def SetupRGroupDecompositionParams():
"""Setup R group decomposition parameters"""
DecompositionParams = rgd.RGroupDecompositionParameters()
DecompositionParams.alignment = OptionsInfo["DecompositionParams"][
"RGroupCoreAlignment"]
DecompositionParams.chunkSize = OptionsInfo["DecompositionParams"][
"chunkSize"]
DecompositionParams.matchingStrategy = OptionsInfo["DecompositionParams"][
"RGroupMatching"]
DecompositionParams.onlyMatchAtRGroups = OptionsInfo[
"DecompositionParams"]["matchOnlyAtRGroups"]
DecompositionParams.removeAllHydrogenRGroups = OptionsInfo[
"DecompositionParams"]["removeHydrogenOnlyGroups"]
DecompositionParams.removeHydrogensPostMatch = OptionsInfo[
"DecompositionParams"]["removeHydrogensPostMatch"]
return DecompositionParams
N - number of rgroups
r2 - regression r squared
descriptors - set of the descriptors for molecules in the training set
used to not enumerate existing molecules
"""
def __init__(self, rgroups, rgroup_to_descriptor_idx, fitter, r2,
descriptors):
self.rgroups = rgroups # dictionary 'Core':[core1, core1], 'R1': [rgroup1, rgroup2], ...
self.rgroup_to_descriptor_idx = rgroup_to_descriptor_idx # dictionary {smi:descriptor_idx}
self.fitter = fitter # fitter rgroup indices -> prediction
self.N = len(rgroup_to_descriptor_idx)
self.r2 = r2
self.descriptors = set([tuple(d) for d in descriptors])
default_decomp_params = rgd.RGroupDecompositionParameters()
# The default decomposition uses the GeneticAlgorirthm
# fingerprint analysis to break symmetry and make
# more consistent rgroup decompositions
default_decomp_params.matchingStrategy = rgd.GA
# Also the decomposition is allowed to add new rgroups
default_decomp_params.onlyMatchAtRGroups = False
# use the fingerprint variance method for scoring
default_decomp_params.scoreMethod = rgd.RGroupScore.FingerprintVariance
def FWDecompose(
scaffolds,
def fuzzy_scaffolding(
mols: List[Chem.rdchem.Mol],
enforce_subs: List[str] = None,
n_atom_cuttoff: int = 8,
additional_templates: List[Chem.rdchem.Mol] = None,
ignore_non_ring: bool = False,
mcs_params: Dict[Any, Any] = None,
):
"""Generate fuzzy scaffold with enforceable group that needs to appear
in the core, forcing to keep the full side chain if required.
NOTE(hadim): consider parallelize this (if possible).
Args:
mols: List of all molecules
enforce_subs: List of substructure to enforce on the scaffold.
n_atom_cuttoff: Minimum number of atom a core should have.
additional_templates: Additional template to use to generate scaffolds.
ignore_non_ring: Whether to ignore atom no in murcko ring system, even if they are in the framework.
mcs_params: Arguments of MCS algorithm.
Returns:
scaffolds: set
All found scaffolds in the molecules as valid smiles
scaffold_infos: dict of dict
Infos on the scaffold mapping, ignoring any side chain that had to be enforced.
Key corresponds to generic scaffold smiles
Values at ['smarts'] corresponds to smarts representation of the true scaffold (from MCS)
Values at ['mols'] corresponds to list of molecules matching the scaffold
scaffold_to_group: dict of list
Map between each generic scaffold and the R-groups decomposition row
"""
if enforce_subs is None:
enforce_subs = []
if additional_templates is None:
additional_templates = []
if mcs_params is None:
mcs_params = {}
rg_params = rdRGroupDecomposition.RGroupDecompositionParameters()
rg_params.removeAllHydrogenRGroups = True
rg_params.removeHydrogensPostMatch = True
rg_params.alignment = rdRGroupDecomposition.RGroupCoreAlignment.MCS
rg_params.matchingStrategy = rdRGroupDecomposition.RGroupMatching.Exhaustive
rg_params.rgroupLabelling = rdRGroupDecomposition.RGroupLabelling.AtomMap
rg_params.labels = rdRGroupDecomposition.RGroupLabels.AtomIndexLabels
core_query_param = AdjustQueryParameters()
core_query_param.makeDummiesQueries = True
core_query_param.adjustDegree = False
core_query_param.makeBondsGeneric = True
# group molecules by they generic Murcko scaffold, allowing
# side chain that contains cycle (might be a bad idea)
scf2infos = collections.defaultdict(dict)
scf2groups = {}
all_scaffolds = set([])
for m in mols:
generic_m = MurckoScaffold.MakeScaffoldGeneric(m)
scf = MurckoScaffold.GetScaffoldForMol(m)
try:
scf = MurckoScaffold.MakeScaffoldGeneric(scf)
except:
pass
if ignore_non_ring:
rw_scf = Chem.RWMol(scf)
atms = [a.GetIdx() for a in rw_scf.GetAtoms() if not a.IsInRing()]
atms.sort(reverse=True)
for a in atms:
rw_scf.RemoveAtom(a)
scfs = list(rdmolops.GetMolFrags(rw_scf, asMols=False))
else:
scfs = [dm.to_smiles(scf)]
# add templates mols if exists:
for tmp in additional_templates:
tmp = dm.to_mol(tmp)
tmp_scf = MurckoScaffold.MakeScaffoldGeneric(tmp)
if generic_m.HasSubstructMatch(tmp_scf):
scfs.append(dm.to_smiles(tmp_scf))
for scf in scfs:
if scf2infos[scf].get("mols"):
scf2infos[scf]["mols"].append(m)
else:
scf2infos[scf]["mols"] = [m]
for scf in scf2infos:
# cheat by adding murcko as last mol always
popout = False
mols = scf2infos[scf]["mols"]
if len(mols) < 2:
mols = mols + [MurckoScaffold.GetScaffoldForMol(mols[0])]
popout = True
# compute the MCS of the cluster
mcs = rdFMCS.FindMCS(
mols,
atomCompare=rdFMCS.AtomCompare.CompareAny,
bondCompare=rdFMCS.BondCompare.CompareAny,
completeRingsOnly=True,
**mcs_params,
)
mcsM = Chem.MolFromSmarts(mcs.smartsString)
mcsM.UpdatePropertyCache(False)
Chem.SetHybridization(mcsM)
if mcsM.GetNumAtoms() < n_atom_cuttoff:
continue
scf2infos[scf]["smarts"] = dm.to_smarts(mcsM)
if popout:
mols = mols[:-1]
core_groups = []
# generate rgroups based on the mcs core
success_mols = []
try:
rg = rdRGroupDecomposition.RGroupDecomposition(mcsM, rg_params)
for i, analog in enumerate(mols):
analog.RemoveAllConformers()
res = rg.Add(analog)
if not (res < 0):
success_mols.append(i)
rg.Process()
core_groups = rg.GetRGroupsAsRows()
except Exception:
pass
mols = [mols[i] for i in success_mols]
scf2groups[scf] = core_groups
for mol, gp in zip(mols, core_groups):
core = gp["Core"]
acceptable_groups = [
a.GetAtomMapNum() for a in core.GetAtoms()
if (a.GetAtomMapNum() and not a.IsInRing())
]
rgroups = [
gp[f"R{k}"] for k in acceptable_groups if f"R{k}" in gp.keys()
]
if enforce_subs:
rgroups = [
rgp for rgp in rgroups if not any([
len(rgp.GetSubstructMatch(frag)) > 0
for frag in enforce_subs
])
]
try:
scaff = trim_side_chain(
mol, AdjustQueryProperties(core, core_query_param),
rgroups)
except:
continue
all_scaffolds.add(dm.to_smiles(scaff))
return all_scaffolds, scf2infos, scf2groups