# -*- Mode: python; tab-width: 4; indent-tabs-mode:nil; coding:utf-8 -*-
# vim: tabstop=4 expandtab shiftwidth=4 softtabstop=4
#
# MDAnalysis --- https://www.mdanalysis.org
# Copyright (c) 2006-2017 The MDAnalysis Development Team and contributors
# (see the file AUTHORS for the full list of names)
#
# Released under the GNU Public Licence, v2 or any higher version
#
# Please cite your use of MDAnalysis in published work:
#
# R. J. Gowers, M. Linke, J. Barnoud, T. J. E. Reddy, M. N. Melo, S. L. Seyler,
# D. L. Dotson, J. Domanski, S. Buchoux, I. M. Kenney, and O. Beckstein.
# MDAnalysis: A Python package for the rapid analysis of molecular dynamics
# simulations. In S. Benthall and S. Rostrup editors, Proceedings of the 15th
# Python in Science Conference, pages 102-109, Austin, TX, 2016. SciPy.
# doi: 10.25080/majora-629e541a-00e
#
# N. Michaud-Agrawal, E. J. Denning, T. B. Woolf, and O. Beckstein.
# MDAnalysis: A Toolkit for the Analysis of Molecular Dynamics Simulations.
# J. Comput. Chem. 32 (2011), 2319--2327, doi:10.1002/jcc.21787
#
"""
Generation and Analysis of X3DNA helicoidal parameter profiles --- :mod:`MDAnalysis.analysis.legacy.x3dna`
==========================================================================================================
:Author: Elizabeth Denning
:Year: 2013-2014
:Copyright: GNU Public License v2
.. versionadded:: 0.8
.. versionchanged:: 0.16.0
This module is difficult to test due to restrictions on the X3DNA_ code. It
is therefore considered unmaintained and legacy code. It was moved to the
:mod:`MDAnalysis.analysis.legacy` package (see `issue 906`_)
.. _`issue 906`: https://github.com/MDAnalysis/mdanalysis/issues/906
With the help of this module, X3DNA_ can be run on frames in a trajectory. Data
can be combined and analyzed. X3DNA_ [Lu2003]_ [Lu2008]_ must be installed
separately.
.. rubric:: References
.. [Lu2003] Xiang-Jun Lu & Wilma K. Olson (2003).
3DNA: a software package for the analysis, rebuilding and visualization
for three-dimensional nucleic acid structure
Nucleic Acids Res. 31(17), 5108-21.
.. [Lu2008] Xiang-Jun Lu & Wilma K. Olson (2008).
3DNA: a versatile, integrated software system for the analysis, rebuilding
and visualization of three-dimensional nucleic-acid structures.
Nat Protoc. 3(7), 1213-27.
.. _X3DNA: http://x3dna.org/
Example applications
--------------------
Single structure
~~~~~~~~~~~~~~~~
B-DNA structure::
from MDAnalysis.analysis.x3dna import X3DNA, X3DNAtraj
from MDAnalysis.tests.datafiles import PDB_X3DNA
# set path to your x3dna binary in bashrc file
H = X3DNA(PDB_X3DNA, executable="x3dna_ensemble analyze -b 355d.bps -p pdbfile")
H.run()
H.collect()
H.plot()
Trajectory
~~~~~~~~~~
Analyzing a trajectory::
u = MDAnalysis.Universe(psf, trajectory)
H = X3DNAtraj(u, ...)
H.run()
H.plot()
H.save()
The profiles are available as the attribute :attr:`X3DNAtraj.profiles`
(``H.profiles`` in the example) and are indexed by frame number but
can also be indexed by an arbitrary order parameter as shown in the
next example.
Analysis classes
----------------
.. autoclass:: X3DNA
:members:
:inherited-members:
.. attribute:: profiles
``x3dna_ensemble analyze -b 355d.bps -p pdbfile attribute``:
After running :meth:`X3DNA.collect`, this dict contains all the
X3DNA profiles, indexed by the frame number. If only a single
frame was analyzed then this will be ``X3DNA.profiles[0]``. Note
that the order is random; one needs to sort the keys first.
.. autoclass:: X3DNAtraj
:members:
:inherited-members:
.. attribute:: profiles
After running :meth:`X3DNA.collect`, this dict contains all the
X3DNA profiles, indexed by the frame number.
Utilities
---------
.. autoexception:: ApplicationError
"""
from __future__ import print_function, absolute_import
from six.moves import range
import glob
import os
import errno
import shutil
import warnings
import os.path
import subprocess
import tempfile
import textwrap
from collections import OrderedDict
import numpy as np
import matplotlib.pyplot as plt
from MDAnalysis import ApplicationError
from MDAnalysis.lib.util import which, realpath, asiterable
import logging
logger = logging.getLogger("MDAnalysis.analysis.x3dna")
def mean_std_from_x3dnaPickle(profile):
"""Get mean and standard deviation of helicoidal parameters from a saved `profile`.
The `profile` should have been saved with :meth:`BaseX3DNA.save`. Then
load it with ::
profile = cPickle.load(open("x3dna.pickle"))
h_mean, h_std = mean_std_from_x3dnaPickle(profile)
Arguments
---------
profile : dict
A :attr:`X3DNA.profiles` dict with results from the
:class:`X3DNA` analysis.
Returns
-------
(list, list)
The tuple contains two lists with the means and the standard deviations
for the helicoidal parameters. The order for both lists is ``[shear,
stretch, stagger, buckle, propeller, opening, shift, slide, rise, tilt,
roll, twist]``.
"""
if profile.x3dna_param is False:
bp_shear, bp_stretch, bp_stagger, bp_rise, bp_shift, bp_slide, bp_buckle, bp_prop, bp_open, bp_tilt, bp_roll,\
bp_twist = [], [], [], [], [], [], [], [], [], [], [], []
for i in range(len(profile)):
bp_shear.append(profile.values()[i].Shear)
bp_stretch.append(profile.values()[i].Stretch)
bp_stagger.append(profile.values()[i].Stagger)
bp_buckle.append(profile.values()[i].Buckle)
bp_prop.append(profile.values()[i].Propeller)
bp_open.append(profile.values()[i].Opening)
bp_rise.append(profile.values()[i].Rise)
bp_shift.append(profile.values()[i].Shift)
bp_slide.append(profile.values()[i].Slide)
bp_tilt.append(profile.values()[i].Tilt)
bp_roll.append(profile.values()[i].Roll)
bp_twist.append(profile.values()[i].Twist)
bp_shear, bp_stretch, bp_stagger, bp_rise, bp_shift, bp_slide, bp_buckle, bp_prop, bp_open, bp_tilt, bp_roll,\
bp_twist = np.array(bp_shear), np.array(bp_stretch), np.array(bp_stagger), np.array(bp_rise),\
np.array(bp_shift), np.array(bp_slide), np.array(bp_buckle), np.array(bp_prop), \
np.array(bp_open), np.array(bp_tilt), np.array(bp_roll), np.array(bp_twist)
na_avg, na_std = [], []
for j in range(len(bp_shear[0])):
na_avg.append([
np.mean(bp_shear[:, j]), np.mean(bp_stretch[:, j]), np.mean(bp_stagger[:, j]),
np.mean(bp_buckle[:, j]), np.mean(bp_prop[:, j]), np.mean(bp_open[:, j]),
np.mean(bp_shift[:, j]), np.mean(bp_slide[:, j]), np.mean(bp_rise[:, j]),
np.mean(bp_tilt[:, j]), np.mean(bp_roll[:, j]), np.mean(bp_twist[:, j])])
na_std.append([
np.std(bp_shear[:, j]), np.std(bp_stretch[:, j]), np.std(bp_stagger[:, j]),
np.std(bp_buckle[:, j]), np.std(bp_prop[:, j]), np.std(bp_open[:, j]),
np.std(bp_shift[:, j]), np.std(bp_slide[:, j]), np.std(bp_rise[:, j]),
np.std(bp_tilt[:, j]), np.std(bp_roll[:, j]), np.std(bp_twist[:, j])])
else:
bp_rise, bp_shift, bp_slide, bp_tilt, bp_roll, bp_twist = [], [], [], [], [], [], [], [], [], [], [], []
for i in range(len(profile)):
#print i
bp_rise.append(profile.values()[i].Rise)
bp_shift.append(profile.values()[i].Shift)
bp_slide.append(profile.values()[i].Slide)
bp_tilt.append(profile.values()[i].Tilt)
bp_roll.append(profile.values()[i].Roll)
bp_twist.append(profile.values()[i].Twist)
bp_rise, bp_shift, bp_slide, bp_tilt, bp_roll, bp_twist = np.array(bp_shear),np.array(bp_stretch),\
np.array(bp_stagger), np.array(bp_rise), np.array(bp_shift), np.array(bp_slide),\
np.array(bp_buckle), np.array(bp_prop), np.array(bp_open), np.array(bp_tilt),\
np.array(bp_roll), np.array(bp_twist)
na_avg, na_std = [], []
for j in range(len(bp_shift[0])):
na_avg.append([
np.mean(bp_shift[:, j]), np.mean(bp_slide[:, j]), np.mean(bp_rise[:, j]),
np.mean(bp_tilt[:, j]), np.mean(bp_roll[:, j]), np.mean(bp_twist[:, j])])
na_std.append([
np.std(bp_shift[:, j]), np.std(bp_slide[:, j]), np.std(bp_rise[:, j]),
np.std(bp_tilt[:, j]), np.std(bp_roll[:, j]), np.std(bp_twist[:, j])])
na_avg, na_std = np.array(na_avg), np.array(na_std)
return na_avg, na_std
class BaseX3DNA(object):
"""Baseclass for X3DNA_ analysis, providing plotting and utility functions.
When methods return helicoidal basepair parameter as lists, then the order
is always
====== ==============
index parameter
====== ==============
0 shear
1 stretch
2 stagger
3 buckle
4 propeller
5 opening
6 shift
7 slide
8 rise
9 tilt
10 roll
11 twist
====== ==============
for each nucleic acid pair.
.. _X3DNA: http://x3dna.org
"""
def save(self, filename="x3dna.pickle"):
"""Save :attr:`profiles` as a Python pickle file *filename*.
Load profiles dictionary with ::
import cPickle
profiles = cPickle.load(open(filename))
"""
import cPickle
cPickle.dump(self.profiles, open(filename, "wb"), cPickle.HIGHEST_PROTOCOL)
def mean_std(self):
"""Returns the mean and standard deviation of base parameters.
Returns
-------
(list, list)
The tuple contains two lists with the means and the standard deviations
for the helicoidal parameters. The order for both lists is ``[shear,
stretch, stagger, buckle, propeller, opening, shift, slide, rise, tilt,
roll, twist]``.
"""
bp_shear, bp_stretch, bp_stagger, bp_rise, bp_shift, bp_slide, bp_buckle, bp_prop, bp_open, bp_tilt, bp_roll,\
bp_twist = [], [], [], [], [], [], [], [], [], [], [], []
for i in range(len(self.profiles)):
bp_shear.append(self.profiles.values()[i].Shear)
bp_stretch.append(self.profiles.values()[i].Stretch)
bp_stagger.append(self.profiles.values()[i].Stagger)
bp_buckle.append(self.profiles.values()[i].Buckle)
bp_prop.append(self.profiles.values()[i].Propeller)
bp_open.append(self.profiles.values()[i].Opening)
bp_rise.append(self.profiles.values()[i].Rise)
bp_shift.append(self.profiles.values()[i].Shift)
bp_slide.append(self.profiles.values()[i].Slide)
bp_tilt.append(self.profiles.values()[i].Tilt)
bp_roll.append(self.profiles.values()[i].Roll)
bp_twist.append(self.profiles.values()[i].Twist)
bp_shear, bp_stretch, bp_stagger, bp_rise, bp_shift, bp_slide, bp_buckle, bp_prop, bp_open, bp_tilt, bp_roll,\
bp_twist = np.array(bp_shear), np.array(bp_stretch), np.array(bp_stagger), np.array(bp_rise),\
np.array(bp_shift), np.array(bp_slide), np.array(bp_buckle), np.array(bp_prop),\
np.array(bp_open), np.array(bp_tilt), np.array(bp_roll), np.array(bp_twist)
na_avg, na_std = [], []
for j in range(len(bp_shear[0])):
na_avg.append([
np.mean(bp_shear[:, j]), np.mean(bp_stretch[:, j]), np.mean(bp_stagger[:, j]),
np.mean(bp_buckle[:, j]), np.mean(bp_prop[:, j]), np.mean(bp_open[:, j]),
np.mean(bp_shift[:, j]), np.mean(bp_slide[:, j]), np.mean(bp_rise[:, j]),
np.mean(bp_tilt[:, j]), np.mean(bp_roll[:, j]), np.mean(bp_twist[:, j])])
na_std.append([
np.std(bp_shear[:, j]), np.std(bp_stretch[:, j]), np.std(bp_stagger[:, j]),
np.std(bp_buckle[:, j]), np.std(bp_prop[:, j]), np.std(bp_open[:, j]),
np.std(bp_shift[:, j]), np.std(bp_slide[:, j]), np.std(bp_rise[:, j]),
np.std(bp_tilt[:, j]), np.std(bp_roll[:, j]), np.std(bp_twist[:, j])])
na_avg, na_std = np.array(na_avg), np.array(na_std)
return na_avg, na_std
def mean(self):
"""Returns the mean value for the base parameters.
Returns
-------
list
The list contains the means for the helicoidal parameters. The
order is ``[shear, stretch, stagger, buckle, propeller, opening,
shift, slide, rise, tilt, roll, twist]``.
"""
bp_shear, bp_stretch, bp_stagger, bp_rise, bp_shift, bp_slide, bp_buckle, bp_prop, bp_open, bp_tilt, bp_roll,\
bp_twist = [], [], [], [], [], [], [], [], [], [], [], []
for i in range(len(self.profiles)):
bp_shear.append(self.profiles.values()[i].Shear)
bp_stretch.append(self.profiles.values()[i].Stretch)
bp_stagger.append(self.profiles.values()[i].Stagger)
bp_buckle.append(self.profiles.values()[i].Buckle)
bp_prop.append(self.profiles.values()[i].Propeller)
bp_open.append(self.profiles.values()[i].Opening)
bp_rise.append(self.profiles.values()[i].Rise)
bp_shift.append(self.profiles.values()[i].Shift)
bp_slide.append(self.profiles.values()[i].Slide)
bp_tilt.append(self.profiles.values()[i].Tilt)
bp_roll.append(self.profiles.values()[i].Roll)
bp_twist.append(self.profiles.values()[i].Twist)
bp_shear, bp_stretch, bp_stagger, bp_rise, bp_shift, bp_slide, bp_buckle, bp_prop, bp_open, bp_tilt, bp_roll,\
bp_twist = np.array(bp_shear), np.array(bp_stretch), np.array(bp_stagger), np.array(bp_rise),\
np.array(bp_shift), np.array(bp_slide), np.array(bp_buckle), np.array(bp_prop),\
np.array(bp_open), np.array(bp_tilt), np.array(bp_roll), np.array(bp_twist)
na_avg = []
for j in range(len(bp_shear[0])):
na_avg.append([
np.mean(bp_shear[:, j]), np.mean(bp_stretch[:, j]), np.mean(bp_stagger[:, j]),
np.mean(bp_buckle[:j]), np.mean(bp_prop[:, j]), np.mean(bp_open[:, j]),
np.mean(bp_shift[:, j]), np.mean(bp_slide[:, j]), np.mean(bp_rise[:, j]),
np.mean(bp_tilt[:, j]), np.mean(bp_roll[:, j]), np.mean(bp_twist[:, j])])
na_avg = np.array(na_avg)
return na_avg
def std(self):
"""Returns the standard deviation for the base parameters.
Returns
-------
list
The list contains the standard deviations for the helicoidal
parameters. The order is ``[shear, stretch, stagger, buckle,
propeller, opening, shift, slide, rise, tilt, roll, twist]``.
"""
bp_shear, bp_stretch, bp_stagger, bp_rise, bp_shift, bp_slide, bp_buckle, bp_prop, bp_open, bp_tilt, bp_roll,\
bp_twist = [], [], [], [], [], [], [], [], [], [], [], []
for i in range(len(self.profiles)):
bp_shear.append(self.profiles.values()[i].Shear)
bp_stretch.append(self.profiles.values()[i].Stretch)
bp_stagger.append(self.profiles.values()[i].Stagger)
bp_buckle.append(self.profiles.values()[i].Buckle)
bp_prop.append(self.profiles.values()[i].Propeller)
bp_open.append(self.profiles.values()[i].Opening)
bp_rise.append(self.profiles.values()[i].Rise)
bp_shift.append(self.profiles.values()[i].Shift)
bp_slide.append(self.profiles.values()[i].Slide)
bp_tilt.append(self.profiles.values()[i].Tilt)
bp_roll.append(self.profiles.values()[i].Roll)
bp_twist.append(self.profiles.values()[i].Twist)
bp_shear, bp_stretch, bp_stagger, bp_rise, bp_shift, bp_slide, bp_buckle, bp_prop, bp_open, bp_tilt, bp_roll,\
bp_twist = np.array(bp_shear), np.array(bp_stretch), np.array(bp_stagger), np.array(bp_rise),\
np.array(bp_shift), np.array(bp_slide), np.array(bp_buckle), np.array(bp_prop),\
np.array(bp_open), np.array(bp_tilt), np.array(bp_roll), np.array(bp_twist)
na_std = []
for j in range(len(bp_shear[0])):
na_std.append([
np.std(bp_shear[:, j]), np.std(bp_stretch[:, j]), np.std(bp_stagger[:, j]),
np.std(bp_buckle[:j]), np.std(bp_prop[:, j]), np.std(bp_open[:, j]), np.std(bp_shift[:, j]),
np.std(bp_slide[:, j]), np.std(bp_rise[:, j]), np.std(bp_tilt[:, j]), np.std(bp_roll[:, j]),
np.std(bp_twist[:, j])])
na_std = np.array(na_std)
return na_std
def plot(self, **kwargs):
"""Plot time-averaged base parameters for each basse pair in a 1D graph.
One plot is produced for each parameter. It shows the the mean and
standard deviation for each individual base pair. Each plot is saved to
PNG file with name "<parameter_name>.png".
Parameters
----------
ax : matplotlib.pyplot.Axes (optional)
Provide `ax` to have all plots plotted in the same axes.
"""
na_avg, na_std = self.mean_std()
for k in range(len(na_avg[0])):
ax = kwargs.pop('ax', plt.subplot(111))
x = list(range(1, len(na_avg[:, k]) + 1))
ax.errorbar(x, na_avg[:, k], yerr=na_std[:, k], fmt='-o')
ax.set_xlim(0, len(na_avg[:, k]) + 1)
ax.set_xlabel(r"Nucleic Acid Number")
param = self.profiles.values()[0].dtype.names[k]
if param in ["Shear", "Stretch", "Stagger", "Rise", "Shift", "Slide"]:
ax.set_ylabel("{0!s} ($\AA$)".format((param)))
else:
ax.set_ylabel("{0!s} (deg)".format((param)))
ax.figure.savefig("{0!s}.png".format((param)))
ax.figure.clf()
def sorted_profiles_iter(self):
"""Return an iterator over profiles sorted by frame/order parameter.
The iterator produces tuples ``(q, profile)``. Typically, `q` is the
frame number.
"""
if self.profiles is None:
return
for q in sorted(self.profiles):
yield (q, self.profiles[q])
__iter__ = sorted_profiles_iter
[docs]class X3DNA(BaseX3DNA):
"""Run X3DNA_ on a single frame or a DCD trajectory.
Only a subset of all X3DNA control parameters is supported and can be set
with keyword arguments. For further details on X3DNA_ see the `X3DNA docs`_.
Running X3DNA with the :class:`X3DNA` class is a 3-step process:
1. set up the class with all desired parameters
2. run X3DNA with :meth:`X3DNA.run`
3. collect the data from the output file with :meth:`X3DNA.collect`
The class also provides some simple plotting functions of the collected
data such as :meth:`X3DNA.plot` or :meth:`X3DNA.plot3D`.
When methods return helicoidal basepair parameter as lists, then the order
is always
====== ==============
index parameter
====== ==============
0 shear
1 stretch
2 stagger
3 buckle
4 propeller
5 opening
6 shift
7 slide
8 rise
9 tilt
10 roll
11 twist
====== ==============
.. versionadded:: 0.8
.. _`X3DNA docs`: http://forum.x3dna.org/
"""
def __init__(self, filename, **kwargs):
"""Set up parameters to run X3DNA_ on PDB *filename*.
Parameters
----------
filename : str
The `filename` is used as input for X3DNA in the
:program:`xdna_ensemble` command. It specifies the name of a
PDB coordinate file to be used. This must be in Brookhaven
protein databank format or something closely approximating
this.
executable : str (optional)
Path to the :program:`xdna_ensemble` executable directories
(e.g. ``/opt/x3dna/2.1 and /opt/x3dna/2.1/bin``) must be set
and then added to export in bashrc file. See X3DNA
documentation for set-up instructions.
x3dna_param : bool (optional)
Determines whether base step or base pair parameters will be
calculated. If ``True`` (default) then stacked *base step*
parameters will be analyzed. If ``False`` then stacked *base
pair* parameters will be analyzed.
logfile : str (optional)
Write output from X3DNA to `logfile` (default: "bp_step.par")
See Also
--------
:class:`X3DNAtraj`
"""
# list of temporary files, to be cleaned up on __del__
self.tempfiles = [
"auxiliary.par", "bestpairs.pdb", "bp_order.dat", "bp_helical.par", "cf_7methods.par",
"col_chains.scr", "col_helices.scr", "hel_regions.pdb", "ref_frames.dat", "hstacking.pdb", "stacking.pdb"
]
self.tempdirs = []
self.filename = filename
logger.info("Setting up X3DNA analysis for %(filename)r", vars(self))
# guess executables
self.exe = {}
x3dna_exe_name = kwargs.pop('executable', 'xdna_ensemble')
self.x3dna_param = kwargs.pop('x3dna_param', True)
self.exe['xdna_ensemble'] = which(x3dna_exe_name)
if self.exe['xdna_ensemble'] is None:
errmsg = "X3DNA binary {x3dna_exe_name!r} not found.".format(**vars())
logger.fatal(errmsg)
logger.fatal("%(x3dna_exe_name)r must be on the PATH or provided as keyword argument 'executable'.",
vars())
raise OSError(errno.ENOENT, errmsg)
x3dnapath = os.path.dirname(self.exe['xdna_ensemble'])
self.logfile = kwargs.pop("logfile", "bp_step.par")
if self.x3dna_param is False:
self.template = textwrap.dedent("""x3dna_ensemble analyze -b 355d.bps --one %(filename)r """)
else:
self.template = textwrap.dedent("""find_pair -s %(filename)r stdout |analyze stdin """)
# sanity checks
for program, path in self.exe.items():
if path is None or which(path) is None:
logger.error("Executable %(program)r not found, should have been %(path)r.",
vars())
# results
self.profiles = OrderedDict()
[docs] def run(self, **kwargs):
"""Run X3DNA on the input file."""
inpname = kwargs.pop("inpfile", None)
outname = kwargs.pop("outfile", self.logfile)
x3dnaargs = vars(self).copy()
x3dnaargs.update(kwargs)
x3dna_param = kwargs.pop('x3dna_param', self.x3dna_param)
inp = self.template % x3dnaargs
if inpname:
with open(inpname, "w") as f:
f.write(inp)
logger.debug("Wrote X3DNA input file %r for inspection", inpname)
logger.info("Starting X3DNA on %(filename)r (trajectory: %(dcd)r)", x3dnaargs)
logger.debug("%s", self.exe['xdna_ensemble'])
with open(outname, "w") as output:
x3dna = subprocess.call([inp], shell=True)
with open(outname, "r") as output:
# X3DNA is not very good at setting returncodes so check ourselves
for line in output:
if line.strip().startswith(('*** ERROR ***', 'ERROR')):
x3dna.returncode = 255
break
if x3dna.bit_length != 0:
logger.fatal("X3DNA Failure (%d). Check output %r", x3dna.bit_length, outname)
logger.info("X3DNA finished: output file %(outname)r", vars())
[docs] def collect(self, **kwargs):
"""Parse the output from a X3DNA run into numpy recarrays.
Can deal with outputs containing multiple frames.
The method saves the result as :attr:`X3DNA.profiles`, a dictionary
indexed by the frame number. Each entry is a
:class:`np.recarray`.
If the keyword `outdir` is supplied (e.g. ".") then each profile is
saved to a gzipped data file.
Parameters
----------
run : str, int (optional
identifier, free form [1]
outdir : str (optional)
save output data under `outdir`/`run` if set to any other
value but ``None`` [``None``]
"""
# Shear Stretch Stagger Buckle Prop-Tw Opening Shift Slide Rise Tilt Roll Twist
#0123456789.0123456789.0123456789.0123456789.0123456789.0123456789.123456789.123456789.123456789.123456789.123456789.123456789.123456789.
# 11 22 33 44
#T-A -0.033 -0.176 0.158 -12.177 -8.979 1.440 0.000 0.000 0.000 0.000 0.000 0.000
#C-G -0.529 0.122 -0.002 -7.983 -10.083 -0.091 -0.911 1.375 3.213 -0.766 -4.065 41.492
# only parse bp_step.par
x3dna_output = kwargs.pop("x3dnaout", self.logfile)
run = kwargs.pop("run", 1) # id number
outdir = kwargs.pop("outdir", os.path.curdir)
logger.info("Collecting X3DNA profiles for run with id %s", run)
length = 1 # length of trajectory --- is this really needed?? No... just for info
if '*' in self.filename:
import glob
filenames = glob.glob(self.filename)
length = len(filenames)
if length == 0:
logger.error("Glob pattern %r did not find any files.", self.filename)
raise ValueError("Glob pattern {0!r} did not find any files.".format(self.filename))
logger.info("Found %d input files based on glob pattern %s", length, self.filename)
# one recarray for each frame, indexed by frame number
self.profiles = OrderedDict()
logger.info("Run %s: Reading %d X3DNA profiles from %r", run, length, x3dna_output)
x3dna_profile_no = 0
records = []
with open(x3dna_output, "r") as x3dna:
read_data = False
for line in x3dna:
line = line.rstrip() # preserve columns (FORTRAN output...)
if self.x3dna_param is False:
if line.startswith("# Shear"):
read_data = True
logger.debug("Started reading data")
fields = line.split()
x3dna_profile_no = int(1) # useless int value code based off hole plugin
records = []
continue
if read_data:
if len(line.strip()) != 0:
try:
Sequence, Shear, Stretch, Stagger, Buckle, Propeller, Opening, Shift, Slide, Rise, \
Tilt, Roll, Twist = line.split()
except:
logger.critical("Run %d: Problem parsing line %r", run, line.strip())
logger.exception("Check input file %r.", x3dna_output)
raise
records.append(
[float(Shear), float(Stretch), float(Stagger), float(Buckle), float(Propeller),
float(Opening), float(Shift), float(Slide), float(Rise), float(Tilt), float(Roll),
float(Twist)])
continue
else:
# end of records (empty line)
read_data = False
else:
if line.startswith("# Shift"):
read_data = True
logger.debug("Started reading data")
fields = line.split()
x3dna_profile_no = int(1) # useless int value code based off hole plugin
records = []
continue
if read_data:
if len(line.strip()) != 0:
try:
Sequence, Shift, Slide, Rise, Tilt, Roll, Twist = line.split()
except:
logger.critical("Run %d: Problem parsing line %r", run, line.strip())
logger.exception("Check input file %r.", x3dna_output)
raise
records.append(
[float(Shift), float(Slide), float(Rise), float(Tilt), float(Roll), float(Twist)])
continue
else:
# end of records (empty line)
read_data = False
if self.x3dna_param is False:
frame_x3dna_output = np.rec.fromrecords(records, formats="f8,f8,f8,f8,f8,f8,f8,f8,f8,f8,f8,f8",
names="Shear,Stretch,Stagger,Buckle,Propeller,Opening,"
"Shift,Slide,Rise,Tilt,Roll,Twist")
else:
frame_x3dna_output = np.rec.fromrecords(records, formats="f8,f8,f8,f8,f8,f8",
names="Shift,Slide,Rise,Tilt,Roll,Twist")
# store the profile
self.profiles[x3dna_profile_no] = frame_x3dna_output
logger.debug("Collected X3DNA profile for frame %d (%d datapoints)",
x3dna_profile_no, len(frame_x3dna_output))
# save a profile for each frame (for debugging and scripted processing)
# a tmp folder for each trajectory
if outdir is not None:
rundir = os.path.join(outdir, "run_" + str(run))
os.system("rm -f tmp*.out")
if not os.path.exists(rundir):
os.makedirs(rundir)
frame_x3dna_txt = os.path.join(rundir, "bp_step_{0!s}_{1:04d}.dat.gz".format(run, x3dna_profile_no))
np.savetxt(frame_x3dna_txt, frame_x3dna_output)
logger.debug("Finished with frame %d, saved as %r", x3dna_profile_no, frame_x3dna_txt)
# if we get here then we haven't found anything interesting
if len(self.profiles) == length:
logger.info("Collected X3DNA profiles for %d frames", len(self.profiles))
else:
logger.warning("Missing data: Found %d X3DNA profiles from %d frames.", len(self.profiles), length)
def __del__(self):
for f in self.tempfiles:
try:
os.unlink(f)
except OSError:
pass
for d in self.tempdirs:
shutil.rmtree(d, ignore_errors=True)
[docs]class X3DNAtraj(BaseX3DNA):
"""Analyze all frames in a trajectory.
The :class:`X3DNA` class provides a direct interface to X3DNA. X3DNA itself
has limited support for analysing trajectories but cannot deal with all the
trajectory formats understood by MDAnalysis. This class can take any
universe and feed it to X3DNA. By default it sequentially creates a PDB for
each frame and runs X3DNA on the frame.
"""
def __init__(self, universe, **kwargs):
"""Set up the class.
Parameters
----------
universe : Universe
The input trajectory as part of a
:class:`~MDAnalysis.core.universe.Universe`; the trajectory is
converted to a sequence of PDB files and X3DNA is run on each
individual file. (Use the `start`, `stop`, and `step` keywords
to slice the trajectory.)
selection : str (optional)
MDAnalysis selection string (default: "nucleic") to select the
atoms that should be analyzed.
start : int (optional)
stop : int (optional)
step : int (optional)
frame indices to slice the trajectory as
``universe.trajectory[start, stop, step]``; by default, the whole
trajectory is analyzed.
x3dna_param : bool (optional)
indicates whether stacked bases or stacked base-pairs will be
analyzed. ``True`` is bases and ``False`` is stacked base-pairs
[Default is ``True``].
kwargs : keyword arguments (optional)
All other keywords are passed on to :class:`X3DNA` (see there
for description).
See Also
--------
:class:`X3DNA`
"""
self.universe = universe
self.selection = kwargs.pop("selection", "nucleic")
self.x3dna_param = kwargs.pop('x3dna_param', True)
self.start = kwargs.pop('start', None)
self.stop = kwargs.pop('stop', None)
self.step = kwargs.pop('step', None)
self.x3dna_kwargs = kwargs
# processing
[docs] def run(self, **kwargs):
"""Run X3DNA on the whole trajectory and collect profiles.
Keyword arguments `start`, `stop`, and `step` can be used to only
analyse part of the trajectory. The defaults are the values provided to
the class constructor.
"""
start = kwargs.pop('start', self.start)
stop = kwargs.pop('stop', self.stop)
step = kwargs.pop('step', self.step)
x3dna_param = kwargs.pop('x3dna_param', self.x3dna_param)
x3dna_kw = self.x3dna_kwargs.copy()
x3dna_kw.update(kwargs)
profiles = OrderedDict()
nucleic = self.universe.select_atoms(self.selection)
for ts in self.universe.trajectory[start:stop:step]:
logger.info("X3DNA analysis frame %4d ", ts.frame)
fd, pdbfile = tempfile.mkstemp(suffix=".pdb")
os.close(fd)
nucleic.write(pdbfile)
os.system("find_pair {0!s} 355d.bps".format(pdbfile))
try:
nucleic.write(pdbfile)
x3dna_profiles = self.run_x3dna(pdbfile, **x3dna_kw)
finally:
try:
os.unlink(pdbfile)
except OSError:
pass
if len(x3dna_profiles) != 1:
err_msg = "Got {0} profiles ({1}) --- should be 1 (time step {2})".format(
len(x3dna_profiles), x3dna_profiles.keys(), ts)
logger.error(err_msg)
warnings.warn(err_msg)
profiles[ts.frame] = x3dna_profiles.values()[0]
self.profiles = profiles
[docs] def run_x3dna(self, pdbfile, **kwargs):
"""Run X3DNA on a single PDB file `pdbfile`."""
kwargs['x3dna_param'] = self.x3dna_param
H = X3DNA(pdbfile, **kwargs)
H.run()
H.collect()
return H.profiles