... mrc.header.map = b'map ' # standard requires b'MAP '
...
Now when we try to open the file, an exception is raised:
>>> # Opening an invalid file raises an exception:
>>> mrc = mrcfile.open('invalid.mrc')
Traceback (most recent call last):
...
ValueError: Map ID string not found - not an MRC file, or file is corrupt
If we use permissive mode, we can open the file, and we'll see a warning about the problem (except that here, we have to catch the warning and print the message manually, because warnings don't play nicely with doctests!):
>>> # Opening in permissive mode succeeds, with a warning:
>>> with warnings.catch_warnings(record=True) as w:
... mrc = mrcfile.open('invalid.mrc', permissive=True)
... print(w[0].message)
...
Map ID string not found - not an MRC file, or file is corrupt
Now let's fix the file:
>>> # Fix the invalid file by correcting the header
>>> with mrcfile.open('invalid.mrc', mode='r+', permissive=True) as mrc:
... mrc.header.map = mrcfile.constants.MAP_ID
...
And now we should be able to open the file again normally:
>>> # Now we don't need permissive mode to open the file any more:
>>> mrc = mrcfile.open('invalid.mrc')
>>> mrc.close()
The problems that can cause an exception when opening an MRC file are:
In the last two cases, the data block will not be read and the data attribute will be set to None.
Fixing invalid files can be quite complicated! This usage guide might be expanded in future to explain how to analyse and fix problems, or the library itself might be improved to fix certain problems automatically. For now, if you have trouble with an invalid file, inspecting the code in this library might help you to work out how to approach the problem (start with MrcInterpreter._read_header()), or you could try asking on the CCP-EM mailing list for advice.
mrcfile follows the Python / C-style convention for axis ordering. This means that the first index is the slowest axis (typically Z for volume data or Y for images) and the last index is the fastest axis (typically X), and the numpy arrays are C-contiguous:
>>> data = mrcfile.read('tmp.mrc')
>>> data
array([[ 0, 1, 2, 3],
[ 4, 5, 6, 7],
[ 8, 9, 10, 11]], dtype=int8)
>>> data[1, 0] # x = 0, y = 1
4
>>> data[2, 3] # x = 3, y = 2
11
>>> data.flags.c_contiguous
True
>>> data.flags.f_contiguous
False
Note that this axis order is the opposite of the FORTRAN-style convention that is used by some other software in structural biology. This can cause confusing errors!
The header and data arrays can be accessed using the header, extended_header and data attributes:
>>> mrc = mrcfile.open('tmp.mrc')
>>> mrc.header
rec.array((4, 3, 1, ...),
dtype=[('nx', ...)])
>>> mrc.extended_header
array([],
dtype='|V1')
>>> mrc.data
array([[ 0, 1, 2, 3],
[ 4, 5, 6, 7],
[ 8, 9, 10, 11]], dtype=int8)
>>> mrc.close()
These attributes are read-only and cannot be assigned to directly, but (unless the file mode is r) the arrays can be modified in-place:
>>> mrc = mrcfile.open('tmp.mrc', mode='r+')
>>> # A new data array cannot be assigned directly to the data attribute
>>> mrc.data = np.ones_like(example_data)
Traceback (most recent call last):
...
AttributeError: can't set attribute
>>> # But the data can be modified by assigning to a slice or index
>>> mrc.data[0, 0] = 10
>>> mrc.data
array([[10, 1, 2, 3],
[ 4, 5, 6, 7],
[ 8, 9, 10, 11]], dtype=int8)
>>> # All of the data values can be replaced this way, as long as the data
>>> # size, shape and type are not changed
>>> mrc.data[:] = np.ones_like(example_data)
>>> mrc.data
array([[1, 1, 1, 1],
[1, 1, 1, 1],
[1, 1, 1, 1]], dtype=int8)
>>> mrc.close()
To replace the data or extended header completely, call the set_data() and set_extended_header() methods:
>>> mrc = mrcfile.open('tmp.mrc', mode='r+')
>>> data_3d = np.linspace(-1000, 1000, 20, dtype=np.int16).reshape(2, 2, 5)
>>> mrc.set_data(data_3d)
>>> mrc.data
array([[[-1000, -894, -789, -684, -578],
[ -473, -368, -263, -157, -52]],
[[ 52, 157, 263, 368, 473],
[ 578, 684, 789, 894, 1000]]], dtype=int16)
>>> # Setting a new data array updates the header dimensions to match
>>> mrc.header.nx
array(5, dtype=int32)
>>> mrc.header.ny
array(2, dtype=int32)
>>> mrc.header.nz
array(2, dtype=int32)
>>> # We can also set the extended header in the same way
>>> string_array = np.fromstring(b'The extended header can hold any kind of numpy array', dtype='S52')
>>> mrc.set_extended_header(string_array)
>>> mrc.extended_header
array([b'The extended header can hold any kind of numpy array'],
dtype='|S52')
>>> # Setting the extended header updates the header's nsymbt field to match
>>> mrc.header.nsymbt
array(52, dtype=int32)
>>> mrc.close()
Note that setting an extended header does not automatically set or change the header's exttyp field. You should set this yourself to identify the type of extended header you are using.
For a quick overview of the contents of a file's header, call print_header():
>>> with mrcfile.open('tmp.mrc') as mrc:
... mrc.print_header()
...
nx : 5
ny : 2
nz : 2
mode : 1
nxstart ...
The voxel (or pixel) size in the file can be accessed using the voxel_size attribute, which returns a numpy record array with three fields, x, y and z, for the voxel size in each dimension:
>>> with mrcfile.open('tmp.mrc') as mrc:
... mrc.voxel_size
...
rec.array((0., 0., 0.),
dtype=[('x', '<f4'), ('y', '<f4'), ('z', '<f4')])
In a new file, the voxel size is zero by default. To set the voxel size, you can assign to the voxel_size attribute, using a single number (for an isotropic voxel size), a 3-tuple or a single-item record array with x, y and z fields (which must be in that order):
>>> mrc = mrcfile.open('tmp.mrc', mode='r+')
>>> # Set a new isotropic voxel size:
>>> mrc.voxel_size = 1.0
>>> mrc.voxel_size
rec.array((1., 1., 1.),
dtype=[('x', '<f4'), ('y', '<f4'), ('z', '<f4')])
>>> # Set an anisotropic voxel size using a tuple:
>>> mrc.voxel_size = (1.0, 2.0, 3.0)
>>> mrc.voxel_size
rec.array((1., 2., 3.),
dtype=[('x', '<f4'), ('y', '<f4'), ('z', '<f4')])
>>> # And set a different anisotropic voxel size using a record array:
>>> mrc.voxel_size = np.rec.array(( 4., 5., 6.), dtype=[('x', '<f4'), ('y', '<f4'), ('z', '<f4')])
>>> mrc.voxel_size
rec.array((4., 5., 6.),
dtype=[('x', '<f4'), ('y', '<f4'), ('z', '<f4')])
>>> mrc.close()
The sizes are not stored directly in the MRC header, but are calculated when required from the header's cell and grid size fields. The voxel size can therefore be changed by altering the cell size:
>>> mrc = mrcfile.open('tmp.mrc', mode='r+')
>>> # Check the current voxel size in X:
>>> mrc.voxel_size.x
array(4., dtype=float32)
>>> # And check the current cell dimensions:
>>> mrc.header.cella
rec.array((20., 10., 6.),
dtype=[('x', '<f4'), ('y', '<f4'), ('z', '<f4')])
>>> # Now change the cell's X length:
>>> mrc.header.cella.x = 10
>>> # And we see the voxel size has also changed:
>>> mrc.voxel_size.x
array(2., dtype=float32)
>>> mrc.close()
Equivalently, the cell size will be changed if a new voxel size is given:
>>> mrc = mrcfile.open('tmp.mrc', mode='r+')
>>> # Check the current cell dimensions:
>>> mrc.header.cella
rec.array((10., 10., 6.),
dtype=[('x', '<f4'), ('y', '<f4'), ('z', '<f4')])
>>> # Set a new voxel size:
>>> mrc.voxel_size = 1.0
>>> # And our cell size has been updated:
>>> mrc.header.cella
rec.array((5., 2., 1.),
dtype=[('x', '<f4'), ('y', '<f4'), ('z', '<f4')])
>>> mrc.close()
Because the voxel size array is calculated on demand, assigning back to it wouldn't work so it's flagged as read-only:
>>> mrc = mrcfile.open('tmp.mrc', mode='r+')
>>> # This doesn't work
>>> mrc.voxel_size.x = 2.0
Traceback (most recent call last):
...
ValueError: assignment destination is read-only
>>> # But you can do this
>>> vsize = mrc.voxel_size.copy()
>>> vsize.x = 2.0
>>> mrc.voxel_size = vsize
>>> mrc.voxel_size
rec.array((2., 1., 1.),
dtype=[('x', '<f4'), ('y', '<f4'), ('z', '<f4')])
>>> mrc.close()
Note that the calculated voxel size will change if the grid size is changed by replacing the data array:
>>> mrc = mrcfile.open('tmp.mrc', mode='r+')
>>> # Check the current voxel size:
>>> mrc.voxel_size
rec.array((2., 1., 1.),
dtype=[('x', '<f4'), ('y', '<f4'), ('z', '<f4')])
>>> # And the current data dimensions:
>>> mrc.data.shape
(2, 2, 5)
>>> # Replace the data with an array with a different shape:
>>> mrc.set_data(example_data)
>>> mrc.data.shape
(3, 4)
>>> # ...and the voxel size has changed:
>>> mrc.voxel_size
rec.array((2.5, 0.6666667, 1.),
dtype=[('x', '<f4'), ('y', '<f4'), ('z', '<f4')])
>>> mrc.close()
When a new data array is given (using set_data() or the data argument to mrcfile.new()), the header is automatically updated to ensure the file is is valid:
>>> mrc = mrcfile.open('tmp.mrc', mode='r+')
>>> # Check the current data shape and header dimensions match
>>> mrc.data.shape
(3, 4)
>>> mrc.header.nx
array(4, dtype=int32)
>>> mrc.header.nx == mrc.data.shape[-1] # X axis is always the last in shape
True
>>> # Let's also check the maximum value recorded in the header
>>> mrc.header.dmax
array(11., dtype=float32)
>>> mrc.header.dmax == mrc.data.max()
True
>>> # Now set a data array with a different shape, and check the header again
>>> mrc.set_data(data_3d)
>>> mrc.data.shape
(2, 2, 5)
>>> mrc.header.nx
array(5, dtype=int32)
>>> mrc.header.nx == mrc.data.shape[-1]
True
>>> # The data statistics are updated as well
>>> mrc.header.dmax
array(1000., dtype=float32)
>>> mrc.header.dmax == mrc.data.max()
True
>>> mrc.close()
If the data array is modified in place, for example by editing values or changing the shape or dtype attributes, the header will no longer be correct:
>>> mrc = mrcfile.open('tmp.mrc', mode='r+')
>>> mrc.data.shape
(2, 2, 5)
>>> # Change the data shape in-place and check the header
>>> mrc.data.shape = (5, 4)
>>> mrc.header.nx == mrc.data.shape[-1]
False
>>> # We'll also change some values and check the data statistics
>>> mrc.data[2:] = 0
>>> mrc.data.max()
0
>>> mrc.header.dmax == mrc.data.max()
False
>>> mrc.close()
Note that the header is deliberately not updated automatically except when set_data() is called, so if you need to override any of the automatic header values you can do.
To keep the header in sync with the data, three methods can be used to update the header:
The file we just saved had an invalid header, but of course, that's what's used by mrcfile to work out how to read the file from disk! When we open the file again, our change to the shape has disappeared:
>>> mrc = mrcfile.open('tmp.mrc', mode='r+')
>>> mrc.data.shape
(2, 2, 5)
>>> # Let's change the shape again, as we did before
>>> mrc.data.shape = (5, 4)
>>> mrc.header.nx == mrc.data.shape[-1]
False
>>> # Now let's update the dimensions:
>>> mrc.update_header_from_data()
>>> mrc.header.nx
array(4, dtype=int32)
>>> mrc.header.nx == mrc.data.shape[-1]
True
>>> # The data statistics are still incorrect:
>>> mrc.header.dmax
array(1000., dtype=float32)
>>> mrc.header.dmax == mrc.data.max()
False
>>> # So let's update those as well:
>>> mrc.update_header_stats()
>>> mrc.header.dmax
array(0., dtype=float32)
>>> mrc.header.dmax == mrc.data.max()
True
>>> mrc.close()
In general, if you're changing the shape, type or endianness of the data, it's easiest to use set_data() and the header will be kept up to date for you. If you start changing values in the data, remember that the statistics in the header will be out of date until you call update_header_stats() or reset_header_stats().
MRC files can be used to store several types of data: single images, image stacks, volumes and volume stacks. These are distinguished by the dimensionality of the data array and the space group number (the header's ispg field):
| Data type | Dimensions | Space group |
| Single image | 2 | 0 |
| Image stack | 3 | 0 |
| Volume | 3 | 1--230 (1 for normal EM data) |
| Volume stack | 4 | 401--630 (401 for normal EM data) |
MrcFile objects have methods to allow easy identification of the data type: is_single_image(), is_image_stack(), is_volume() and is_volume_stack().
>>> mrc = mrcfile.open('tmp.mrc')
>>> # The file currently contains two-dimensional data
>>> mrc.data.shape
(5, 4)
>>> len(mrc.data.shape)
2
>>> # This is intepreted as a single image
>>> mrc.is_single_image()
True
>>> mrc.is_image_stack()
False
>>> mrc.is_volume()
False
>>> mrc.is_volume_stack()
False
>>> mrc.close()
If a file already contains image or image stack data, new three-dimensional data is treated as an image stack; otherwise, 3D data is treated as a volume by default:
>>> mrc = mrcfile.open('tmp.mrc', mode='r+')
>>> # New 3D data in an existing image file is treated as an image stack:
>>> mrc.set_data(data_3d)
>>> len(mrc.data.shape)
3
>>> mrc.is_volume()
False
>>> mrc.is_image_stack()
True
>>> int(mrc.header.ispg)
0
>>> mrc.close()
>>> # But normally, 3D data is treated as a volume:
>>> mrc = mrcfile.new('tmp.mrc', overwrite=True)
>>> mrc.set_data(data_3d)
>>> mrc.is_volume()
True
>>> mrc.is_image_stack()
False
>>> int(mrc.header.ispg)
1
>>> mrc.close()
Call set_image_stack() and set_volume() to change the interpretation of 3D data. (Note: as well as changing ispg, these methods also change mz to be 1 for image stacks and equal to nz for volumes.)
>>> mrc = mrcfile.open('tmp.mrc', mode='r+')
>>> # Change the file to represent an image stack:
>>> mrc.set_image_stack()
>>> mrc.is_volume()
False
>>> mrc.is_image_stack()
True
>>> int(mrc.header.ispg)
0
>>> # And now change it back to representing a volume:
>>> mrc.set_volume()
>>> mrc.is_volume()
True
>>> mrc.is_image_stack()
False
>>> int(mrc.header.ispg)
1
>>> mrc.close()
Note that the MRC format allows the data axes to be swapped using the header's mapc, mapr and maps fields. This library does not attempt to swap the axes and simply assigns the columns to X, rows to Y and sections to Z. (The data array is indexed in C style, so data values can be accessed using mrc.data[z][y][x].) In general, EM data is written using the default axes, but crystallographic data files might use swapped axes in certain space groups -- if this might matter to you, you should check the mapc, mapr and maps fields after opening the file and consider transposing the data array if necessary.
Various numpy data types can be used for MRC data arrays. The conversions to MRC mode numbers are:
| Data type | MRC mode |
| float16 | 12 (see note below) |
| float32 | 2 |
| int8 | 0 |
| int16 | 1 |
| uint8 | 6 (note that data will be widened to 16 bits in the file) |
| uint16 | 6 |
| complex64 | 4 |
(Mode 3 and the proposed 4-bit mode 101 are not supported since there are no corresponding numpy dtypes.)
Note that mode 12 is a proposed extension to the MRC2014 format and is not yet widely supported by other software. If you need to write float16 data to MRC files in a compatible way, you should cast to float32 first and use mode 2.
No other data types are accepted, including integer types of more than 16 bits, or float types of more than 32 bits. Many numpy array creation routines use int64 or float64 dtypes by default, which means you will need to give a dtype argument to ensure the array can be used in an MRC file:
>>> mrc = mrcfile.open('tmp.mrc', mode='r+')
>>> # This does not work
>>> mrc.set_data(np.zeros((4, 5)))
Traceback (most recent call last):
...
ValueError: dtype 'float64' cannot be converted to an MRC file mode
>>> # But this does
>>> mrc.set_data(np.zeros((4, 5), dtype=np.int16))
>>> mrc.data
array([[0, 0, 0, 0, 0],
[0, 0, 0, 0, 0],
[0, 0, 0, 0, 0],
[0, 0, 0, 0, 0]], dtype=int16)
>>> mrc.close()
Warning: be careful if you have an existing numpy array in float64, int64 or int32 data types. If they try to convert them into one of the narrower types supported by mrcfile and they contain values outside the range of the target type, the values will silently overflow. For floating point formats this can lead to inf values, and with integers it can lead to entirely meaningless values. A full discussion of this issue is outside the scope of this guide; see the numpy documentation for more information.
mrcfile provides two ways of improving performance when handling large files: memory mapping and asynchronous (background) opening. Memory mapping treats the file's data on disk as if it is already in memory, and only actually loads the data in small chunks when it is needed. Asynchronous opening uses a separate thread to open the file, allowing the main thread to carry on with other work while the file is loaded from disk in parallel.
Which technique is better depends on what you intend to do with the file and the characteristics of your computer, and it's usually worth testing both approaches and seeing what works best for your particular task. In general, memory mapping gives better performance when dealing with a single file, particularly if the file is very large. If you need to process several files, asynchronous opening can be faster because you can work on one file while loading the next one.
With very large files, it might be helpful to use the mrcfile.mmap() function to open the file, which will open the data as a memory-mapped numpy array. The contents of the array are only read from disk as needed, so this allows large files to be opened very quickly. Parts of the data can then be read and written by slicing the array:
>>> # Let's make a new file to work with (only small for this example!)
>>> mrcfile.write('maybe_large.mrc', example_data)
>>> # Open the file in memory-mapped mode
>>> mrc = mrcfile.mmap('maybe_large.mrc', mode='r+')
>>> # Now read part of the data by slicing
>>> mrc.data[1:3]
memmap([[ 4, 5, 6, 7],
[ 8, 9, 10, 11]], dtype=int8)
>>> # Set some values by assigning to a slice
>>> mrc.data[1:3,1:3] = 0
>>> # Read the entire array - with large files this might take a while!
>>> mrc.data[:]
memmap([[ 0, 1, 2, 3],
[ 4, 0, 0, 7],
[ 8, 0, 0, 11]], dtype=int8)
>>> mrc.close()
To create new large, empty files quickly, use the mrcfile.new_mmap() function. This creates an empty file with a given shape and data mode. An optional fill value can be provided but filling a very large mmap array can take a long time, so it's best to use this only when needed. If you plan to fill the array with other data anyway, it's better to leave the fill value as None. A typical use case would be to create a new file and then fill it slice by slice:
>>> # Make a new, empty memory-mapped MRC file
>>> mrc = mrcfile.new_mmap('mmap.mrc', shape=(3, 3, 4), mrc_mode=0)
>>> # Fill each slice with a different value
>>> for val in range(len(mrc.data)):
... mrc.data[val] = val
...
>>> mrc.data[:]
memmap([[[0, 0, 0, 0],
[0, 0, 0, 0],
[0, 0, 0, 0]],
[[1, 1, 1, 1],
[1, 1, 1, 1],
[1, 1, 1, 1]],
[[2, 2, 2, 2],
[2, 2, 2, 2],
[2, 2, 2, 2]]], dtype=int8)
When processing several files in a row, asynchronous (background) opening can improve performance by allowing you to open multiple files in parallel. The mrcfile.open_async() function starts a background thread to open a file, and returns a FutureMrcFile object which you can call later to get the file after it's been opened:
>>> # Open the first example file
>>> mrc1 = mrcfile.open('maybe_large.mrc')
>>> # Start opening the second example file before we process the first
>>> future_mrc2 = mrcfile.open_async('tmp.mrc.gz')
>>> # Now we'll do some calculations with the first file
>>> mrc1.data.sum()
36
>>> # Get the second file from its "Future" container ('result()' will wait
>>> # until the file is ready)
>>> mrc2 = future_mrc2.result()
>>> # Before we process the second file, we'll start the third one opening
>>> future_mrc3 = mrcfile.open_async('tmp.mrc.bz2')
>>> mrc2.data.max()
22
>>> # Finally, we'll get the third file and process it
>>> mrc3 = future_mrc3.result()
>>> mrc3.data
array([[ 0, 3, 6, 9],
[12, 15, 18, 21],
[24, 27, 30, 33]], dtype=int8)
As we saw in that example, calling result() will give us the MrcFile from the file opening operation. If the file hasn't been fully opened yet, result() will simply wait until it's ready. To avoid waiting, call done() to check if it's finished.
MRC files can be validated with mrcfile.validate(), which prints an explanation of what is happening and also returns True if the file is valid or False otherwise:
>>> mrcfile.validate('tmp.mrc')
Checking if tmp.mrc is a valid MRC2014 file...
File appears to be valid.
True
This works equally well for gzip- or bzip2-compressed files:
>>> mrcfile.validate('tmp.mrc.gz')
Checking if tmp.mrc.gz is a valid MRC2014 file...
File appears to be valid.
True
>>> mrcfile.validate('tmp.mrc.bz2')
Checking if tmp.mrc.bz2 is a valid MRC2014 file...
File appears to be valid.
True
Errors will cause messages to be printed to the console, and validate() will return False:
>>> # Let's make a file which is valid except for the header's mz value
>>> with mrcfile.new('tmp.mrc', overwrite=True) as mrc:
... mrc.set_data(example_data)
... mrc.header.mz = -1
...
>>> # Now it should fail validation and print a helpful message
>>> mrcfile.validate('tmp.mrc')
Checking if tmp.mrc is a valid MRC2014 file...
Header field 'mz' is negative
False
(More serious errors might also cause warnings to be printed to sys.stderr.)
Normally, messages are printed to sys.stdout (as normal for Python print() calls). validate() has an optional print_file argument which allows any text stream to be used for the output instead:
>>> # Create a text stream to capture the output
>>> import io
>>> output = io.StringIO()
>>> # Now validate the file...
>>> mrcfile.validate('tmp.mrc', print_file=output)
False
>>> # ...and check the output separately
>>> print(output.getvalue().strip())
Checking if tmp.mrc is a valid MRC2014 file...
Header field 'mz' is negative
Behind the scenes, mrcfile.validate() opens the file in permissive mode using mrcfile.open() and then calls MrcFile.validate(). If you already have an MrcFile open, you can call its validate() method directly to check the file -- but note that the file size test might be inaccurate unless you call flush() first. To ensure the file is completely valid, it's best to flush or close the file and then validate it from scratch using mrcfile.validate().
If you find that a file created with this library is invalid, and you haven't altered anything in the header in a way that might cause problems, please file a bug report on the issue tracker!
Some mrcfile functionality is available directly from the command line, via scripts that are installed along with the package, or in some cases by running modules with python -m.
(If you've downloaded the source code instead of installing via pip, run pip install <path-to-mrcfile> or python setup.py install to make the command line scripts available.)
MRC files can be validated with the mrcfile-validate script:
$ mrcfile-validate tests/test_data/EMD-3197.map Checking if tests/test_data/EMD-3197.map is a valid MRC2014 file... File does not declare MRC format version 20140 or 20141: nversion = 0 $ # Exit status is 1 if file is invalid $ echo $? 1
This script wraps the mrcfile.validator module, which can also be called directly:
$ python -m mrcfile.validator valid_file.mrc Checking if valid_file.mrc is a valid MRC2014 file... File appears to be valid. $ echo $? 0
Multiple file names can be passed to either form of the command, and because these commands call mrcfile.validate() behind the scenes, gzip- and bzip2-compressed files can be validated as well:
$ mrcfile-validate valid_file_1.mrc valid_file_2.mrc.gz valid_file_3.mrc.bz2 Checking if valid_file_1.mrc is a valid MRC2014 file... File appears to be valid. Checking if valid_file_2.mrc is a valid MRC2014 file... File appears to be valid. Checking if valid_file_3.mrc is a valid MRC2014 file... File appears to be valid.
MRC file headers can be printed to the console with the mrcfile-header script:
$ mrcfile-header tests/test_data/EMD-3197.map MRC header for tests/test_data/EMD-3197.map: nx : 20 ny : 20 nz : 20 mode : 2 nxstart : -2 nystart : 0 nzstart : 0 mx : 20 my : 20 mz : 20 cella : (228.0, 228.0, 228.0) cellb : (90.0, 90.0, 90.0) ... ...
Like mrcfile-validate, this also works for multiple files. If you want to access the same functionality from within Python, call print_header() on an open MrcFile object, or mrcfile.command_line.print_headers() with a list of file names.
The following classes are provided by the mrcfile.py library:
Attributes:
Methods:
A pure Python implementation of the MRC2014 file format.
For a full introduction and documentation, see http://mrcfile.readthedocs.io/
Examples assume that this package has been imported as mrcfile and numpy has been imported as np.
To open an MRC file and read a slice of data:
>>> with mrcfile.open('tests/test_data/EMD-3197.map') as mrc:
... mrc.data[10,10]
...
array([ 2.58179283, 3.1406002 , 3.64495397, 3.63812137, 3.61837363,
4.0115056 , 3.66981959, 2.07317996, 0.1251585 , -0.87975615,
0.12517013, 2.07319379, 3.66982722, 4.0115037 , 3.61837196,
3.6381247 , 3.64495087, 3.14059472, 2.58178973, 1.92690361], dtype=float32)
To create a new file with a 2D data array, and change some values:
>>> with mrcfile.new('tmp.mrc') as mrc:
... mrc.set_data(np.zeros((5, 5), dtype=np.int8))
... mrc.data[1:4,1:4] = 10
... mrc.data
...
array([[ 0, 0, 0, 0, 0],
[ 0, 10, 10, 10, 0],
[ 0, 10, 10, 10, 0],
[ 0, 10, 10, 10, 0],
[ 0, 0, 0, 0, 0]], dtype=int8)
The MRC2014 format was described in the Journal of Structural Biology: http://dx.doi.org/10.1016/j.jsb.2015.04.002
The format specification is available on the CCP-EM website: http://www.ccpem.ac.uk/mrc_format/mrc2014.php
This function opens both normal and compressed MRC files. Supported compression formats are: gzip, bzip2.
It is possible to use this function to create new MRC files (using mode w+) but the new() function is more flexible.
This function offers a permissive read mode for attempting to open corrupt or invalid files. In permissive mode, warnings are issued instead of exceptions if problems with the file are encountered. See MrcInterpreter or the usage guide for more information.
This is a convenience function to read the data from an MRC file when there is no need for the file's header information. To read the headers as well, or if you need access to an MrcFile object representing the file, use mrcfile.open() instead.
This is a convenience function to allow data to be quickly written to a file (with optional compression) using just a single function call. However, there is no control over the file's metadata except for optionally setting the voxel size. For more control, or if you need access to an MrcFile object representing the new file, use mrcfile.new() instead.
This allows a file to be opened in the background while the main thread continues with other work. This can be a good way to improve performance if the main thread is busy with intensive computation, but will be less effective if the main thread is itself busy with disk I/O.
Multiple files can be opened in the background simultaneously. However, this implementation is relatively crude; each call to this function will start a new thread and immediately use it to start opening a file. If you try to open many large files at the same time, performance will decrease as all of the threads attempt to access the disk at once. You'll also risk running out of memory to store the data from all the files.
This function returns a FutureMrcFile object, which deliberately mimics the API of the Future object from Python 3's concurrent.futures module. (Future versions of this library might return genuine Future objects instead.)
To get the real MrcFile object from a FutureMrcFile, call result(). This will block until the file has been read and the MrcFile object is ready. To check if the MrcFile is ready without blocking, call running() or done().
This allows much faster opening of large files, because the data is only accessed on disk when a slice is read or written from the data array. See the MrcMemmap class documentation for more information.
Because the memory-mapped data array accesses the disk directly, compressed files cannot be opened with this function. In all other ways, mmap() behaves in exactly the same way as open(). The MrcMemmap object returned by this function can be used in exactly the same way as a normal MrcFile object.
This function is useful for creating very large files. The initial contents of the data array can be set with the fill parameter if needed, but be aware that filling a large array can take a long time.
If fill is not set, the new data array's contents are unspecified and system-dependent. (Some systems fill a new empty mmap with zeros, others fill it with the bytes from the disk at the newly-mapped location.) If you are definitely going to fill the entire array with new data anyway you can safely leave fill as None, otherwise it is advised to use a sensible fill value (or ensure you are on a system that fills new mmaps with a reasonable default value).
The MRC mode to use for the new file. One of 0, 1, 2, 4 or 6, which correspond to numpy dtypes as follows:
The default is 0.
This function first opens the file by calling open() (with permissive=True), then calls validate(), which runs a series of tests to check whether the file complies with the MRC2014 format specification.
If the file is completely valid, this function returns True, otherwise it returns False. Messages explaining the validation result will be printed to sys.stdout by default, but if a text stream is given (using the print_file argument) output will be printed to that instead.
Badly invalid files will also cause warning messages to be issued, which will be written to sys.stderr by default. See the documentation of the warnings module for information on how to suppress or capture warning output.
Because the file is opened by calling open(), gzip- and bzip2-compressed MRC files can be validated easily using this function.
After the file has been opened, it is checked for problems. The tests are:
Module which exports the Bzip2MrcFile class.
MrcFile subclass for handling bzip2-compressed files.
Usage is the same as for MrcFile.
Module for functions used as command line entry points.
The names of the corresponding command line scripts can be found in the entry_points section of setup.py.
This function opens files in permissive mode to allow headers of invalid files to be examined.
Constants used by the mrcfile.py library.
numpy dtypes used by the mrcfile.py library.
The dtypes are defined in a separate module because they do not interact nicely with the from __future__ import unicode_literals feature used in the rest of the package.
Module which exports the FutureMrcFile class.
Object representing an MRC file being opened asynchronously.
This API deliberately mimics a Future object from the concurrent.futures module in Python 3.2+ (which we do not use directly because this code still needs to run in Python 2.7).
This constructor starts a new thread which will invoke the callable given in open_function with the given arguments.
(For internal use only.)
(See concurrent.futures.Future.cancel() for more details. This implementation does not allow jobs to be cancelled.)
(See concurrent.futures.Future.cancelled() for more details. This implementation does not allow jobs to be cancelled.)
(See concurrent.futures.Future.running() for more details.)
(See concurrent.futures.Future.done() for more details.)
(See concurrent.futures.Future.result() for more details.)
(See concurrent.futures.Future.exception() for more details.)
(For internal use only.)
(See concurrent.futures.Future.add_done_callback() for more details.)
Module which exports the GzipMrcFile class.
MrcFile subclass for handling gzipped files.
Usage is the same as for MrcFile.
Module for top-level functions that open MRC files and form the main API of the package.
This function opens both normal and compressed MRC files. Supported compression formats are: gzip, bzip2.
It is possible to use this function to create new MRC files (using mode w+) but the new() function is more flexible.
This function offers a permissive read mode for attempting to open corrupt or invalid files. In permissive mode, warnings are issued instead of exceptions if problems with the file are encountered. See MrcInterpreter or the usage guide for more information.
This is a convenience function to read the data from an MRC file when there is no need for the file's header information. To read the headers as well, or if you need access to an MrcFile object representing the file, use mrcfile.open() instead.
This is a convenience function to allow data to be quickly written to a file (with optional compression) using just a single function call. However, there is no control over the file's metadata except for optionally setting the voxel size. For more control, or if you need access to an MrcFile object representing the new file, use mrcfile.new() instead.
This allows a file to be opened in the background while the main thread continues with other work. This can be a good way to improve performance if the main thread is busy with intensive computation, but will be less effective if the main thread is itself busy with disk I/O.
Multiple files can be opened in the background simultaneously. However, this implementation is relatively crude; each call to this function will start a new thread and immediately use it to start opening a file. If you try to open many large files at the same time, performance will decrease as all of the threads attempt to access the disk at once. You'll also risk running out of memory to store the data from all the files.
This function returns a FutureMrcFile object, which deliberately mimics the API of the Future object from Python 3's concurrent.futures module. (Future versions of this library might return genuine Future objects instead.)
To get the real MrcFile object from a FutureMrcFile, call result(). This will block until the file has been read and the MrcFile object is ready. To check if the MrcFile is ready without blocking, call running() or done().
This allows much faster opening of large files, because the data is only accessed on disk when a slice is read or written from the data array. See the MrcMemmap class documentation for more information.
Because the memory-mapped data array accesses the disk directly, compressed files cannot be opened with this function. In all other ways, mmap() behaves in exactly the same way as open(). The MrcMemmap object returned by this function can be used in exactly the same way as a normal MrcFile object.
This function is useful for creating very large files. The initial contents of the data array can be set with the fill parameter if needed, but be aware that filling a large array can take a long time.
If fill is not set, the new data array's contents are unspecified and system-dependent. (Some systems fill a new empty mmap with zeros, others fill it with the bytes from the disk at the newly-mapped location.) If you are definitely going to fill the entire array with new data anyway you can safely leave fill as None, otherwise it is advised to use a sensible fill value (or ensure you are on a system that fills new mmaps with a reasonable default value).
The MRC mode to use for the new file. One of 0, 1, 2, 4 or 6, which correspond to numpy dtypes as follows:
The default is 0.
Module which exports the MrcFile class.
An object which represents an MRC file.
The header and data are handled as numpy arrays - see MrcObject for details.
MrcFile supports a permissive read mode for attempting to open corrupt or invalid files. See mrcfile.mrcinterpreter.MrcInterpreter or the usage guide for more information.
>>> with MrcFile('tmp.mrc', 'w+') as mrc:
... mrc.set_data(np.zeros((10, 10), dtype=np.int8))
In mode r or r+, the named file is opened from disk and read. In mode w+ a new empty file is created and will be written to disk at the end of the with block (or when flush() or close() is called).
The given file name is opened in the given mode. For mode r or r+ the header, extended header and data are read from the file. For mode w+ a new file is created with a default header and empty extended header and data arrays.
This override calls MrcInterpreter.close() to ensure the stream is flushed and closed, then closes the file object.
The tests are:
Module which exports the MrcInterpreter class.
An object which interprets an I/O stream as MRC / CCP4 map data.
The header and data are handled as numpy arrays - see MrcObject for details.
MrcInterpreter can be used directly, but it is mostly intended as a superclass to provide common stream-handling functionality. This can be used by subclasses which will handle opening and closing the stream.
This class implements the __enter__() and __exit__() special methods which allow it to be used by the Python context manager in a with block. This ensures that close() is called after the object is finished with.
When reading the I/O stream, a ValueError is raised if the data is invalid in one of the following ways:
MrcInterpreter offers a permissive read mode for handling problematic files. If permissive is set to True and any of the validity checks fails, a warning is issued instead of an exception, and file interpretation continues. If the mode number is invalid or the data block is too small, the data attribute will be set to None. In this case, it might be possible to inspect and correct the header, and then call _read() again to read the data correctly. See the usage guide for more details.
Methods:
Methods relevant to subclasses:
This initialiser reads the stream if it is given. In general, subclasses should call __init__() without giving an iostream argument, then set the _iostream attribute themselves and call _read() when ready.
To use the MrcInterpreter class directly, pass a stream when creating the object (or for a write-only stream, create an MrcInterpreter with no stream, call _create_default_attributes() and set the _iostream attribute directly).
Before calling this method, the stream should be open and positioned at the start of the header. This method will advance the stream to the end of the data block (or the end of the extended header if header_only is True.
The header will be read from the current stream position, and the stream will be advanced by 1024 bytes.
If there is no extended header, a zero-length array is assigned to the extended_header attribute.
If the extended header is recognised as FEI microscope metadata (by 'FEI1' or 'FEI2' in the header's exttyp field), its dtype is set appropriately. Otherwise, the dtype is set as void ('V1').
This method uses information from the header to set the data array's shape and dtype.
This default implementation relies on the stream implementing the readinto() method to avoid copying the new array while creating the mutable bytearray. Subclasses should override this if their stream does not support readinto().
This implementation seeks to the start of the stream, writes the header, extended header and data arrays, and then truncates the stream.
Subclasses should override this implementation for streams which do not support seek() or truncate().
Module which exports the MrcMemmap class.
MrcFile subclass that uses a numpy memmap array for the data.
Using a memmap means that the disk access is done lazily: the data array will only be read or written in small chunks when required. To access the contents of the array, use the array slice operator.
Usage is the same as for MrcFile.
Note that memmap arrays use a fairly small chunk size and so performance could be poor on file systems that are optimised for infrequent large I/O operations.
If required, it is possible to create a very large empty file by creating a new MrcMemmap and then calling _open_memmap() to create the memmap array, which can then be filled slice-by-slice. Be aware that the contents of a new, empty memmap array depend on your platform: the data values could be garbage or zeros.
Note that the file's entire data block must be moved if the extended header size changes. Setting a new extended header can therefore be very time consuming with large files, if the new extended header occupies a different number of bytes than the previous one.
This method first calculates the parameters needed to read the data (block start position, endian-ness, file mode, array shape) and then opens the data as a numpy memmap array.
The array is flagged as read-only before deletion, so if a reference to it has been kept elsewhere, changes to it should no longer be able to change the file contents.
Module which exports the MrcObject class.
An object representing image or volume data in the MRC format.
The header, extended header and data are stored as numpy arrays and exposed as read-only attributes. To replace the data or extended header, call set_data() or set_extended_header(). The header cannot be replaced but can be modified in place.
Voxel size is exposed as a writeable attribute, but is calculated on-the-fly from the header's cella and mx/my/mz fields.
Three-dimensional data can represent either a stack of 2D images, or a 3D volume. This is indicated by the header's ispg (space group) field, which is set to 0 for image data and >= 1 for volume data. The is_single_image(), is_image_stack(), is_volume() and is_volume_stack() methods can be used to identify the type of information stored in the data array. For 3D data, the set_image_stack() and set_volume() methods can be used to switch between image stack and volume interpretations of the data.
If the data contents have been changed, you can use the update_header_from_data() and update_header_stats() methods to make the header consistent with the data. These methods are called automatically if the data array is replaced by calling set_data(). update_header_from_data() is fast, even with very large data arrays, because it only examines the shape and type of the data array. update_header_stats() calculates statistics from all items in the data array and so can be slow for very large arrays. If necessary, the reset_header_stats() method can be called to set the header fields to indicate that the statistics are undetermined.
Attributes:
Methods:
Attributes and methods relevant to subclasses:
This initialiser deliberately avoids creating any arrays and simply sets the header, extended header and data attributes to None. This allows subclasses to call __init__() at the start of their initialisers and then set the attributes themselves, probably by reading from a file, or by calling _create_default_attributes() for a new empty object.
Note that this behaviour might change in future: this initialiser could take optional arguments to allow the header and data to be provided by the caller, or might create the standard empty defaults rather than setting the attributes to None.
The header is initialised with standard file type and version information, default values for some essential fields, and zeros elsewhere. The first text label is also set to indicate the file was created by this module.
If this MrcObject was read from a file and the extended header type was recognised, its dtype will be set appropriately. (Currently the only supported types are 'FEI1' and 'FEI2'.) Otherwise, the dtype will be void (raw data, dtype 'V'). If the actual data type of the extended header is known, the dtype of the array can be changed to match.
The extended header may be modified in place. To replace it completely, call set_extended_header().
If you set the extended header you should also set the header.exttyp field to indicate the type of extended header.
This replaces the current data with the given array (or a copy of it), and updates the header to match the new data dimensions. The data statistics (min, max, mean and rms) stored in the header will also be updated.
The new data array is not checked - it must already be valid for use in an MRC file.
The voxel size is returned as a structured NumPy record array with three fields (x, y and z). For example:
>>> mrc.voxel_size rec.array((0.44825, 0.3925, 0.45874998),
dtype=[('x', '<f4'), ('y', '<f4'), ('z', '<f4')]) >>> mrc.voxel_size.x array(0.44825, dtype=float32)
Note that changing the voxel_size array in-place will not change the voxel size in the file -- to prevent this being overlooked accidentally, the writeable flag is set to False on the voxel_size array.
To set the voxel size, assign a new value to the voxel_size attribute. You may give a single number, a 3-tuple (x, y ,z) or a modified version of the voxel_size array. The following examples are all equivalent:
>>> mrc.voxel_size = 1.0
>>> mrc.voxel_size = (1.0, 1.0, 1.0)
>>> vox_sizes = mrc.voxel_size
>>> vox_sizes.flags.writeable = True
>>> vox_sizes.x = 1.0
>>> vox_sizes.y = 1.0
>>> vox_sizes.z = 1.0
>>> mrc.voxel_size = vox_sizes
This provides a convenient way to get and set the values of the header's nxstart, nystart and nzstart fields. Note that these fields are integers and are measured in voxels, not angstroms. The start locations are returned as a structured NumPy record array with three fields (x, y and z). For example:
>>> mrc.header.nxstart array(0, dtype=int32) >>> mrc.header.nystart array(-21, dtype=int32) >>> mrc.header.nzstart array(-12, dtype=int32) >>> mrc.nstart rec.array((0, -21, -12),
dtype=[('x', '<i4'), ('y', '<i4'), ('z', '<i4')]) >>> mrc.nstart.y array(-21, dtype=int32)
Note that changing the nstart array in-place will not change the values in the file -- to prevent this being overlooked accidentally, the writeable flag is set to False on the nstart array.
To set the start locations, assign a new value to the nstart attribute. You may give a single number, a 3-tuple (x, y ,z) or a modified version of the nstart array. The following examples are all equivalent:
>>> mrc.nstart = -150
>>> mrc.nstart = (-150, -150, -150)
>>> starts = mrc.nstart
>>> starts.flags.writeable = True
>>> starts.x = -150
>>> starts.y = -150
>>> starts.z = -150
>>> mrc.nstart = starts
This method changes the space group number (header.ispg) to zero.
If the space group was previously zero (representing an image stack), this method sets it to one. Otherwise the space group is not changed.
This function updates the header byte order and machine stamp to match the byte order of the data. It also updates the file mode, space group and the dimension fields nx, ny, nz, mx, my and mz.
If the data is 2D, the space group is set to 0 (image stack). For 3D data the space group is not changed, and for 4D data the space group is set to 401 (simple P1 volume stack) unless it is already in the volume stack range (401--630).
This means that new 3D data will be treated as an image stack if the previous data was a single image or image stack, or as a volume if the previous data was a volume or volume stack.
Note that this function does not update the data statistics fields in the header (dmin, dmax, dmean and rms). Use the update_header_stats() function to update the statistics. (This is for performance reasons -- updating the statistics can take a long time for large data sets, but updating the other header information is always fast because only the type and shape of the data array need to be inspected.)
Note that this can take some time with large files, particularly with files larger than the currently available memory.
Up to ten labels are stored in the header as arrays of 80 bytes. This method returns the labels as Python strings, filtered to remove non-printable characters. To access the raw bytes (including any non-printable characters) use the header.label attribute (and note that header.nlabl stores the number of labels currently set).
The new label will be stored after any labels already in the header. If all ten labels are already in use, an exception will be raised.
Future versions of this method might add checks to ensure that labels containing valid text are not overwritten even if the nlabl value is incorrect.
This method runs a series of tests to check whether this object complies strictly with the MRC2014 format specification:
Utility functions used by the other modules in the mrcfile package.
This function calls dtype_from_mode() to get the basic dtype, and then makes sure that the byte order of the new dtype matches the byte order of the header's mode field.
The conversion is as follows:
Note that there is no numpy dtype which corresponds to MRC mode 3.
The mode parameter may be given as a Python scalar, numpy scalar or single-item numpy array.
The conversion is as follows:
Note that modes 3 and 101 are not supported as there is no matching numpy dtype.
Even though this is a one-liner, the details are tricky to get right so things work properly in both Python 2 and 3. It's broken out as a separate function so it can be thoroughly tested.
Module for top-level functions that validate MRC files.
This module is runnable to allow files to be validated easily from the command line.
The return value is used as the process exit code when this function is called by running this module or from the corresponding console_scripts entry point.
This function calls validate() for each file name in the given list.
This function first opens the file by calling open() (with permissive=True), then calls validate(), which runs a series of tests to check whether the file complies with the MRC2014 format specification.
If the file is completely valid, this function returns True, otherwise it returns False. Messages explaining the validation result will be printed to sys.stdout by default, but if a text stream is given (using the print_file argument) output will be printed to that instead.
Badly invalid files will also cause warning messages to be issued, which will be written to sys.stderr by default. See the documentation of the warnings module for information on how to suppress or capture warning output.
Because the file is opened by calling open(), gzip- and bzip2-compressed MRC files can be validated easily using this function.
After the file has been opened, it is checked for problems. The tests are:
Colin Palmer
2023, Science and Technology Facilities Council
| February 14, 2023 | 1.4.3 |