def test_projection_for_pca(self):
traj = pt.load("./data/tz2.nc", "./data/tz2.parm7")
state = pt.load_cpptraj_state(command)
state.run()
cpp_modes = state.data['MyEvecs']
cpp_arr_crd = np.array(cpp_modes._get_avg_crd())
cpp_arr_crd = cpp_arr_crd.reshape(117, 3)
mask = '!@H='
pt.superpose(traj, mask=mask)
avg = pt.mean_structure(traj)
atom_indices = traj.top(mask).indices
strip_avg_coords = avg.xyz[atom_indices]
pt.superpose(traj, mask=mask, ref=avg)
avg2 = pt.mean_structure(traj, mask=mask)
mat = pt.matrix.covar(traj, mask)
modes = pt.matrix.diagonalize(mat, n_vecs=2, dtype='dataset')[0]
aa_eq(cpp_arr_crd, avg2.xyz)
aa_eq(np.abs(modes.eigenvalues), np.abs(state.data['MyEvecs'].eigenvalues))
aa_eq(np.abs(modes.eigenvectors), np.abs(state.data['MyEvecs'].eigenvectors))
projection_data = pt.all_actions.projection(traj, mask=mask, average_coords=avg2.xyz,
eigenvalues=modes.eigenvalues,
eigenvectors=modes.eigenvectors,
scalar_type='covar')
aa_eq(np.abs(projection_data), np.abs(state.data[-2:].values), decimal=3)
def test_mass(self):
traj = self.traj
# cpptraj output
cm = '''
reference avg.pdb
rms R0 reference @CA,C,N,O savematrices mass
'''
state = pt.load_batch(traj, cm)
state.run()
saved_mat = state.data[-1].values
# pytraj output
avg = pt.mean_structure(traj)
mat = pt.calc_rotation_matrix(traj,
ref=avg,
mask='@CA,C,N,O',
mass=True)
assert mat.shape == (traj.n_frames, 3, 3), 'mat shape'
aa_eq(mat.flatten(), saved_mat.flatten())
# with rmsd
avg = pt.mean_structure(traj)
mat2, rmsd_ = pt.calc_rotation_matrix(traj,
ref=avg,
mask='@CA,C,N,O',
mass=True,
with_rmsd=True)
aa_eq(mat2.flatten(), saved_mat.flatten())
assert pt.tools.rmsd(rmsd_, state.data['R0']) < 1E-3
def test_autoimage(self):
traj = pt.iterload(fn('tz2.ortho.nc'), fn('tz2.ortho.parm7'))
t0 = traj[:]
t0.autoimage()
avg_0 = pt.mean_structure(t0, '@CA')
avg_1 = pt.mean_structure(traj(autoimage=True), '@CA')
aa_eq(avg_0.xyz, avg_1.xyz)
def test_autoimage(self):
traj = pt.iterload("data/tz2.ortho.nc", "data/tz2.ortho.parm7")
t0 = traj[:]
t0.autoimage()
avg_0 = pt.mean_structure(t0, '@CA')
avg_1 = pt.mean_structure(traj(autoimage=True), '@CA')
aa_eq(avg_0.xyz, avg_1.xyz)
def test_comprehensive(self):
traj = pt.iterload(fn('Tc5b.x'), fn('Tc5b.top'))
# make sure we DO reproducing cpptraj output
f_saved = pt.iterload(fn("avg.Tc5b.pdb"), traj.top)[0]
# shorter
frame2 = mean_structure(traj)
aa_eq(frame2.xyz, f_saved.xyz, decimal=3)
frame3 = mean_structure(traj=traj)
aa_eq(frame3.xyz, f_saved.xyz, decimal=3)
# test list
frame4 = mean_structure(traj=[traj, traj[:3]], top=traj.top)
# test iter
frame5 = mean_structure(traj=traj(1, 8, 2), top=traj.top)
f5_saved = pt.iterload(fn("avg.Tc5b.frame_2_to_8_skip_2.pdb"),
traj.top)[0]
aa_eq(frame5.xyz, f5_saved.xyz, decimal=3)
# test iter CA
frame5 = mean_structure(traj[[0, 3, 7]], '@CA', top=traj.top)
# use atom_indices
ca_indices = pt.select('@CA', traj.top)
frame5_1 = mean_structure(traj[[0, 3, 7]], ca_indices, top=traj.top)
# test frame_indices
frame6 = mean_structure(traj, mask='@CA', frame_indices=[0, 3, 7])
aa_eq(frame5.xyz, frame6.xyz, decimal=3)
aa_eq(frame5_1.xyz, frame6.xyz, decimal=3)
xyz_0 = pt.get_coordinates(traj(1, 8, 2))
xyz_1 = np.array([
frame.xyz.copy()
for frame in traj.iterframe(frame_indices=range(1, 8, 2))
])
aa_eq(xyz_0, xyz_1, decimal=3)
# test as traj
out_traj = mean_structure(traj,
mask='@CA',
frame_indices=[0, 3, 7],
dtype='traj')
assert isinstance(out_traj, Trajectory), 'must be Trajectory'
aa_eq(out_traj.xyz, [frame6.xyz], decimal=3)
# raise if not trajectory, traj or frame
with pytest.raises(ValueError):
pt.mean_structure(traj, dtype='trajxyz')
def prepare(self):
'''Load trajectory, print information and
perform initial alignment and fit.'''
self.load_traj()
self.atoms = self.traj.n_atoms
self.frames = self.traj.n_frames
print('\nPROTEIN INFORMATION')
print('PDB code: {}'.format(self.PDB))
print('residues: {}'.format(self.atoms))
print('frames: {}'.format(self.frames))
print('temperature: {}'.format(self.temperature))
# Align molecule to principal axes
pt.principal_axes(self.traj, dorotation=True)
# Center molecule at the origin
pt.center(self.traj, center='origin')
# First rmsd fit
pt.rmsd(self.traj)
# Calculate average structure
self.avg = pt.mean_structure(self.traj)
# Perform rms fit to average structure
pt.rmsd(self.traj, ref=self.avg)
# Initialize arrays
self.initialize_arrays()
def test_nomass(self):
traj = self.traj
# cpptraj output
cm = '''
reference avg.pdb
rms R0 reference @CA,C,N,O savematrices
'''
state = pt.load_batch(traj, cm)
state.run()
saved_mat = state.data[-1].values
# pytraj output
avg = pt.mean_structure(traj)
mat = pt.calc_rotation_matrix(traj, ref=avg, mask='@CA,C,N,O')
assert mat.shape == (traj.n_frames, 3, 3), 'mat shape'
aa_eq(mat.flatten(), saved_mat.flatten())
# not specify reference
cm = '''
rms R0 @CA,C,N,O savematrices
'''
state = pt.load_batch(traj, cm)
state.run()
saved_mat = state.data[-1].values
mat = pt.calc_rotation_matrix(traj, mask='@CA,C,N,O')
aa_eq(mat.flatten(), saved_mat.flatten())
def test_pmap_average_structure(self):
traj = pt.iterload("data/tz2.nc", "data/tz2.parm7")
saved_frame = pt.mean_structure(traj, '@CA')
saved_xyz = saved_frame.xyz
for n_cores in [2, 3, 4]:
frame = pt.pmap(pt.mean_structure, traj, '@CA', n_cores=n_cores)
aa_eq(frame.xyz, saved_xyz)
def test_pmap_average_structure(self):
traj = pt.iterload("data/tz2.nc", "data/tz2.parm7")
saved_frame = pt.mean_structure(traj, '@CA')
saved_xyz = saved_frame.xyz
for n_cores in [2, 3, 4]:
frame = pt.pmap(pt.mean_structure, traj, '@CA', n_cores=n_cores)
aa_eq(frame.xyz, saved_xyz)
def test_dataset_coords_ref(self):
traj = pt.iterload('data/tz2.nc', 'data/tz2.parm7')
avg_frame = pt.mean_structure(traj(rmsfit=(0, '!@H=')))
state = self.state
# need to loop several times to make sure this does not fail
# due to memory free
for _ in range(20):
cpp_ref = state.data['AVG'].get_frame()
aa_eq(avg_frame.xyz, cpp_ref.xyz)
def test_dataset_coords_ref(self):
traj = pt.iterload('data/tz2.nc', 'data/tz2.parm7')
avg_frame = pt.mean_structure(traj(rmsfit=(0, '!@H=')))
state = self.state
# need to loop several times to make sure this does not fail
# due to memory free
for _ in range(20):
cpp_ref = state.data['AVG'].get_frame()
aa_eq(avg_frame.xyz, cpp_ref.xyz)
def test_comprehensive(self):
traj = pt.iterload("./data/Tc5b.x", "./data/Tc5b.top")
# make sure we DO reproducing cpptraj output
f_saved = pt.iterload("./data/avg.Tc5b.pdb", traj.top)[0]
# shorter
frame2 = mean_structure(traj)
aa_eq(frame2.xyz, f_saved.xyz, decimal=3)
frame3 = mean_structure(traj=traj)
aa_eq(frame3.xyz, f_saved.xyz, decimal=3)
# test list
frame4 = mean_structure(traj=[traj, traj[:3]], top=traj.top)
# test iter
frame5 = mean_structure(traj=traj(1, 8, 2), top=traj.top)
f5_saved = pt.iterload("./data/avg.Tc5b.frame_2_to_8_skip_2.pdb",
traj.top)[0]
aa_eq(frame5.xyz, f5_saved.xyz, decimal=3)
# test iter CA
frame5 = mean_structure(traj[[0, 3, 7]], '@CA', top=traj.top)
# use atom_indices
ca_indices = pt.select('@CA', traj.top)
frame5_1 = mean_structure(traj[[0, 3, 7]], ca_indices, top=traj.top)
# test frame_indices
frame6 = mean_structure(traj, mask='@CA', frame_indices=[0, 3, 7])
aa_eq(frame5.xyz, frame6.xyz, decimal=3)
aa_eq(frame5_1.xyz, frame6.xyz, decimal=3)
xyz_0 = pt.get_coordinates(traj(1, 8, 2))
xyz_1 = np.array([frame.xyz.copy(
) for frame in traj.iterframe(frame_indices=range(1, 8, 2))])
aa_eq(xyz_0, xyz_1, decimal=3)
# test as traj
out_traj = mean_structure(traj,
mask='@CA',
frame_indices=[0, 3, 7],
dtype='traj')
assert isinstance(out_traj, Trajectory), 'must be Trajectory'
aa_eq(out_traj.xyz, frame6.xyz, decimal=3)
# raise if not trajectory, traj or frame
self.assertRaises(ValueError, lambda: pt.mean_structure(traj, dtype='trajxyz'))
def test_autoimage_with_rmsfit(self):
traj = pt.iterload(fn('tz2.ortho.nc'), fn('tz2.ortho.parm7'))
t0 = traj[:]
pt.autoimage(t0).superpose()
avg_0 = pt.mean_structure(t0, '@CA')
avg_1 = pt.mean_structure(traj(autoimage=True, rmsfit=0), '@CA')
aa_eq(avg_0.xyz, avg_1.xyz)
# 3rd frame
# assign traj again
t0 = traj[:]
pt.autoimage(t0).superpose(ref=3)
avg_0 = pt.mean_structure(t0, '@CA')
avg_1 = pt.mean_structure(traj(autoimage=True, rmsfit=3), '@CA')
avg_2 = pt.mean_structure(traj, autoimage=True, rmsfit=3, mask='@CA')
aa_eq(avg_0.xyz, avg_1.xyz)
aa_eq(avg_0.xyz, avg_2.xyz)
# 3rd frame, frame_indices
# assign traj again
frame_indices = [0, 8, 5]
t0 = traj[frame_indices]
t1 = traj[frame_indices]
t0.autoimage().superpose(ref=-1)
avg_0 = pt.mean_structure(t0, '@CA')
# use ref=5 which correspond to original index
# try with pytraj.TrajectoryIterator
avg_1 = pt.mean_structure(
traj,
autoimage=True,
rmsfit=5,
mask='@CA',
frame_indices=frame_indices)
# try with pytraj.Trajectory
avg_2 = pt.mean_structure(t1, autoimage=True, rmsfit=-1, mask='@CA')
avg_3 = pt.mean_structure(
traj[:],
autoimage=True,
rmsfit=5,
mask='@CA',
frame_indices=frame_indices)
aa_eq(avg_0.xyz, avg_1.xyz)
aa_eq(avg_0.xyz, avg_2.xyz)
aa_eq(avg_0.xyz, avg_3.xyz)
def clean_traj(self):
# Align molecule to principal axes
pt.principal_axes(self.traj, dorotation=True)
# Center molecule at the origin
pt.center(self.traj, center='origin')
# First rmsd fit
pt.rmsd(self.traj)
# Calculate average structure
self.avg = pt.mean_structure(self.traj)
# Perform rms fit to average structure
pt.rmsd(self.traj, ref=self.avg)
def test_autoimage_with_rmsfit(self):
traj = pt.iterload("data/tz2.ortho.nc", "data/tz2.ortho.parm7")
t0 = traj[:]
pt.autoimage(t0).superpose()
avg_0 = pt.mean_structure(t0, '@CA')
avg_1 = pt.mean_structure(traj(autoimage=True, rmsfit=0), '@CA')
aa_eq(avg_0.xyz, avg_1.xyz)
# 3rd frame
# assign traj again
t0 = traj[:]
pt.autoimage(t0).superpose(ref=3)
avg_0 = pt.mean_structure(t0, '@CA')
avg_1 = pt.mean_structure(traj(autoimage=True, rmsfit=3), '@CA')
avg_2 = pt.mean_structure(traj, autoimage=True, rmsfit=3, mask='@CA')
aa_eq(avg_0.xyz, avg_1.xyz)
aa_eq(avg_0.xyz, avg_2.xyz)
# 3rd frame, frame_indices
# assign traj again
frame_indices = [0, 8, 5]
t0 = traj[frame_indices]
t1 = traj[frame_indices]
t0.autoimage().superpose(ref=-1)
avg_0 = pt.mean_structure(t0, '@CA')
# use ref=5 which correspond to original index
# try with pytraj.TrajectoryIterator
avg_1 = pt.mean_structure(traj,
autoimage=True,
rmsfit=5,
mask='@CA',
frame_indices=frame_indices)
# try with pytraj.Trajectory
avg_2 = pt.mean_structure(t1, autoimage=True, rmsfit=-1, mask='@CA')
avg_3 = pt.mean_structure(traj[:],
autoimage=True,
rmsfit=5,
mask='@CA',
frame_indices=frame_indices)
aa_eq(avg_0.xyz, avg_1.xyz)
aa_eq(avg_0.xyz, avg_2.xyz)
aa_eq(avg_0.xyz, avg_3.xyz)
def test_projection_for_pca(self):
traj = pt.load(fn('tz2.nc'), fn('tz2.parm7'))
state = pt.load_cpptraj_state(command)
state.run()
cpp_modes = state.data['MyEvecs']
cpp_arr_crd = np.array(cpp_modes._get_avg_crd())
cpp_arr_crd = cpp_arr_crd.reshape(117, 3)
mask = '!@H='
pt.superpose(traj, mask=mask)
avg = pt.mean_structure(traj)
atom_indices = traj.top(mask).indices
avg.xyz[atom_indices]
pt.superpose(traj, mask=mask, ref=avg)
avg2 = pt.mean_structure(traj, mask=mask)
mat = pt.matrix.covar(traj, mask)
modes = pt.matrix.diagonalize(mat, n_vecs=2, dtype='dataset')[0]
aa_eq(cpp_arr_crd, avg2.xyz)
aa_eq(np.abs(modes.eigenvalues),
np.abs(state.data['MyEvecs'].eigenvalues))
aa_eq(np.abs(modes.eigenvectors),
np.abs(state.data['MyEvecs'].eigenvectors))
projection_data = pt.all_actions.projection(
traj,
mask=mask,
average_coords=avg2.xyz,
eigenvalues=modes.eigenvalues,
eigenvectors=modes.eigenvectors,
scalar_type='covar')
aa_eq(np.abs(projection_data),
np.abs(state.data[-2:].values),
decimal=3)
def test_RMSF(self):
traj = pt.iterload("./data/tz2.nc", "./data/tz2.parm7")
state = pt.load_batch(traj, '''
rms first
average crdset MyAvg
run
rms ref MyAvg
atomicfluct out fluct.agr''')
state.run()
t0 = traj[:]
pt.superpose(t0, ref=0)
avg = pt.mean_structure(t0)
pt.superpose(t0, ref=avg)
data = pt.rmsf(t0)
aa_eq(data, state.data[-1].values)
def test_iter_options(self):
traj = pt.iterload("data/tz2.ortho.nc", "data/tz2.ortho.parm7")
t0 = traj[:].autoimage().rmsfit(ref=0)
saved_avg = pt.mean_structure(t0)
saved_radgyr = pt.radgyr(traj, '@CA')
# perform autoimage, then rms fit to 1st frame, then compute mean structure
iter_options = {'autoimage': True, 'rmsfit': 0}
for n_cores in [2, 3]:
avg = pt.pmap(pt.mean_structure,
traj,
iter_options=iter_options,
n_cores=n_cores)
aa_eq(saved_avg.xyz, avg.xyz)
radgyr_ = pt.tools.dict_to_ndarray(
pt.pmap(pt.radgyr, traj, iter_options={'mask': '@CA'}))
aa_eq(radgyr_[0], saved_radgyr)
def test_RMSF(self):
traj = pt.iterload(fn('tz2.nc'), fn('tz2.parm7'))
state = pt.load_batch(traj, '''
rms first
average crdset MyAvg
run
rms ref MyAvg
atomicfluct out fluct.agr''')
state.run()
t0 = traj[:]
pt.superpose(t0, ref=0)
avg = pt.mean_structure(t0)
pt.superpose(t0, ref=avg)
data = pt.rmsf(t0)
aa_eq(data, state.data[-1].values)
def test_iter_options(self):
traj = pt.iterload("data/tz2.ortho.nc", "data/tz2.ortho.parm7")
t0 = traj[:].autoimage().rmsfit(ref=0)
saved_avg = pt.mean_structure(t0)
saved_radgyr = pt.radgyr(traj, '@CA')
# perform autoimage, then rms fit to 1st frame, then compute mean structure
iter_options = {'autoimage': True, 'rmsfit': 0}
for n_cores in [2, 3]:
avg = pt.pmap(pt.mean_structure,
traj,
iter_options=iter_options,
n_cores=n_cores)
aa_eq(saved_avg.xyz, avg.xyz)
radgyr_ = pt.tools.dict_to_ndarray(pt.pmap(pt.radgyr,
traj,
iter_options={'mask':
'@CA'}))
aa_eq(radgyr_[0], saved_radgyr)
import pytraj as pt
pdb = pt.load_pdb_rcsb("1l2y")
avg_pdb = pt.mean_structure(pdb, '@CA')
print(avg_pdb)
def calc_(traj=traj):
frame = pt.mean_structure(traj)
print(frame)
from sklearn.decomposition import PCA
import pytraj as pt
# we use `load` method to load all data to memory. This is good for small data size.
# use `pytraj.iterload` for out-of-core traj.
traj = pt.load('../tests/data/tz2.nc', '../tests/data/tz2.parm7')
pca = PCA(n_components=2)
# superpose to 1st frame
pt.superpose(traj, ref=0, mask='!@H=')
# create average structure
avg = pt.mean_structure(traj)
# superpose all structures to average frame
pt.superpose(traj, ref=avg, mask='!@H=')
# perform PCA calculation and get transformed coords
# we need to reshape 3D traj.xyz array to 2D to make sklearn happy
# make a new traj by stripping all H atoms
traj_new = traj['!@H=']
xyz_2d = traj_new.xyz.reshape(traj_new.n_frames, traj_new.n_atoms * 3)
print(xyz_2d.shape) # (n_frames, n_dimensions)
reduced_cartesian = pca.fit_transform(xyz_2d)
print(reduced_cartesian.shape) # (n_frames, n_dimensions)
plt.figure()
def calc_(traj=traj):
frame = pt.mean_structure(traj)
print(frame)
from sklearn.decomposition import PCA
import pytraj as pt
# we use `load` method to load all data to memory. This is good for small data size.
# use `pytraj.iterload` for out-of-core traj.
traj = pt.load('../tests/data/tz2.nc', '../tests/data/tz2.parm7')
pca = PCA(n_components=2)
# superpose to 1st frame
pt.superpose(traj, ref=0, mask='!@H=')
# create average structure
avg = pt.mean_structure(traj)
# superpose all structures to average frame
pt.superpose(traj, ref=avg, mask='!@H=')
# perform PCA calculation and get transformed coords
# we need to reshape 3D traj.xyz array to 2D to make sklearn happy
# make a new traj by stripping all H atoms
traj_new = traj['!@H=']
xyz_2d = traj_new.xyz.reshape(traj_new.n_frames, traj_new.n_atoms * 3)
print(xyz_2d.shape) # (n_frames, n_dimensions)
reduced_cartesian = pca.fit_transform(xyz_2d)
print(reduced_cartesian.shape) # (n_frames, n_dimensions)
plt.figure()
