def compare_ffs(in_dict, conf_id_tag, out_prefix, mol_slice=None):
"""
For 2+ SDF files that are analogous in terms of molecules and their
conformers, assess them by RMSD, TFD, and relative energy differences.
Parameters
----------
in_dict : OrderedDict
dictionary from input file, where key is method and value is dictionary
first entry should be reference method
in sub-dictionary, keys are 'sdfile' and 'sdtag'
conf_id_tag : string
label of the SD tag that should be the same for matching conformers
in different files
out_prefix : string
prefix appended to sdf file name to write out new SDF file
with RMSD and TFD info added as SD tags
mol_slice : numpy slice object
The resulting integers are numerically sorted and duplicates removed.
e.g., slices = np.s_[0, 3:5, 6::3] would be parsed to return
[0, 3, 4, 6, 9, 12, 15, 18, ...]
Can also parse from end: [-3:] gets the last 3 molecules, and
[-2:-1] is the same as [-2] to get just next to last molecule.
Returns
-------
enes_full : 3D list
enes_full[i][j][k] = ddE of ith method, jth mol, kth conformer.
ddE = (dE of query method) - (dE of ref method),
where the dE is computed as conformer M - conformer N,
and conformer N is chosen from the lowest energy of the ref confs.
the reference method is not present; i.e., self-comparison is skipped,
so the max i value represents total number of files minus one.
rmsds_full : 3D list
same format as that of enes_full but with conformer RMSDs
tfds_full : 3D list
same format as that of enes_full but with conformer TFDs
smiles_full : 3D list
same format as that of enes_full but with conformer SMILES strings
"""
# set RMSD calculation parameters
automorph = True # take into acct symmetry related transformations
heavyOnly = False # do consider hydrogen atoms for automorphisms
overlay = True # find the lowest possible RMSD
# initiate final data lists
enes_full = []
rmsds_full = []
tfds_full = []
smiles_full = []
# get first filename representing the reference geometries
sdf_ref = list(in_dict.values())[0]['sdfile']
tag_ref = list(in_dict.values())[0]['sdtag']
# assess each file against reference
for ff_label, ff_dict in in_dict.items():
# get details of queried file
sdf_que = ff_dict['sdfile']
tag_que = ff_dict['sdtag']
if sdf_que == sdf_ref:
continue
# initiate new sublists
enes_method = []
rmsds_method = []
tfds_method = []
smiles_method = []
# open an output file to store query molecules with new SD tags
out_file = f'{out_prefix}_{os.path.basename(sdf_que)}'
ofs = oechem.oemolostream()
if not ofs.open(out_file):
oechem.OEThrow.Fatal(f"Unable to open {out_file} for writing")
# load molecules from open reference and query files
print(f"\n\nOpening reference file {sdf_ref}")
mols_ref = reader.read_mols(sdf_ref, mol_slice)
print(f"Opening query file {sdf_que} for [ {ff_label} ] energies")
mols_que = reader.read_mols(sdf_que, mol_slice)
# loop over each molecule in reference and query files
for rmol, qmol in zip(mols_ref, mols_que):
# initial check that they have same title and number of confs
rmol_name = rmol.GetTitle()
rmol_nconfs = rmol.NumConfs()
if (rmol_name != qmol.GetTitle()) or (rmol_nconfs !=
qmol.NumConfs()):
raise ValueError(
"ERROR: Molecules not aligned in iteration. "
"Offending molecules and number of conformers:\n"
f"\'{rmol_name}\': {rmol_nconfs} nconfs\n"
f"\'{qmol.GetTitle()}\': {qmol.NumConfs()} nconfs")
# initialize lists to store conformer energies
enes_ref = []
enes_que = []
rmsds_mol = []
tfds_mol = []
smiles_mol = []
# loop over each conformer of this mol
for ref_conf, que_conf in zip(rmol.GetConfs(), qmol.GetConfs()):
# check confomer match from the specified tag
ref_id = oechem.OEGetSDData(ref_conf, conf_id_tag)
que_id = oechem.OEGetSDData(que_conf, conf_id_tag)
if ref_id != que_id:
raise ValueError(
"ERROR: Conformers not aligned in iteration"
f" for mol: '{rmol_name}'. The conformer "
f"IDs ({conf_id_tag}) for ref and query are:"
f"\n{ref_id}\n{que_id}.")
# note the smiles id
smiles_mol.append(ref_id)
# get energies
enes_ref.append(float(oechem.OEGetSDData(ref_conf, tag_ref)))
enes_que.append(float(oechem.OEGetSDData(que_conf, tag_que)))
# compute RMSD between reference and query conformers
rmsd = oechem.OERMSD(ref_conf, que_conf, automorph, heavyOnly,
overlay)
rmsds_mol.append(rmsd)
# compute TFD between reference and query conformers
tfd = calc_tfd(ref_conf, que_conf, conf_id_tag)
tfds_mol.append(tfd)
# store data in SD tags for query conf, and write conf to file
oechem.OEAddSDData(que_conf, f'RMSD to {sdf_ref}', str(rmsd))
oechem.OEAddSDData(que_conf, f'TFD to {sdf_ref}', str(tfd))
oechem.OEWriteConstMolecule(ofs, que_conf)
# compute relative energies against lowest E reference conformer
lowest_ref_idx = enes_ref.index(min(enes_ref))
rel_enes_ref = np.array(enes_ref) - enes_ref[lowest_ref_idx]
rel_enes_que = np.array(enes_que) - enes_que[lowest_ref_idx]
# subtract them to get ddE = dE (query method) - dE (ref method)
enes_mol = np.array(rel_enes_que) - np.array(rel_enes_ref)
# store then move on
enes_method.append(enes_mol)
rmsds_method.append(np.array(rmsds_mol))
tfds_method.append(np.array(tfds_mol))
smiles_method.append(smiles_mol)
#print(rmsds_method, len(rmsds_method))
#print(enes_method, len(enes_method))
enes_full.append(enes_method)
rmsds_full.append(rmsds_method)
tfds_full.append(tfds_method)
smiles_full.append(smiles_method)
ofs.close()
return enes_full, rmsds_full, tfds_full, smiles_full
def tailed_parameters(in_sdf, ffxml, cutoff, tag, tag_smiles, metric_type):
"""
Extract data from SD tags, identify outlier molecules above cutoff,
and get the associated force field parameters for each structure.
Parameters
----------
in_sdf : string
name of the input SDF molecule file with RMSD or TFD info as SD tags
ffxml : string
name of the FFXML force field file
cutoff : float
cutoff value to use for the metric; structures with value above
the cutoff are considered outliers
tag : string
name of the SD tag in the SDF file with the metric information
tag_smiles : string
name of the SD tag in the SDF file with the molecule identifier
metric_type : string
what metric the tag and cutoff refer to (e.g., TFD or RMSD)
for plot and figure labeling
Returns
-------
data_all : dict
key of 'count' has int for count of all structures
key of 'mols_dict' has dict for dictionary of mols (see note)
key of 'params_mol' has dict of isosmiles and list of parameter IDs
key of 'params_id' has dict of parameter IDs and list of isosmiles
key of 'smi_dict' has dict of isosmiles keys and SD tag identifier
data_out : dict
same format as data_all but only containing outlier molecules
Note
----
The mols_dict dictionary is itself a dict of dicts where the first
level key is the SMILES string (or specified molecular identifier),
and the value of that key is a dict with the following key/value pairs:
metric geometric measurement
structure OEGraphMol of the structure
"""
# load molecules from open reference and query files
print(f"\n\n\nOpening SDF file {in_sdf}...")
mols = reader.read_mols(in_sdf)
print(
f"Looking for outlier molecules with {metric_type.upper()} above {cutoff}...\n"
)
# find the molecules with the metric above the cutoff
all_smiles = []
mols_all = OrderedDict()
mols_out = OrderedDict()
count_all = 0
count_out = 0
for mol in mols:
for conf in mol.GetConfs():
smiles = oechem.OEGetSDData(conf, tag_smiles)
try:
value = float(oechem.OEGetSDData(conf, tag))
except ValueError as e:
raise ValueError(
"There was an error while obtaining the SD "
f"tag data of '{oechem.OEGetSDData(conf, tag)}'. Did you "
f"specify the correct SD tag of '{tag}'?")
if value >= cutoff:
mols_out[smiles] = {
'metric': value,
'structure': oechem.OEGraphMol(conf)
}
count_out += 1
mols_all[smiles] = {'structure': oechem.OEGraphMol(conf)}
all_smiles.append(smiles)
count_all += 1
# save outliers molecules to file
write_mols(mols_out, f'outliers_{metric_type}.mol2')
# analyze parameters in the outlier and full sets
params_mol_out, params_id_out, smi_dict_out = get_parameters(
mols_out, ffxml)
params_mol_all, params_id_all, smi_dict_all = get_parameters(
mols_all, ffxml)
# organize all computed data to encompassing dictionary
# all values in data_* are dictionaries except for data_*['count']
data_all = {
'count': count_all,
'mols_dict': mols_all,
'params_mol': params_mol_all,
'params_id': params_id_all,
'smi_dict': smi_dict_all
}
data_out = {
'count': count_out,
'mols_dict': mols_out,
'params_mol': params_mol_out,
'params_id': params_id_out,
'smi_dict': smi_dict_out
}
# save the params organized by id to pickle
with open(f'tailed_{metric_type}.pickle', 'wb') as f:
pickle.dump((data_all, data_out), f)
return data_all, data_out
def match_minima(in_dict, rmsd_cutoff):
"""
For different methods, match the conformer minima to those of the reference
method. Ex. Conf G of reference method matches with conf R of method 2.
Parameters
----------
in_dict : OrderedDict
dictionary from input file, where key is method and value is dictionary
first entry should be reference method
in sub-dictionary, keys are 'sdfile' and 'sdtag'
rmsd_cutoff : float
cutoff above which two structures are considered diff conformers
Returns
-------
mol_dict : dict of dicts
mol_dict['mol_name']['energies'] =
[[file1_conf1_E file1_conf2_E] [file2_conf1_E file2_conf2_E]]
An analogous structure is followed for mol_dict['mol_name']['indices'].
"""
# nested dictionary: 1st layer of mol names, 2nd layer of method energies
mol_dict = {}
# get first filename representing the reference geometries
sdf_ref = list(in_dict.values())[0]['sdfile']
# assess each file against reference
for ff_label, ff_dict in in_dict.items():
sdf_query = ff_dict['sdfile']
sdf_tag = ff_dict['sdtag']
# load molecules from open reference and query files
print(f"\n\nOpening reference file {sdf_ref}")
mols_ref = reader.read_mols(sdf_ref)
print(f"Opening query file {sdf_query} for [ {ff_label} ] energies")
mols_query = reader.read_mols(sdf_query)
# loop over each molecule in reference and query files
for rmol in mols_ref:
mol_name = rmol.GetTitle()
ref_nconfs = rmol.NumConfs()
run_match = False
for qmol in mols_query:
# same mol titles should mean same molecular identity;
# when same molecular identity found, break out of loop to
# start matching conformers
if rmol.GetTitle() == qmol.GetTitle():
run_match = True
break
# create entry for this mol in mol_dict if not already present
# energies [i][j] will be 2d list of ith method and jth conformer
if mol_name not in mol_dict:
mol_dict[mol_name] = {'energies': [], 'indices': []}
# no same molecules were found bt ref and query methods
# for N reference minima of each mol, P matching indices for each ref minimia
if not run_match:
print(f"No \"{mol_name}\" molecule found in {sdf_query}")
# fill in -2 error values for conformer indices
mol_dict[mol_name]['indices'].append([-2] * ref_nconfs)
# fill in nan values for conformer energies and ref_nconfs
mol_dict[mol_name]['energies'].append([np.nan] * ref_nconfs)
# reset mols_query generator
mols_query = reader.read_mols(sdf_query)
# continue with the next rmol
continue
# get data from specified sd tag for all conformers
data_confs = reader.get_sd_list(qmol, sdf_tag)
# format sd tag data to float types
float_data_confs = list(map(float, data_confs))
# store sd data from tags into dictionary
mol_dict[mol_name]['energies'].append(float_data_confs)
# don't run match if query method is same as reference method
# keep this section after sd tag extraction of energies
if sdf_query == sdf_ref:
print("Skipping comparison against self.")
mol_dict[mol_name]['indices'].append([-1] * ref_nconfs)
continue
# run the match here
# get indices of qmol conformers that match rmol conformers
molIndices = compare_two_mols(rmol, qmol, rmsd_cutoff)
mol_dict[mol_name]['indices'].append(molIndices)
return mol_dict
