def __get_mutation_center(struct_df, label_info, center_at_mut=True):
if center_at_mut:
# Use CA position of the mutated residue as center for subgrid center
sel = ((struct_df.chain == label_info.chain) &
(struct_df.residue == label_info.residue) &
(struct_df.name == 'CA'))
mutation_pos = struct_df[sel][['x', 'y', 'z']].astype(np.float32)
mutation_center = util.get_center(mutation_pos)
else:
pos = struct_df[['x', 'y', 'z']].astype(np.float32)
mutation_center = util.get_center(pos)
return mutation_center
def get_data_stats(data_filename):
"""
Get the furthest distance from the ligand's center and the number of
atoms for each structure in the dataset.
"""
data_df = pd.read_hdf(data_filename, 'structures')
data = []
for pdbcode, struct_df in data_df.groupby(['structure']):
pos = struct_df[['x', 'y', 'z']].astype(np.float32)
ligand_pos = struct_df[struct_df.chain == 'LIG'][['x', 'y', 'z'
]].astype(np.float32)
ligand_center = util.get_center(ligand_pos)
max_dist = util.get_max_distance_from_center(pos, ligand_center)
num_atoms = struct_df.shape[0]
data.append((pdbcode, max_dist, num_atoms))
df = pd.DataFrame(data, columns=['pdbcode', 'max_dist', 'num_atoms'])
df = df.sort_values(by=['max_dist', 'num_atoms'],
ascending=[False, False]).reset_index(drop=True)
print(df.describe())
print(df[df.max_dist < 20].shape[0] * 100.0 / df.shape[0])
return df
def df_to_feature(struct_df, grid_config, random_seed=None):
pos = struct_df[['x', 'y', 'z']].astype(np.float32)
center = util.get_center(pos)
rot_mat = subgrid_gen.gen_rot_matrix(grid_config, random_seed=random_seed)
grid = subgrid_gen.get_grid(
struct_df, center, config=grid_config, rot_mat=rot_mat)
return grid
def get_data_stats(sharded_list):
"""
Get the furthest distance from the protein's center and max residue ID for
every protein in the sharded dataset.
"""
data = []
all_elements = []
labels = []
for i, sharded in enumerate(sharded_list):
for shard_num, shard_df in sharded.iter_shards():
labels_df = sharded.read_shard(shard_num, key='labels')
for ensemble, ensemble_df in shard_df.groupby(['ensemble']):
all_elements.extend(ensemble_df.element.values)
subunits = ensemble_df.subunit.unique()
inactive = __get_subunit_name(subunits, mode='inactive')
active = __get_subunit_name(subunits, mode='active')
for subunit_name in [inactive, active]:
struct_df = ensemble_df[ensemble_df.subunit ==
subunit_name]
pos = struct_df[['x', 'y', 'z']].astype(np.float32)
ligand_pos = struct_df[struct_df.chain == 'L'][[
'x', 'y', 'z'
]].astype(np.float32)
ligand_center = util.get_center(ligand_pos)
max_dist = util.get_max_distance_from_center(
pos, ligand_center)
num_atoms = struct_df.shape[0]
data.append((ensemble, subunit_name, max_dist, num_atoms))
labels.append((i, shard_num, labels_df[
labels_df.ensemble == ensemble].label.values[0] == 'A'))
all_elements_df = pd.DataFrame(all_elements, columns=['element'])
unique_elements = all_elements_df.element.unique()
print('Unique elements ({:}): {:}'.format(len(unique_elements),
unique_elements))
print('\nElement counts:')
print(all_elements_df.element.value_counts())
print('\n')
all_labels_df = pd.DataFrame(labels,
columns=['sharded', 'shard_num', 'label'])
print('\nLabel dist by dataset:')
print(all_labels_df.groupby(['sharded', 'shard_num']).label.value_counts())
print('\n')
df = pd.DataFrame(data,
columns=['ensemble', 'subunit', 'max_dist', 'num_atoms'])
df = df.sort_values(by=['max_dist', 'num_atoms'],
ascending=[False, False]).reset_index(drop=True)
print(df.describe())
print(df[df.max_dist < 90].shape[0] * 100.0 / df.shape[0])
return df
def df_to_feature(struct_df, grid_config, random_seed=None):
# Use center of ligand for subgrid center
ligand_pos = struct_df[struct_df.chain == 'LIG'][['x', 'y',
'z']].astype(np.float32)
ligand_center = util.get_center(ligand_pos)
rot_mat = subgrid_gen.gen_rot_matrix(grid_config, random_seed=random_seed)
grid = subgrid_gen.get_grid(struct_df,
ligand_center,
config=grid_config,
rot_mat=rot_mat)
return grid
def get_data_stats(data_filename):
"""
Get the furthest distance from the molecule's center and the number of
atoms for each molecule in the dataset.
"""
data_df = pd.read_hdf(data_filename, 'structures')
data = []
for mol_id, mol_df in data_df.groupby(['structure']):
pos = mol_df[['x', 'y', 'z']].astype(np.float32)
max_dist = util.get_max_distance_from_center(pos, util.get_center(pos))
num_atoms = mol_df.shape[0]
data.append((mol_id, max_dist, num_atoms))
df = pd.DataFrame(data, columns=['mol_id', 'max_dist', 'num_atoms'])
df = df.sort_values(by=['max_dist', 'num_atoms'],
ascending=[False, False]).reset_index(drop=True)
print(df.describe())
print(df[df.max_dist < 7.5].shape[0] * 100.0 / df.shape[0])
return df
def get_data_stats(sharded_list):
data = []
for sharded in sharded_list:
for _, shard_df in sharded.iter_shards():
for (target,
decoy), struct_df in shard_df.groupby(['ensemble',
'subunit']):
pos = struct_df[['x', 'y', 'z']].astype(np.float32)
max_dist = util.get_max_distance_from_center(
pos, util.get_center(pos))
max_res = struct_df.residue.max()
data.append((target, decoy, max_dist, max_res))
df = pd.DataFrame(data, columns=['target', 'decoy', 'max_dist', 'max_res'])
df = df.sort_values(by=['max_dist', 'max_res'],
ascending=[False, False]).reset_index(drop=True)
print(df.describe())
print(df[df.max_dist < 50].shape[0] * 100.0 / df.shape[0])
print(df[df.max_dist < 50].target.unique().shape[0] * 100.0 /
float(df.target.unique().shape[0]))
return df
def get_data_stats(sharded_list):
"""
Get the furthest distance from the protein's center and max residue ID for
every protein in the sharded dataset.
"""
data = []
for i, sharded in enumerate(sharded_list):
for _, shard_df in sharded.iter_shards():
for (target, decoy), struct_df in shard_df.groupby(['ensemble', 'subunit']):
pos = struct_df[['x', 'y', 'z']].astype(np.float32)
max_dist = util.get_max_distance_from_center(
pos, util.get_center(pos))
max_res = struct_df.residue.max()
data.append((i, target, decoy, max_dist, max_res))
df = pd.DataFrame(data, columns=['sharded', 'target', 'decoy', 'max_dist', 'max_res'])
df = df.sort_values(by=['sharded', 'max_dist', 'max_res'],
ascending=[True, False, False]).reset_index(drop=True)
print(df.describe())
print(df[df.max_dist < 90].shape[0]*100.0/df.shape[0])
print(df[df.max_dist < 90].target.unique().shape[0]*100.0/float(df.target.unique().shape[0]))
return df
def df_to_feature(struct_df, grid_config, center_around_Cs, random_seed=None):
# Consider only atoms that have mapping for computing center.
# If <center_around_Cs> is set, consider only carbon atoms.
if center_around_Cs:
pruned_struct_df = struct_df[struct_df.element == 'C']
else:
pruned_struct_df = struct_df[struct_df.element.isin(
grid_config.element_mapping.keys())]
pos = pruned_struct_df[['x', 'y', 'z']].astype(np.float32)
# Use center of ligand for subgrid center
ligand_pos = pruned_struct_df[pruned_struct_df.chain == 'L'][[
'x', 'y', 'z'
]].astype(np.float32)
ligand_center = util.get_center(ligand_pos)
rot_mat = subgrid_gen.gen_rot_matrix(grid_config, random_seed=random_seed)
grid = subgrid_gen.get_grid(struct_df,
ligand_center,
config=grid_config,
rot_mat=rot_mat)
return grid
