def __init__(self,
pair_asu_table,
site_labels,
sites_frac=None,
sites_cart=None,
covariance_matrix=None,
cell_covariance_matrix=None,
parameter_map=None,
include_bonds_to_hydrogen=False,
fixed_distances=None,
eps=2e-16):
assert [sites_frac, sites_cart].count(None) == 1
fmt = "%.4f"
asu_mappings = pair_asu_table.asu_mappings()
space_group_info = sgtbx.space_group_info(group=asu_mappings.space_group())
unit_cell = asu_mappings.unit_cell()
if sites_cart is not None:
sites_frac = unit_cell.fractionalize(sites_cart)
self.loop = model.loop(header=(
"_geom_bond_atom_site_label_1",
"_geom_bond_atom_site_label_2",
"_geom_bond_distance",
"_geom_bond_site_symmetry_2"
))
distances = crystal.calculate_distances(
pair_asu_table, sites_frac,
covariance_matrix=covariance_matrix,
cell_covariance_matrix=cell_covariance_matrix,
parameter_map=parameter_map)
for d in distances:
if (not include_bonds_to_hydrogen
and (site_labels[d.i_seq].startswith('H') or
site_labels[d.j_seq].startswith('H'))):
continue
if (d.variance is not None and d.variance > eps
and not(fixed_distances is not None and
((d.i_seq, d.j_seq) in fixed_distances or
(d.j_seq, d.i_seq) in fixed_distances))):
distance = format_float_with_su(d.distance, math.sqrt(d.variance))
else:
distance = fmt % d.distance
sym_code = space_group_info.cif_symmetry_code(d.rt_mx_ji)
if sym_code == "1": sym_code = "."
self.loop.add_row((site_labels[d.i_seq],
site_labels[d.j_seq],
distance,
sym_code))
self.distances = distances.distances
self.variances = distances.variances
self.pair_counts = distances.pair_counts
def find_basis_vector_combinations_cluster_analysis(self):
# hijack the xray.structure class to facilitate calculation of distances
xs = xray.structure(crystal_symmetry=self.crystal_symmetry)
for i, site in enumerate(self.sites):
xs.add_scatterer(xray.scatterer("C%i" %i, site=site))
xs = xs.sites_mod_short()
xs = xs.select(xs.sites_frac().norms() < 0.45)
cell_multiplier = 10
xs1 = xs.customized_copy(
unit_cell=uctbx.unit_cell([xs.unit_cell().parameters()[0]*cell_multiplier]*3))
xs1.set_sites_cart(xs.sites_cart())
xs = xs1
sites_cart = xs.sites_cart()
lengths = flex.double([matrix.col(sc).length() for sc in sites_cart])
xs = xs.select(flex.sort_permutation(lengths))
if self.params.debug:
with open('peaks.pdb', 'wb') as f:
print >> f, xs.as_pdb_file()
vector_heights = flex.double()
sites_frac = xs.sites_frac()
pair_asu_table = xs.pair_asu_table(distance_cutoff=self.params.max_cell)
asu_mappings = pair_asu_table.asu_mappings()
distances = crystal.calculate_distances(pair_asu_table, sites_frac)
vectors = []
difference_vectors = []
pairs = []
for di in distances:
if di.distance < self.params.min_cell: continue
i_seq, j_seq = di.i_seq, di.j_seq
if i_seq > j_seq: continue
pairs.append((i_seq, j_seq))
rt_mx_ji = di.rt_mx_ji
site_frac_ji = rt_mx_ji * sites_frac[j_seq]
site_cart_ji = xs.unit_cell().orthogonalize(site_frac_ji)
site_cart_i = xs.unit_cell().orthogonalize(sites_frac[i_seq])
vectors.append(matrix.col(site_cart_ji))
diff_vec = matrix.col(site_cart_i) - matrix.col(site_cart_ji)
if diff_vec[0] < 0:
# only one hemisphere of difference vector space
diff_vec = -diff_vec
difference_vectors.append(diff_vec)
params = self.params.multiple_lattice_search.cluster_analysis
if params.method == 'dbscan':
i_cluster = self.cluster_analysis_dbscan(difference_vectors)
min_cluster_size = 1
elif params.method == 'hcluster':
i_cluster = self.cluster_analysis_hcluster(difference_vectors)
i_cluster -= 1 # hcluster starts counting at 1
min_cluster_size = params.min_cluster_size
if self.params.debug_plots:
self.debug_plot_clusters(
difference_vectors, i_cluster, min_cluster_size=min_cluster_size)
clusters = []
min_cluster_size = params.min_cluster_size
for i in range(max(i_cluster)+1):
isel = (i_cluster == i).iselection()
if len(isel) < min_cluster_size:
continue
clusters.append(isel)
cluster_point_sets = []
centroids = []
cluster_sizes = flex.int()
difference_vectors = flex.vec3_double(difference_vectors)
from libtbx.utils import flat_list
for cluster in clusters:
points = flat_list([pairs[i] for i in cluster])
cluster_point_sets.append(set(points))
d_vectors = difference_vectors.select(cluster)
cluster_sizes.append(len(d_vectors))
centroids.append(d_vectors.mean())
# build a graph where each node is a centroid from the difference vector
# cluster analysis above, and an edge is defined when there is a
# significant overlap between the sets of peaks in the FFT map that
# contributed to the difference vectors in two clusters
import networkx as nx
G = nx.Graph()
G.add_nodes_from(range(len(cluster_point_sets)))
cutoff_frac = 0.25
for i in range(len(cluster_point_sets)):
for j in range(i+1, len(cluster_point_sets)):
intersection_ij = cluster_point_sets[i].intersection(
cluster_point_sets[j])
union_ij = cluster_point_sets[i].union(cluster_point_sets[j])
frac_connected = len(intersection_ij)/len(union_ij)
if frac_connected > cutoff_frac:
G.add_edge(i, j)
# iteratively find the maximum cliques in the graph
# break from the loop if there are no cliques remaining or there are
# fewer than 3 vectors in the remaining maximum clique
# Allow 1 basis vector to be shared between two cliques, to allow for
# cases where two lattices share one basis vectors (e.g. two plate
# crystals exactly aligned in one direction, but not in the other two)
distinct_cliques = []
cliques = list(nx.find_cliques(G))
cliques = sorted(cliques, key=len, reverse=True)
for i, clique in enumerate(cliques):
clique = set(clique)
if len(clique) < 3:
break
is_distinct = True
for c in distinct_cliques:
if len(c.intersection(clique)) > 1:
is_distinct = False
break
if is_distinct:
distinct_cliques.append(clique)
this_set = set()
for i_cluster in clique:
this_set = this_set.union(cluster_point_sets[i_cluster])
logger.info("Clique %i: %i lattice points" %(i+1, len(this_set)))
assert len(distinct_cliques) > 0
logger.info("Estimated number of lattices: %i" %len(distinct_cliques))
self.candidate_basis_vectors = []
self.candidate_crystal_models = []
for clique in distinct_cliques:
sel = flex.size_t(list(clique))
vectors = flex.vec3_double(centroids).select(sel)
perm = flex.sort_permutation(vectors.norms())
vectors = [matrix.col(vectors[p]) for p in perm]
# exclude vectors that are (approximately) integer multiples of a shorter
# vector
unique_vectors = []
for v in vectors:
is_unique = True
for v_u in unique_vectors:
if is_approximate_integer_multiple(v_u, v,
relative_tolerance=0.01,
angular_tolerance=0.5):
is_unique = False
break
if is_unique:
unique_vectors.append(v)
vectors = unique_vectors
self.candidate_basis_vectors.extend(vectors)
candidate_orientation_matrices \
= self.find_candidate_orientation_matrices(
vectors,
max_combinations=self.params.basis_vector_combinations.max_try)
if len(candidate_orientation_matrices) == 0:
continue
crystal_model, n_indexed = self.choose_best_orientation_matrix(
candidate_orientation_matrices)
if crystal_model is None: continue
# map to minimum reduced cell
crystal_symmetry = crystal.symmetry(
unit_cell=crystal_model.get_unit_cell(),
space_group=crystal_model.get_space_group())
cb_op = crystal_symmetry.change_of_basis_op_to_minimum_cell()
crystal_model = crystal_model.change_basis(cb_op)
self.candidate_crystal_models.append(crystal_model)
if self.params.debug:
file_name = "vectors.pdb"
a = self.params.max_cell
cs = crystal.symmetry(unit_cell=(a,a,a,90,90,90), space_group="P1")
xs = xray.structure(crystal_symmetry=cs)
for v in difference_vectors:
v = matrix.col(v)
xs.add_scatterer(xray.scatterer("C", site=v/(a/10)))
xs.sites_mod_short()
with open(file_name, 'wb') as f:
print >> f, xs.as_pdb_file()
for crystal_model in self.candidate_crystal_models:
logger.debug(crystal_model)
