print 'Percent overlap:', ovr
if ovr < 0.0: ovr = 0.0
except:
print 'Exception for lccc and overlap score'
print_exc()
ovr = 0.0
if ovr < 0.02:
print "Maps do not overlap: ", Name2
continue
#SCCC
print 'Local correlation score: ', ccc_mask
if ccc_mask < -1.0 or ccc_mask > 1.0:
ccc_mask = 0.0
#LMI
try:
mi_mask = sc.MI(emmap1, emmap2, c1, c2, 3)
print 'Local Mutual information score: ', mi_mask
if mi_mask < 0.0: mi_mask = 0.0
except:
print 'Exception for MI score'
print_exc()
mi_mask = 0.0
#NMI
try:
nmi = sc.MI(emmap1, emmap2, c1, c2, 1, None, None, True)
print 'Normalized Mutual information score:', nmi
if nmi < 0.0: nmi = 0.0
except:
print 'Exception for NMI score'
print_exc()
nmi = 0.0
def rank_fit_ensemble(self,ensemble_list,score,res_target_map,sigma_coeff,number_top_mod=0,\
write=False,targetMap=False,cont_targetMap=None):
"""
RMSD clustering of the multiple "fits" accordingly with a chosen score.
Cluster the fits based on Calpha RMSD (starting from the best scoring model)
Arguments:
*ensemble_list*
Input list of Structure Instances.
*targetMap*
Target Map Instance.
*score*
Scoring function to use.
See ScoringFunctions class for a list of the available Scoring Function.
E.g. set score='CCC' to use the Cross-correlation coefficient.
Score option are:
i 'CCC' - Cross-correlation coefficient;
ii 'LAP' - Laplacian-filtered cross-correlation coefficient: useful for maps with resolutions worse than 10-15 A;
iii 'MI' - Mutual information score: a good and robust score but relatively slow to calculate;
iv 'ENV' - Envelope score: the fastest score to calculate due to binarisation of the map.
v-vii 'NV','NV_Sobel','NV_Laplace'- Normal vector score: a vector-based surface superimposition score with or without Sobel/Laplace filter.
viii 'CD' - Chamfer Distance: a score used in computer vision algorithms as a fast similarity metric
*rms_cutoff*
float, the Calpha RMSD cutoff based on which you want to cluster the solutions. For example 3.5 (for 3.5 A).
*res_target_map*
the resolution, in Angstroms, of the target Map.
*sigma_coeff*
the sigma value (multiplied by the resolution) that controls the width of the Gaussian.
Default values is 0.356.
Other values used :
0.187R corresponding with the Gaussian width of the Fourier transform falling to half the maximum at 1/resolution, as used in Situs (Wriggers et al, 1999);
0.225R which makes the Fourier transform of the distribution fall to 1/e of its maximum value at wavenumber 1/resolution, the default in Chimera (Petterson et al, 2004)
0.356R corresponding to the Gaussian width at 1/e maximum height equaling the resolution, an option in Chimera (Petterson et al, 2004);
0.425R the fullwidth half maximum being equal to the resolution, as used by FlexEM (Topf et al, 2008);
0.5R the distance between the two inflection points being the same length as the resolution, an option in Chimera (Petterson et al, 2004);
1R where the sigma value simply equal to the resolution, as used by NMFF (Tama et al, 2004).
*number_top_mod*
Number of Fits to cluster. Default is all.
*write*
True will write out a file that contains the list of the structure instances representing different fits scored and clustered.
note the lrms column is the Calpha RMSD of each fit from the first fit in its class
"""
blurrer = StructureBlurrer()
scorer = ScoringFunctions()
cluster = Cluster()
count = 0
dict_ensembl = {}
list_to_order = []
#print targetMap
if targetMap == False:
#targetMap = self.protMap(prot, min(resolution/4., 3.5), resolution)
print("WARNING:Need target map")
sys.exit()
if score not in [
'CCC', 'LAP', 'MI', 'NV', 'NV_Sobel', 'NV_Laplace', 'ENV', 'CD'
]:
print('Incorrect Scoring Function: %s', score)
print(
'Please select from one of the following scoring functions: %s',
''.join([
'CCC', 'LAP', 'MI', 'NV', 'NV_Sobel', 'NV_Laplace', 'ENV',
'CD'
]))
sys.exit()
targetMap = targetMap.copy()
if score == 'CCC':
for mod1 in ensemble_list:
count += 1
name_mod = mod1[0]
mod = mod1[1]
sim_map = blurrer.gaussian_blur(mod,
res_target_map,
densMap=targetMap,
sigma_coeff=sigma_coeff)
if not cont_targetMap is None:
score_mod = scorer.CCC_map(
sim_map, targetMap, 0.5 * sim_map.fullMap.std(),
cont_targetMap, 2, True)[0] #CCC(sim_map,targetMap)
else:
score_mod = scorer.CCC_map(sim_map, targetMap, 0.0, 0.0,
True)[0]
#else: score_mod=scorer.CCC(sim_map,targetMap)
#'name_file','structure_instance','score','lrmsd','class'
list_to_order.append([name_mod, mod, score_mod, 0, 0])
if score == 'LAP':
for mod1 in ensemble_list:
count += 1
name_mod = mod1[0]
mod = mod1[1]
sim_map = blurrer.gaussian_blur(mod,
res_target_map,
densMap=targetMap,
sigma_coeff=sigma_coeff)
score_mod = scorer.laplace_CCC(sim_map, targetMap)
#'name_file','structure_instance','score','lrmsd','class'
list_to_order.append([name_mod, mod, score_mod, 0, 0])
if score == 'MI':
for mod1 in ensemble_list:
count += 1
name_mod = mod1[0]
mod = mod1[1]
sim_map = blurrer.gaussian_blur(mod,
res_target_map,
densMap=targetMap,
sigma_coeff=sigma_coeff)
if not cont_targetMap is None:
score_mod = scorer.MI(sim_map, targetMap,
0.5 * sim_map.fullMap.std(),
cont_targetMap, 1)
else:
score_mod = scorer.MI(sim_map, targetMap)
list_to_order.append([name_mod, mod, score_mod, 0, 0])
if score == 'NV':
for mod1 in ensemble_list:
count += 1
name_mod = mod1[0]
mod = mod1[1]
#These two values should be calculated for the experimental map, and only
#need to be calculated once, at the beginning
sim_map = blurrer.gaussian_blur(mod,
res_target_map,
densMap=targetMap,
sigma_coeff=sigma_coeff)
if not cont_targetMap is None:
score_mod = scorer.normal_vector_score(
targetMap,
sim_map,
cont_targetMap - (0.1 * targetMap.std()),
cont_targetMap + (0.1 * targetMap.std()),
Filter=None)
else:
min_thr = targetMap.get_primary_boundary(
mod.get_prot_mass_from_atoms(), targetMap.min(),
targetMap.max())
points = targetMap.get_point_map(min_thr, percentage=0.2)
max_thr = targetMap.get_second_boundary(min_thr,
points,
min_thr,
targetMap.max(),
err_percent=1)
score_mod = scorer.normal_vector_score(targetMap,
sim_map,
min_thr,
max_thr,
Filter=None)
score_mod = 1 - (score_mod / 3.14)
list_to_order.append([name_mod, mod, score_mod, 0, 0])
if score == 'NV_Sobel':
for mod1 in ensemble_list:
count += 1
name_mod = mod1[0]
mod = mod1[1]
sim_map = blurrer.gaussian_blur(mod,
res_target_map,
densMap=targetMap,
sigma_coeff=sigma_coeff)
if not cont_targetMap is None:
score_mod = scorer.normal_vector_score(
targetMap,
sim_map,
cont_targetMap - (0.1 * targetMap.std()),
cont_targetMap + (0.1 * targetMap.std()),
Filter='Sobel')
else:
min_thr = targetMap.get_primary_boundary(
mod.get_prot_mass_from_atoms(), targetMap.min(),
targetMap.max())
points = targetMap.get_point_map(min_thr, percentage=0.2)
max_thr = targetMap.get_second_boundary(min_thr,
points,
min_thr,
targetMap.max(),
err_percent=1)
score_mod = scorer.normal_vector_score(targetMap,
sim_map,
min_thr,
max_thr,
Filter='Sobel')
score_mod = 1 - (score_mod / 3.14)
list_to_order.append([name_mod, mod, score_mod, 0, 0])
if score == 'NV_Laplace':
for mod1 in ensemble_list:
count += 1
name_mod = mod1[0]
mod = mod1[1]
sim_map = blurrer.gaussian_blur(mod,
res_target_map,
densMap=targetMap,
sigma_coeff=sigma_coeff)
if not cont_targetMap is None:
score_mod = scorer.normal_vector_score(
targetMap,
sim_map,
cont_targetMap - (0.1 * targetMap.std()),
cont_targetMap + (0.1 * targetMap.std()),
Filter='Laplace')
else:
min_thr = targetMap.get_primary_boundary(
mod.get_prot_mass_from_atoms(), targetMap.min(),
targetMap.max())
points = targetMap.get_point_map(min_thr, percentage=0.2)
max_thr = targetMap.get_second_boundary(min_thr,
points,
min_thr,
targetMap.max(),
err_percent=1)
score_mod = scorer.normal_vector_score(targetMap,
sim_map,
min_thr,
max_thr,
Filter='Laplace')
score_mod = 1 - (score_mod / 3.14)
list_to_order.append([name_mod, mod, score_mod, 0, 0])
if score == 'ENV':
for mod1 in ensemble_list:
count += 1
name_mod = mod1[0]
mod = mod1[1]
min_thr = targetMap.get_primary_boundary(
mod.get_prot_mass_from_atoms(), targetMap.min(),
targetMap.max())
score_mod = scorer.envelope_score(targetMap, min_thr, mod)
#'name_file','structure_instance','score','lrmsd','class'
list_to_order.append([name_mod, mod, score_mod, 0, 0])
if score == 'CD':
for mod1 in ensemble_list:
count += 1
name_mod = mod1[0]
mod = mod1[1]
sim_map = blurrer.gaussian_blur(mod,
res_target_map,
densMap=targetMap,
sigma_coeff=sigma_coeff)
if not cont_targetMap is None:
score_mod = scorer._surface_distance_score(
sim_map, targetMap, 0.5 * sim_map.fullMap.std(),
cont_targetMap, 'Minimum')
else:
min_thr = targetMap.get_primary_boundary(
mod.get_prot_mass_from_atoms(), targetMap.min(),
targetMap.max())
points = targetMap.get_point_map(min_thr, percentage=0.2)
max_thr = targetMap.get_second_boundary(min_thr,
points,
min_thr,
targetMap.max(),
err_percent=1)
score_mod = scorer.chamfer_distance(sim_map,
targetMap,
min_thr,
max_thr,
kdtree=None)
score_mod = 1 / score_mod
list_to_order.append([name_mod, mod, score_mod, 0, 0])
if score in ['NV', 'NV_Sobel', 'NV_Laplace']:
list_ordered = sorted(
list_to_order, key=lambda x: x[2],
reverse=True) #was false when NV was negative
else:
list_ordered = sorted(list_to_order,
key=lambda x: x[2],
reverse=True)
if number_top_mod == 0:
if write == True:
return cluster._print_results_cluster2(list_ordered, write)
return list_ordered
else:
x = int(number_top_mod)
if write == True:
return cluster._print_results_cluster2(list_ordered[:x], write)
return list_ordered[:x]
def transform_map(self, matR, transvec, m1, m2, c1, c2):
mat = matR.T
emmap1 = MapParser.readMRC(m1)
emmap2 = MapParser.readMRC(m2)
# geometric centre of map
vec_centre = emmap2.centre()
spacing = emmap2.apix
# to work on the box transformations, get the box centre irrespective of origin
vec_centre.x = vec_centre.x - emmap2.x_origin()
vec_centre.y = vec_centre.y - emmap2.y_origin()
vec_centre.z = vec_centre.z - emmap2.z_origin()
# calculate new box dimensions, after rotation
new_centre = emmap2._box_transform(matR)
output_shape = (int(new_centre.x / spacing),
int(new_centre.y / spacing),
int(new_centre.z / spacing))
new_centre.x = new_centre.x / 2
new_centre.y = new_centre.y / 2
new_centre.z = new_centre.z / 2
# offset for rotation
offset = emmap2._rotation_offset(mat, vec_centre, new_centre)
#APPLY ROTATION
emmap2 = emmap2._matrix_transform_offset(mat, output_shape, offset.x,
offset.y, offset.z)
offset_x = new_centre.x - vec_centre.x
offset_y = new_centre.y - vec_centre.y
offset_z = new_centre.z - vec_centre.z
emmap2 = emmap2.shift_origin(-offset_x, -offset_y, -offset_z)
# TRANSLATION COMPONENT
a14, a24, a34 = transvec[0], transvec[1], transvec[2]
emmap_2 = emmap2.shift_origin(
float(a14) * spacing,
float(a24) * spacing,
float(a34) * spacing)
emmap_1 = emmap1.copy()
# CROP BOX TO REDUCE ARRAY SIZE
emmap_1._crop_box(c1, 2)
emmap_2._crop_box(c2, 2)
# DETERMINE A COMMON ALIGNMENT BOX
spacing = emmap_2.apix
if emmap_2.apix < emmap_1.apix: spacing = emmap_1.apix
grid_shape, new_ori = emmap_1._alignment_box(emmap_2, spacing)
# INTERPOLATE TO NEW GRID
emmap_1 = emmap_1._interpolate_to_grid(grid_shape, spacing, new_ori)
emmap_2 = emmap_2._interpolate_to_grid(grid_shape, spacing, new_ori)
sc = ScoringFunctions()
ccc = sc.CCF_mask_zero(emmap_1, emmap_2, c1, c2)
mi = sc.MI(emmap_1, emmap_2)
env = sc.map_envelope_score(emmap_1, emmap_2, c1, c2)
nv = sc.normal_vector_score(emmap_1, emmap_2,
float(c1) - (emmap1.std() * 0.05),
float(c1) + (emmap1.std() * 0.05))
nv = sc.normal_vector_score(emmap_1,
emmap_2,
float(c1) - (emmap1.std() * 0.05),
float(c1) + (emmap1.std() * 0.05),
Filter='Sobel')
return ccc, mi, env, nv, nv_s
def test_tempy_nmi(self):
''' Test the tempy nmi score based on the files
provided. Use this as a baseline for the second
chimeraX test. '''
path_test = "./"
m = os.path.join(path_test, 'emd_5168.map')
p = os.path.join(path_test, 'emd_5170.map')
sc = ScoringFunctions()
rez1 = 6.6
rez2 = 15.0
Name1, emmap1, c1 = map_contour(m, t=1.5)
Name2, emmap2, c2 = map_contour(p, t=1.5)
print(rez1, rez2, c1, c2, emmap1.apix, emmap2.apix)
if not sc.mapComparison(emmap1, emmap2):
emmap1._crop_box(c1, 0.5)
emmap2._crop_box(c2, 0.5)
if rez1 > 1.25 * rez2:
emmap_2 = lpfilter(emmap2, rez1)
emmap1, emmap2 = match_grid(emmap1, emmap_2, c1, c2)
elif rez2 > 1.25 * rez1:
emmap_1 = lpfilter(emmap1, rez2)
emmap1, emmap2 = match_grid(emmap_1, emmap2, c1, c2)
else:
emmap1, emmap2 = match_grid(emmap1, emmap2, c1, c2)
nmi = 0
try:
nmi = sc.MI(emmap1, emmap2, c1, c2, 1, None, None, True)
if nmi < 0.0: nmi = 0.0
except:
self.assertTrue(False)
print_exc()
nmi = 0.0
self.assertTrue(abs(round(nmi, 5) - 1.0492) < 0.001)
# Now test with a model and map
p = os.path.join(path_test, '1J6Z.pdb')
m = os.path.join(path_test, 'emd_5168_monomer.mrc')
res = 6.6
Name1 = os.path.basename(m).split('.')[0]
Name2 = os.path.basename(p).split('.')[0]
emmap1 = MapParser.readMRC(m)
structure_instance = PDBParser.read_PDB_file(Name2,
p,
hetatm=False,
water=False)
blurrer = StructureBlurrer()
emmap2 = blurrer.gaussian_blur(structure_instance, res, densMap=emmap1)
c1 = 9.7
c2 = 1.0
nmi = 0
try:
nmi = sc.MI(emmap1, emmap2, c1, c2, 1, None, None, True)
if nmi < 0.0: nmi = 0.0
except:
self.assertTrue(False)
print_exc()
nmi = 0.0
self.assertTrue(abs(round(nmi, 5) - 1.0575) < 0.001)
def score_cmd(session, comparators, compared, rez_comparators, rez_compared, contours_comparators, contour_compared):
sc = ScoringFunctions()
blurrer = StructureBlurrer()
# Loop through these to be compared
idx = 0
scores = []
for comparator in comparators:
emmap1 = None
emmap2 = None
if type(comparator) is AtomicStructure:
if type(compared) is AtomicStructure:
# Both models
if None in ([rez_compared] + rez_comparators):
print("Please provide the resolution for all models")
return
bms1 = chimera_to_tempy_model(compared)
bms2 = chimera_to_tempy_model(comparator)
emmap1 = model_contour( bms1, rez_compared, emmap=False,t=0.5)
if contours_comparators[idx] is None:
emmap2 = model_contour(bms2, rez_comparators[idx],emmap=False,t=0.5)
else:
emmap2 = blur_model(bms2, rez_comparators[idx], emmap=False)
else:
# 0 - map, 1 - model
if rez_comparators[idx] == None:
print("Please provide the resolution for the model.")
return
emmap1 = chimera_to_tempy_map(compared)
bms = chimera_to_tempy_model(comparator)
emmap2 = blurrer.gaussian_blur(bms, rez_compared, densMap=emmap1)
else:
if type(compared) is AtomicStructure:
# 0 - model, 1 - map
if rez_compared == None:
print("Please provide the resolution for the model.")
return
emmap2 = chimera_to_tempy_map(comparator)
bms = chimera_to_tempy_model(compared)
emmap1 = blurrer.gaussian_blur(bms, rez_compared, densMap=emmap2)
else:
# 0 - map, 1 - map
emmap1 = chimera_to_tempy_map(compared)
emmap2 = chimera_to_tempy_map(comparator)
c1 = contour_compared
# Contouring
if c1 == None:
c1 = map_contour(emmap1,t=1.5)
c2 = contours_comparators[idx]
# This kinda makes no sense and could be tricky
if c2 == None:
c2 = map_contour(emmap2,t=1.5)
# Some kind of fix if the maps don't match?
# Resize, resample or blur of somekind
if not sc.mapComparison(emmap1,emmap2):
emmap1._crop_box(c1,0.5)
emmap2._crop_box(c2,0.5)
if rez_compared > 1.25*rez_comparators[idx]:
emmap_2 = lpfilter(emmap2,rez_compared)
emmap1, emmap2 = match_grid(emmap1,emmap_2,c1,c2)
elif rez_comparators[idx] > 1.25*rez_compared:
emmap_1 = lpfilter(emmap1,rez_comparators[idx])
emmap1, emmap2 = match_grid(emmap_1,emmap2,c1,c2)
else:
emmap1, emmap2 = match_grid(emmap1,emmap2,c1,c2)
nmi = 0.0
try:
nmi = sc.MI(emmap1,emmap2,c1,c2,1,None,None,True)
if nmi < 0.0: nmi = 0.0
except:
print('Exception for NMI score')
print_exc()
nmi = 0.0
scores.append(nmi)
idx+=1
return scores
def score(session, atomic_model1 = None, map_model1 = None, atomic_model2 = None, map_model2 = None, rez1 = None, rez2 = None, c1 = None, c2 = None):
""" Generate the NMI score for 2 maps or 1 map and 1 model. """
sc = ScoringFunctions()
# We have choices - 1 map and one model, 2 maps or 2 models
emmap1 = None
emmap2 = None
blurrer = StructureBlurrer()
if atomic_model1 != None and map_model1 != None:
# 1 map 1 model
if rez1 == None:
print("Please provide the resolution for the model.")
return
emmap1 = chimera_to_tempy_map(map_model1)
bms = chimera_to_tempy_model(atomic_model1)
emmap2 = blurrer.gaussian_blur(bms, rez1, densMap=emmap1)
elif map_model1 != None and map_model2 != None:
# 2 maps
emmap1 = chimera_to_tempy_map(map_model1)
emmap2 = chimera_to_tempy_map(map_model2)
elif atomic_model1 != None and atomic_model2 != None:
# 2 models
if None in [rez1,rez2]:
print("Please provide the resolution for both model")
return
bms1 = chimera_to_tempy_model(atomic_model1)
bms2 = chimera_to_tempy_model(atomic_model2)
emmap1 = model_contour( bms1, rez1, emmap=False,t=0.5)
if c2 is None:
emmap2 = model_contour(bms2, rez2,emmap=False,t=0.5)
else:
emmap2 = blur_model( bms2, rez2, emmap=False)
else:
print("Error. Must have 1 model and 1 map, 2 maps or 2 models")
return
# Contouring
if c1 == None:
c1 = map_contour(emmap1,t=1.5)
if c2 == None:
c2 = map_contour(emmap2,t=1.5)
# Some kind of fix if the maps don't match?
# Resize, resample or blur of somekind
if not sc.mapComparison(emmap1,emmap2):
emmap1._crop_box(c1,0.5)
emmap2._crop_box(c2,0.5)
if rez1 > 1.25*rez2:
emmap_2 = lpfilter(emmap2,rez1)
emmap1, emmap2 = match_grid(emmap1,emmap_2,c1,c2)
elif rez2 > 1.25*rez1:
emmap_1 = lpfilter(emmap1,rez2)
emmap1, emmap2 = match_grid(emmap_1,emmap2,c1,c2)
else:
emmap1, emmap2 = match_grid(emmap1,emmap2,c1,c2)
nmi = 0.0
try:
nmi = sc.MI(emmap1,emmap2,c1,c2,1,None,None,True)
if nmi < 0.0: nmi = 0.0
except:
print('Exception for NMI score')
print_exc()
nmi = 0.0
return nmi
map_target=MapParser.readMRC('emd_5168_monomer.mrc') #read target map
map_probe = blurrer.gaussian_blur(structure_instance, 6.6,densMap=map_target)#create a simulated map from the structure instance
map_probe.write_to_MRC_file("map_probe_actin.mrc") #write simulated map to a MRC file format
##SCORING FUNCTION
print "Calculate Envelope Score (ENV):"
molecualr_weight=structure_instance.get_prot_mass_from_atoms()
#Mmolecualr_weight=structure_instance.get_prot_mass_from_res()
first_bound=map_target.get_primary_boundary(molecualr_weight, map_target.min(), map_target.max())
#print scorer.envelope_score_APJ(map_target, first_bound, structure_instance,norm=True)
print scorer.envelope_score(map_target, first_bound, structure_instance,norm=True)
print "Calculate Mutual information Score (MI)"
print scorer.MI(map_target,map_probe)
print "Calculate Laplacian cross-correlation Score (LAP)"
print scorer.laplace_CCC(map_target,map_probe)
print "Calculate cross-correlation Score (CCC)"
print scorer.CCC(map_target,map_probe)
print "Calculate Normal Vector (NV):"
#Number of points to use in the normal vector score
points= round((map_target.map_size())*0.01)
first_bound=map_target.get_primary_boundary(structure_instance.get_prot_mass_from_atoms(), map_target.min(), map_target.max())
second_bound=map_target.get_second_boundary(first_bound, points, first_bound, map_target.max(),err_percent=1)
print scorer.normal_vector_score(map_target,map_probe, first_bound, second_bound)
print "Calculate Normal Vector (NV) with Sobel Filter:"