def identify_lipid_leaflets(pts,
vec,
monolayer_cutoff=2.0,
monolayer_cutoff_retry=True,
max_count_asymmetry=0.05,
pbc_rewrap=True,
topologize_tolerance=None):
"""
Identify leaflets in a bilayer by consensus.
Note that the time limit on the topologize call was increased from 10 to 30 for large systems.
"""
#---time limit on the tolerance checker
try:
with time_limit(30):
wrapper = topologize(
pts, vec,
**({
'tol': topologize_tolerance
} if topologize_tolerance else {}))
except TimeoutException, msg:
status(
'topologize failed to join the bilayer. '
'if it is broken over PBCs e.g. a saddle, this is a serious error which may go undetected. '
'make sure you always inspect the topology later.',
tag='error')
wrapper = np.zeros((len(pts), 3))
def identify_leaflets_cluster(self,pts,vec,topologize_time_limit=30,max_count_asymmetry=0.05):
"""
Use scikit-learn clustering methods to separate leaflets.
Note that this method can cluster a tortuous manifold and may work for complex morphologies.
"""
import scipy
import sklearn
import sklearn.neighbors
import sklearn.cluster
nlipids = len(pts)
#---time limit on the topologize function which joins broken bilayers e.g. a saddle that crosses PBCs
try:
with time_limit(topologize_time_limit):
wrapper = topologize(pts,vec,
**({'tol':self.topologize_tolerance} if self.topologize_tolerance else {}))
except TimeoutException, msg:
status('topologize failed to join the bilayer. '
'if it is broken over PBCs e.g. a saddle, this is a serious error which may go undetected. '
'make sure you always inspect the topology later.',tag='error')
wrapper = np.zeros((len(pts),3))
def identify_lipid_leaflets_legacy(pts,vec,monolayer_cutoff,
monolayer_cutoff_retry=True,max_count_asymmetry=0.05,pbc_rewrap=True,
topologize_tolerance=None,topologize_time_limit=30):
"""
Identify leaflets in a bilayer by consensus.
Note that the time limit on the topologize call was increased from 10 to 30 for large systems.
This is the legacy version of this algorithm. Previously it was recursive, lowering the cutoff by small
increments and then calling itself again if the bilayer did not appear to be split correctly. The current
version is called by the LeafletFinder class and throws exceptions to trigger a lower cutoff. We have
tried to preserve the legacy version for other users, but the cluster version is more reliable.
"""
#---previous default was somewhat high, but typically came in from specs, and we reduced it incrementally
if monolayer_cutoff==None: monolayer_cutoff = 2.0
#---time limit on the tolerance checker
try:
with time_limit(topologize_time_limit):
wrapper = topologize(pts,vec,
**({'tol':topologize_tolerance} if topologize_tolerance else {}))
except TimeoutException, msg:
status('topologize failed to join the bilayer. '
'if it is broken over PBCs e.g. a saddle, this is a serious error which may go undetected. '
'make sure you always inspect the topology later.',tag='error')
wrapper = np.zeros((len(pts),3))
def identify_leaflets_cluster(self,
pts,
vec,
topologize_time_limit=30,
max_count_asymmetry=0.05):
"""
Use scikit-learn clustering methods to separate leaflets.
Note that this method can cluster a tortuous manifold and may work for complex morphologies.
"""
import scipy
import sklearn
import sklearn.neighbors
import sklearn.cluster
nlipids = len(pts)
#---time limit on the topologize function which joins broken bilayers e.g. a saddle that crosses PBCs
try:
with time_limit(topologize_time_limit):
wrapper = topologize(
pts, vec,
**({
'tol': self.topologize_tolerance
} if self.topologize_tolerance else {}))
except TimeoutException:
status(
'topologize failed to join the bilayer. '
'if it is broken over PBCs e.g. a saddle, this is a serious error which may go undetected. '
'make sure you always inspect the topology later.',
tag='error')
wrapper = np.zeros((len(pts), 3))
findframe = pts + wrapper * np.array(vec)
#---ensure that all points are in the box
findframe += vec * (findframe < 0) - vec * (findframe > vec)
#---previous calculation of connectivity was done manually
if False:
#---conservative cutoff gets lots of nearby points
cutoff = 10.0
cutoff_short = 2.0
#---make a K-D tree from the points
tree = scipy.spatial.ckdtree.cKDTree(findframe,
boxsize=np.concatenate(
(vec, vec)) + 0. * eps)
#---find the nearest reference points for each instantaneous point
close, nns = tree.query(findframe,
distance_upper_bound=cutoff,
k=20)
#---construct the neighbor list
subjects = np.where(np.all((close < cutoff, close > 0), axis=0))
#---get the pairs of neighbors
subjects, neighbors = subjects[0], nns[subjects]
pds = np.ones((nlipids, nlipids)) * 0.0
pds[tuple((np.arange(nlipids), np.arange(nlipids)))] = 0.0
nears = np.where(np.all((close > 0, close <= cutoff_short),
axis=0))
pds[tuple((nears[0], nns[nears]))] = 1.0 #close[nears]
pds[tuple((nns[nears], nears[0]))] = 1.0 #close[nears]
connectivity = sklearn.neighbors.kneighbors_graph(
findframe, n_neighbors=self.cluster_neighbors, include_self=False)
ward = sklearn.cluster.AgglomerativeClustering(
n_clusters=2, connectivity=connectivity,
linkage='complete').fit(findframe)
imono = ward.labels_
if np.mean(imono) == 0.5:
status('[STATUS] perfect split is %0.5f' % np.mean(imono))
elif (np.all(np.array(imono) == 0) or np.all(np.array(imono) == 1)
or np.abs(np.mean(imono) - 0.5) >= max_count_asymmetry):
status('[STATUS] split is %0.5f' % np.mean(imono))
status('[STATUS] one side has %d' % np.sum(imono))
status('[WARNING] leaflets were not distinguished')
raise Exception(
'[ERROR] failed to identify leaflets. '
'DEVELOPMENT NOTE!? use legacy or a different cutoff?')
else:
status('[STATUS] some lipids might be flipped %d %.5f' %
(np.sum(imono), np.mean(imono)))
return np.array(imono)
def identify_lipid_leaflets_legacy(pts,
vec,
monolayer_cutoff,
monolayer_cutoff_retry=True,
max_count_asymmetry=0.05,
pbc_rewrap=True,
topologize_tolerance=None,
topologize_time_limit=30):
"""
Identify leaflets in a bilayer by consensus.
Note that the time limit on the topologize call was increased from 10 to 30 for large systems.
This is the legacy version of this algorithm. Previously it was recursive, lowering the cutoff by small
increments and then calling itself again if the bilayer did not appear to be split correctly. The current
version is called by the LeafletFinder class and throws exceptions to trigger a lower cutoff. We have
tried to preserve the legacy version for other users, but the cluster version is more reliable.
"""
#---previous default was somewhat high, but typically came in from specs, and we reduced it incrementally
if monolayer_cutoff == None: monolayer_cutoff = 2.0
#---time limit on the tolerance checker
try:
with time_limit(topologize_time_limit):
wrapper = topologize(
pts, vec,
**({
'tol': topologize_tolerance
} if topologize_tolerance else {}))
except TimeoutException:
status(
'topologize failed to join the bilayer. '
'if it is broken over PBCs e.g. a saddle, this is a serious error which may go undetected. '
'make sure you always inspect the topology later.',
tag='error')
wrapper = np.zeros((len(pts), 3))
findframe = pts + wrapper * np.array(vec)
status('this step is somewhat slow. it uses scipy.spatial.pdist.',
tag='warning')
pd = [
scipy.spatial.distance.squareform(
scipy.spatial.distance.pdist(findframe[:, d:d + 1]))
for d in range(3)
]
if pbc_rewrap:
pd3pbc = np.sqrt(
np.sum(np.array([
pd[d] - (pd[d] > vec[d] / 2.) * vec[d] +
(pd[d] < -1 * vec[d] / 2.) * vec[d] for d in range(3)
])**2,
axis=0))
else:
pd3pbc = pd
nbors = np.transpose(np.where(pd3pbc < monolayer_cutoff))
nlipids = len(pts)
imono = np.zeros(nlipids)
nlist = []
for i in range(nlipids):
status('cataloging lipids', i=i, looplen=nlipids, tag='compute')
nlist.append(nbors[np.where(nbors[:, 0] == i)[0], 1])
iref = 0
mono = np.zeros(nlipids)
searched = np.zeros(nlipids)
imono[iref], searched[iref] = 1, 1
imono[nlist[iref]] = 1
while np.any(np.all((imono == 1, searched == 0), axis=0)):
for iref in np.where(np.all((imono == 1, searched == 0), axis=0))[0]:
imono[nlist[iref]] = 1
searched[iref] = 1
#---check that the leaflets were properly distinguished by looking at the number in each monolayer
if np.mean(imono) == 0.5:
status('[STATUS] perfect split is %0.5f' % np.mean(imono))
return imono
elif (monolayer_cutoff_retry
and (np.all(np.array(imono) == 0) or np.all(np.array(imono) == 1)
or np.abs(np.mean(imono) - 0.5) >= max_count_asymmetry)):
status('[STATUS] split is %0.5f' % np.mean(imono))
status('[STATUS] one side has %d' % np.sum(imono))
status('[WARNING] leaflets were not distinguished')
status('[COMPUTE] leaflets = ' + str(np.sum(imono)) + '/' +
str(len(imono)))
status('[WARNING] previous monolayer_cutoff = ' +
str(monolayer_cutoff))
raise Exception(
'[ERROR] failed to identify leaflets so we are returning an exception to the LeafletFinder'
)
else:
status('[STATUS] some lipids might be flipped %d %.5f' %
(np.sum(imono), np.mean(imono)))
return imono