Source code for MDAnalysis.coordinates.ParmEd

# -*- 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
#

"""ParmEd structure I/O --- :mod:`MDAnalysis.coordinates.ParmEd`
================================================================

Read coordinates data from a `ParmEd <https://parmed.github.io/ParmEd/html>`_ :class:`parmed.structure.Structure` with :class:`ParmEdReader` 
into a MDAnalysis Universe. Convert it back to a :class:`parmed.structure.Structure` with 
:class:`ParmEdConverter`.

Example
-------

ParmEd has some neat functions. One such is `HMassRepartition`_. 
This function changes the mass of the hydrogens in your system to your 
desired value. It then adjusts the mass of the atom to which it is 
bonded by the same amount, so that the total mass is unchanged. ::

    >>> import MDAnalysis as mda
    >>> from MDAnalysis.tests.datafiles import PRM
    >>> u = mda.Universe(PRM)
    >>> u.atoms.masses[:10]
    array([14.01 ,  1.008,  1.008,  1.008, 12.01 ,  1.008, 12.01 ,  1.008,
        1.008,  1.008])

We can convert our universe to a ParmEd structure to change our hydrogen masses. ::

    >>> import parmed.tools as pmt
    >>> parm = u.atoms.convert_to('PARMED')
    >>> hmass = pmt.HMassRepartition(parm, 5)  # convert to 5 daltons
    >>> hmass.execute()

We can then convert it back to an MDAnalysis Universe for further analysis. ::

    >>> u2 = mda.Universe(parm)
    >>> u2.atoms.masses[:10]
    array([2.03399992, 5.        , 5.        , 5.        , 8.01799965,
       5.        , 0.034     , 5.        , 5.        , 5.        ])

.. _`HMassRepartition`: http://parmed.github.io/ParmEd/html/parmed.html#hmassrepartition



Classes
-------

.. autoclass:: ParmEdReader
   :members:

.. autoclass:: ParmEdConverter
   :members:


"""
from __future__ import absolute_import
import functools

from . import base
from ..topology.tables import SYMB2Z
from ..core.universe import Universe


[docs]class ParmEdReader(base.SingleFrameReaderBase): """Coordinate reader for ParmEd.""" format = 'PARMED' # Structure.coordinates always in Angstrom units = {'time': None, 'length': 'Angstrom'} @staticmethod def _format_hint(thing): """Can this reader read *thing*? .. versionadded:: 1.0.0 """ try: import parmed as pmd except ImportError: # if we can't import parmed, it's probably not parmed return False else: return isinstance(thing, pmd.Structure) def _read_first_frame(self): self.n_atoms = len(self.filename.atoms) self.ts = ts = self._Timestep(self.n_atoms, **self._ts_kwargs) if self.filename.coordinates is not None: ts._pos = self.filename.coordinates if self.filename.box is not None: # optional field ts.dimensions = self.filename.box else: ts._unitcell = None ts.frame = 0 return ts
MDA2PMD = { 'tempfactor': 'bfactor', 'gbscreen': 'screen', 'altLoc': 'altloc', 'nbindex': 'nb_idx', 'solventradius': 'solvent_radius', 'id': 'number' } def get_indices_from_subset(i, atomgroup=None, universe=None): return atomgroup.index(universe.atoms[i])
[docs]class ParmEdConverter(base.ConverterBase): """Convert MDAnalysis AtomGroup or Universe to ParmEd :class:`~parmed.structure.Structure`. Example ------- .. code-block:: python import parmed as pmd import MDAnalysis as mda from MDAnalysis.tests.datafiles import GRO pgro = pmd.load_file(GRO) mgro = mda.Universe(pgro) parmed_subset = mgro.select_atoms('resname SOL').convert_to('PARMED') """ lib = 'PARMED' units = {'time': None, 'length': 'Angstrom'}
[docs] def convert(self, obj): """Write selection at current trajectory frame to :class:`~parmed.structure.Structure`. Parameters ----------- obj : AtomGroup or Universe or :class:`Timestep` """ try: import parmed as pmd except ImportError: raise ImportError('ParmEd is required for ParmEdConverter but ' 'is not installed. Try installing it with \n' 'pip install parmed') try: # make sure to use atoms (Issue 46) ag_or_ts = obj.atoms except AttributeError: if isinstance(obj, base.Timestep): ag_or_ts = obj.copy() else: raise_from(TypeError("No Timestep found in obj argument"), None) # Check for topology information missing_topology = [] try: names = ag_or_ts.names except (AttributeError, NoDataError): names = itertools.cycle(('X',)) missing_topology.append('names') try: resnames = ag_or_ts.resnames except (AttributeError, NoDataError): resnames = itertools.cycle(('UNK',)) missing_topology.append('resnames') if missing_topology: warnings.warn( "Supplied AtomGroup was missing the following attributes: " "{miss}. These will be written with default values. " "Alternatively these can be supplied as keyword arguments." "".format(miss=', '.join(missing_topology))) try: positions = ag_or_ts.positions except: positions = [None]*ag_or_ts.n_atoms try: velocities = ag_or_ts.velocities except: velocities = [None]*ag_or_ts.n_atoms atom_kwargs = [] for atom, name, resname, xyz, vel in zip(ag_or_ts, names, resnames, positions, velocities): akwargs = {'name': name} chain_seg = {'segid': atom.segid} for attrname in ('mass', 'charge', 'type', 'altLoc', 'tempfactor', 'occupancy', 'gbscreen', 'solventradius', 'nbindex', 'rmin', 'epsilon', 'rmin14', 'epsilon14', 'id'): try: akwargs[MDA2PMD.get(attrname, attrname)] = getattr(atom, attrname) except AttributeError: pass try: el = atom.element.lower().capitalize() akwargs['atomic_number'] = SYMB2Z[el] except (KeyError, AttributeError): try: tp = atom.type.lower().capitalize() akwargs['atomic_number'] = SYMB2Z[tp] except (KeyError, AttributeError): pass try: chain_seg['chain'] = atom.chainID except AttributeError: pass try: chain_seg['inscode'] = atom.icode except AttributeError: pass atom_kwargs.append((akwargs, resname, atom.resid, chain_seg, xyz, vel)) struct = pmd.Structure() for akwarg, resname, resid, kw, xyz, vel in atom_kwargs: atom = pmd.Atom(**akwarg) if xyz is not None: atom.xx, atom.xy, atom.xz = xyz if vel is not None: atom.vx, atom.vy, atom.vz = vel atom.atom_type = pmd.AtomType(akwarg['name'], None, akwarg['mass'], atomic_number=akwargs.get('atomic_number')) struct.add_atom(atom, resname, resid, **kw) try: struct.box = ag_or_ts.dimensions except AttributeError: struct.box = None if hasattr(ag_or_ts, 'universe'): get_atom_indices = functools.partial(get_indices_from_subset, atomgroup=list(ag_or_ts), universe=ag_or_ts.universe) else: get_atom_indices = lambda x: x # bonds try: params = ag_or_ts.bonds.atomgroup_intersection(ag_or_ts, strict=True) except AttributeError: pass else: for p in params: atoms = [struct.atoms[i] for i in map(get_atom_indices, p.indices)] try: for obj in p.type: bond = pmd.Bond(*atoms, type=obj.type, order=obj.order) struct.bonds.append(bond) if isinstance(obj.type, pmd.BondType): struct.bond_types.append(bond.type) bond.type.list = struct.bond_types except (TypeError, AttributeError): order = p.order if p.order is not None else 1 btype = getattr(p.type, 'type', None) bond = pmd.Bond(*atoms, type=btype, order=order) struct.bonds.append(bond) if isinstance(bond.type, pmd.BondType): struct.bond_types.append(bond.type) bond.type.list = struct.bond_types # dihedrals try: params = ag_or_ts.dihedrals.atomgroup_intersection(ag_or_ts, strict=True) except AttributeError: pass else: for p in params: atoms = [struct.atoms[i] for i in map(get_atom_indices, p.indices)] try: for obj in p.type: imp = getattr(obj, 'improper', False) ign = getattr(obj, 'ignore_end', False) dih = pmd.Dihedral(*atoms, type=obj.type, ignore_end=ign, improper=imp) struct.dihedrals.append(dih) if isinstance(dih.type, pmd.DihedralType): struct.dihedral_types.append(dih.type) dih.type.list = struct.dihedral_types except (TypeError, AttributeError): btype = getattr(p.type, 'type', None) imp = getattr(p.type, 'improper', False) ign = getattr(p.type, 'ignore_end', False) dih = pmd.Dihedral(*atoms, type=btype, improper=imp, ignore_end=ign) struct.dihedrals.append(dih) if isinstance(dih.type, pmd.DihedralType): struct.dihedral_types.append(dih.type) dih.type.list = struct.dihedral_types for param, pmdtype, trackedlist, typelist, clstype in ( ('ureybradleys', pmd.UreyBradley, struct.urey_bradleys, struct.urey_bradley_types, pmd.BondType), ('angles', pmd.Angle, struct.angles, struct.angle_types, pmd.AngleType), ('impropers', pmd.Improper, struct.impropers, struct.improper_types, pmd.ImproperType), ('cmaps', pmd.Cmap, struct.cmaps, struct.cmap_types, pmd.CmapType) ): try: params = getattr(ag_or_ts, param) values = params.atomgroup_intersection(ag_or_ts, strict=True) except AttributeError: pass else: for v in values: atoms = [struct.atoms[i] for i in map(get_atom_indices, v.indices)] try: for parmed_obj in v.type: p = pmdtype(*atoms, type=parmed_obj.type) trackedlist.append(p) if isinstance(p.type, clstype): typelist.append(p.type) p.type.list = typelist except (TypeError, AttributeError): vtype = getattr(v.type, 'type', None) p = pmdtype(*atoms, type=vtype) trackedlist.append(p) if isinstance(p.type, clstype): typelist.append(p.type) p.type.list = typelist return struct