def _get_crossminer_pharmacophore(self):
"""
convert a PharmacophoreModel into a crossminer pharmacophore
"""
# TODO: UPDATE WITH CHARGED FEATURES
supported_features = {"acceptor_projected": "acceptor",
"donor_projected": "donor",
"ring": "apolar"}
try:
Pharmacophore.read_feature_definitions()
except:
raise ImportError("Crossminer is only available to CSD-Discovery")
feature_definitions = {supported_features[fd.identifier]: fd for fd in Pharmacophore.feature_definitions.values()
if fd.identifier in supported_features.keys()}
model_features = []
for feat in self._features:
if feat.feature_type == "negative" or feat.feature_type == "positive":
print("Charged feature not currently supported in CrossMiner: Its on the TODO list")
else:
sphere = GeometricDescriptors.Sphere(feat.feature_coordinates, self.settings.radius)
if feat.projected_coordinates:
projected = GeometricDescriptors.Sphere(feat.projected_coordinates, self.settings.radius)
p = Pharmacophore.Feature(feature_definitions[feat.feature_type], *[sphere, projected])
else:
p = Pharmacophore.Feature(feature_definitions[feat.feature_type], sphere)
model_features.append(p)
if self.settings.excluded_volume:
if not self.protein:
print("Pharmacophore Model must have protein to calculate excluded volume")
else:
bs = self._get_binding_site_residues()
for residue in bs.residues:
mol = None
mol = Molecule(identifier="temp_residue")
# for a in residue.backbone_atoms:
# ev = Pharmacophore.ExcludedVolume(GeometricDescriptors.Sphere(a.coordinates, 2))
# model_features.append(ev)
for a in residue.backbone_atoms:
mol.add_atom(a)
centre = mol.centre_of_geometry()
ev = Pharmacophore.ExcludedVolume(GeometricDescriptors.Sphere(centre, 2))
model_features.append(ev)
return Pharmacophore.Query(model_features)
def detect_features(self, crystal):
_csv = ChemistryLib.CrystalStructureView_instantiate(crystal._crystal)
_ssc = MotifPharmacophoreLib.MotifPharmacophoreSearchStructureCreator()
_ssc.register_components_from_feature_definitions(
(self._feature_def, ))
_mss = _ssc.search_structure(_csv)
_ded = self._feature_def.feature_deducer()
_feats = _ded.generate_motif_features(
_mss, self._feature_def.component_label())
features = []
for i in range(_feats.end_index()):
code, feat_pts = _feats.at(i)
for fp in feat_pts:
pts = fp.points()
spheres = tuple(
GeometricDescriptors.Sphere((p[0], p[1], p[2]), 1.0)
for p in pts)
# Skip duplicates
dup = False
if len(spheres) == 2:
for f in features:
if len(f.spheres) == 2:
if (f.spheres[0] == spheres[1]
and f.spheres[1] == spheres[0]):
dup = True
break
if not dup:
feat = Pharmacophore.Feature(self._clone(),
*spheres,
crystal=crystal)
feat.label = f"{crystal.identifier}/{self.identifier}/{i}"
features.append(feat)
return tuple(features)
def create_new_features(self):
"""
create new features from points and projections over the cutoff value
:return: list of (`hotspots.pharmacophore_extension.Feature`)
"""
new_feats = []
point = GeometricDescriptors.Sphere(centre=self.point, radius=1)
if len(self.projection_peaks) == 0:
feat = Pharmacophore.Feature(self.feature_def, point)
feat.point = point
feat.score = self.value
new_feats.append(feat)
else:
for projection in self.projection_peaks:
proj = GeometricDescriptors.Sphere(centre=projection.point,
radius=1)
feat = Pharmacophore.Feature(self.feature_def, point, proj)
feat.point = point
feat.score = self.value
feat.projected = proj
feat.projected_value = projection.value
new_feats.append(feat)
return new_feats
def detect_from_arpeggio(self, protein_path, hetid, chain):
"""
creates a pharmacophore from protein-ligand interactions
TODO: This could be cleaner but for time reasons this is good enough. For example the SMARTS
definitions between Arpeggio and Crossminer are not identical. Also, the SMARTS "grouping"
are subtly different.
1. Create atomtypes using Crossminer
2. Detect bonds using Arepeggio
3. For features in ligand, if bond, create a feature
:param protein:
:param hetid:
:param chainid:
:return:
"""
# Arpeggio needs the protein in file, read the protein to get some information
protein = Protein.from_file(protein_path)
pdb_code = protein.identifier
# assertion 1: protein must be protonated for CrossMiner
# assert("H" in {atom.atomic_symbol for atom in protein.atoms[:50]})
# assertion 2: protein must contain the ligand of interest
assert (len([
l for l in protein.ligands if l.identifier.split(":")[0] == chain
and l.identifier.split(":")[1][:3] == hetid
]) == 1)
lig = [
l for l in protein.ligands if l.identifier.split(":")[0] == chain
and l.identifier.split(":")[1][:3] == hetid
][0]
# CrossMiner needs a `ccdc.crystal.Crystal`
crystal_ligand = self._get_crystal(lig)
# Run Arpeggio
arpeggio = Arpeggio(pdb_code, hetid, chain, protein_path)
arpeggio.run()
atom_features, ring_features = arpeggio.create_feature_list()
interaction_features = []
interaction_features.extend(atom_features)
interaction_features.extend(ring_features)
# CrossMiner atom-typing
new_features = []
ipoints = np.array([
to_array(interaction.point) for interaction in interaction_features
])
for identifier, fd in self.feature_definitions.items():
feats = fd.detect_features(crystal_ligand)
fpoints = np.array(
[to_array(feat.point[0].centre) for feat in feats])
# find the overlapping points
ipoint_index, fpoint_index = np.where(
distance.cdist(ipoints, fpoints) < 0.01)
for i, f in zip(ipoint_index, fpoint_index):
if identifier is "ring":
fd = Pharmacophore.feature_definitions[
"ring_planar_projected"]
# sphere from CrossMiner
point = GeometricDescriptors.Sphere(fpoints[f], 1)
# sphere from Arpeggio
projected = GeometricDescriptors.Sphere(
to_array(interaction_features[i].projected), 1)
print(point, projected)
new = Pharmacophore.Feature(fd, point, projected)
new.point = point
new.point_identifier = interaction_features[i].point_identifier
new.projected = projected
new.projected_identifier = interaction_features[
i].projected_identifier
print(new.projected_identifier)
new_features.append(new)
print(len(new_features))
self.detected_features = new_features
self.protein = self._get_crystal(protein)
self.ligands = [crystal_ligand]
def from_hotspot(self,
hr,
projections=True,
min_distance=2,
radius_dict={"apolar": 2.5},
sigma=1,
override=True):
interaction_dict = {
"donor": ["acceptor_projected"],
"acceptor": ["donor_projected", "donor_ch_projected"]
# "apolar": ["ring_planar_projected"]
}
hotspot_to_cm = {
"projected": {
"apolar": "ring_planar_projected",
"donor": "donor_projected",
"acceptor": "acceptor_projected"
},
"non-projected": {
"apolar": "ring",
"donor": "donor_projected",
"acceptor": "acceptor"
},
}
# get peaks
features = []
for p, g in hr.super_grids.items():
# peak as a sphere
# Keep consistent with vis
if p == "apolar":
h = g.max_value_of_neighbours()
h = h.gaussian(sigma=sigma + 0.5)
#
all_peaks = h.get_peaks(min_distance=min_distance + 2,
cutoff=5)
else:
h = g.max_value_of_neighbours()
h = h.gaussian(sigma=sigma)
#
all_peaks = h.get_peaks(min_distance=min_distance, cutoff=5)
if p in radius_dict:
radius = radius_dict[p]
else:
radius = 1.5
for peak in all_peaks:
point = GeometricDescriptors.Sphere(centre=peak, radius=radius)
score = h.value_at_point(peak)
if p != "apolar" and projections:
# binding site from point (within 4/5 angstrom of peak)
binding_site = hr.protein.copy()
bs = Protein.BindingSiteFromPoint(hr.protein,
peak,
distance=6)
for r in ({r.identifier
for r in binding_site.residues} -
{r.identifier
for r in bs.residues}):
binding_site.remove_residue(r)
# detect projected features
pm = ProteinPharmacophoreModel()
pm.feature_definitions = interaction_dict[p]
pm.detect_from_prot(binding_site)
feats = pm.detected_features
# This returns multiple: ATM the user will then select which one (semi-automated)
# TODO: implement method to pick the best projection
projs = select_projections(feats,
np.array([peak]),
tolerance=4)
else:
projs = None
if projs is None:
# no projections
if p == "donor":
print(
"warning! feature: donor projection used without projection. Will not work in CrossMiner"
)
if override:
print("here")
f = Pharmacophore.Feature(
self.feature_definitions[
hotspot_to_cm["non-projected"][p]], point)
f.point = point
f.score = score
features.append(f)
else:
f = Pharmacophore.Feature(
self.feature_definitions[
hotspot_to_cm["non-projected"][p]], point)
f.point = point
f.score = score
features.append(f)
else:
# for proj in projs:
# just picking the closest now
centre = (projs.spheres[0].centre[0],
projs.spheres[0].centre[1],
projs.spheres[0].centre[2])
s = GeometricDescriptors.Sphere(centre=centre,
radius=radius)
f = Pharmacophore.Feature(
self.feature_definitions[hotspot_to_cm["projected"]
[p]], point, s)
f.point = point
f.projected = s
f.projected_atom = projs.point_atom
f.score = score
features.append(f)
self.detected_features = features
def create_feature(peak):
point = GeometricDescriptors.Sphere(centre=peak.point, radius=1)
feat = Pharmacophore.Feature(peak.feature_def, point)
feat.point = point
feat.score = peak.value
return feat
def create_consensus(pharmacophores, cutoff=2, max_distance=2.0):
"""
"""
new_features = []
# initialise grids from all pharmacophores
feature_point_grids = _create_grids(pharmacophores)
# add point spheres to corresponding grids
features_by_type = {
k:
[f for p in pharmacophores for f in p.features if f.identifier == k]
for k in feature_point_grids.keys()
}
for identifier, all_features in features_by_type.items():
# add spheres to grid
feature_point_grids[identifier] = _features_to_grid(
all_features, feature_point_grids[identifier])
# find peaks
all_peaks = feature_point_grids[identifier].get_peaks(min_distance=2,
cutoff=0)
peak_objs = []
for peak in all_peaks:
value = feature_point_grids[identifier].value_at_point(peak)
if value >= cutoff:
peak_objs.append(
GridPeak(point=peak,
value=value,
feature_def=Pharmacophore.
feature_definitions[identifier]))
peaks_array = np.array([p.point for p in peak_objs])
# projections
if len(peaks_array) > 0:
if len(all_features[0].spheres) > 1:
# if a peak has features with projections try for a concensus projection
# assign features to closest peak
for feature in all_features:
index = closest_peak_index(peaks_array, feature,
max_distance)
if index is not None:
peak_objs[int(index)].features.append(feature)
# create projections
for j, peak in enumerate(peak_objs):
peak.create_projection_grid()
peak.find_projection_peaks()
feats = peak.create_new_features()
for f in feats:
# for a projected feat, if no projection found, scrap
if len(f.spheres) > 1:
new_features.append(f)
else:
for j, peak in enumerate(peak_objs):
point = GeometricDescriptors.Sphere(centre=peak.point,
radius=1)
feat = Pharmacophore.Feature(peak.feature_def, point)
feat.point = point
feat.score = peak.value
new_features.append(feat)
return new_features, feature_point_grids
def run(self):
if not os.path.exists(self.args.output_directory):
os.makedirs(self.args.output_directory)
Pharmacophore.read_feature_definitions()
self.crystals = list(io.CrystalReader(self.args.overlay_file))
if self.args.threshold <= 0.0:
self.args.threshold = (len(self.crystals)) / 2.0
if self.args.feature_definitions:
self.feature_definitions = [
v for k, v in Pharmacophore.feature_definitions.items()
if k in self.args.feature_definitions
]
else:
self.feature_definitions = [
fd for fd in Pharmacophore.feature_definitions.values()
if fd.identifier != 'exit_vector' and
fd.identifier != 'heavy_atom' and fd.identifier != 'hydrophobe'
]
complete_set_of_features = []
for fd in self.feature_definitions:
detected = [fd.detect_features(c) for c in self.crystals]
all_feats = [f for l in detected for f in l]
if not all_feats:
continue
minx = min(f.spheres[0].centre.x() for f in all_feats)
miny = min(f.spheres[0].centre.y() for f in all_feats)
minz = min(f.spheres[0].centre.z() for f in all_feats)
maxx = max(f.spheres[0].centre.x() for f in all_feats)
maxy = max(f.spheres[0].centre.y() for f in all_feats)
maxz = max(f.spheres[0].centre.z() for f in all_feats)
g = utilities.Grid((minx - 1., miny - 1., minz - 1.),
(maxx + 1, maxy + 1, maxz + 1), 0.2)
spheres = []
for f in all_feats:
if f.spheres[0] in spheres:
g.set_sphere(f.spheres[0].centre, f.spheres[0].radius, 0)
else:
spheres.append(f.spheres[0])
g.set_sphere(f.spheres[0].centre, f.spheres[0].radius, 1)
islands = g.islands(self.args.threshold)
print('Feature: %s, max value %.2f, n_features %d' %
(fd.identifier, g.extrema[1], len(islands)))
for island in islands:
# how do I make a feature from an island? Location of highest value
indices = island.indices_at_value(island.extrema[1])
centre = indices[0]
org = island.bounding_box[0]
centre = tuple(org[i] + island.spacing * centre[i]
for i in range(3))
radius = 1.0
# Any other spheres?
if len(all_feats[0].spheres) > 1:
# Pick all features which contain centre
feat_dists = {}
for f in all_feats:
dist, feat = (GeometricDescriptors.point_distance(
f.spheres[0].centre, centre), f)
if feat_dists.has_key(dist):
feat_dists[dist].append(feat)
else:
feat_dists.update({dist: [feat]})
feat_dists = collections.OrderedDict(
sorted(feat_dists.items()))
shortest_distance = feat_dists.keys()[0]
if len(feat_dists[shortest_distance]) > 1:
new_feat = [
Pharmacophore.Feature(
fd,
GeometricDescriptors.Sphere(centre, radius),
feat_dists[shortest_distance][i].spheres[1])
for i in range(len(feat_dists[shortest_distance]))
]
else:
new_feat = [
Pharmacophore.Feature(
fd,
GeometricDescriptors.Sphere(centre, radius),
feat_dists[shortest_distance][0].spheres[1])
]
else:
new_feat = [
Pharmacophore.Feature(
fd, GeometricDescriptors.Sphere(centre, radius))
]
complete_set_of_features.extend(new_feat)
model = Pharmacophore.Query(complete_set_of_features)
model.write(os.path.join(self.args.output_directory, 'model.cm'))