def get_molecules_from_pdb(protein_pdb, ligand_pdb):
"""
Input:
protein_pdb: str. Path to protein .pdb file
ligand_pdb: str. Path to ligand .pdb file
Output:
protein: rdkit.Chem.rdchem.Mol
ligand: rdkit.Chem.rdchem.Mol
compl: rdkit.Chem.rdchem.Mol
"""
# TODO Currently combining protein and ligand with
# mdtraj. Verify this method.
complex_traj = combine_mdtraj(md.load(protein_pdb), md.load(ligand_pdb))
tmpfile = tempfile.mkstemp(suffix='.pdb')
f, comp_pdb = tmpfile
complex_traj.save(comp_pdb)
protein = rdmolfiles.MolFromPDBFile(protein_pdb)
ligand = rdmolfiles.MolFromPDBFile(ligand_pdb)
compl = rdmolfiles.MolFromPDBFile(comp_pdb)
os.close(f)
os.remove(comp_pdb)
return (protein, ligand, compl)
def pdb2graph(pdbid, data_dir='./data/pdbbind/v2018'):
"""
Input:
pdbid: str. protein code from PDBBind
Returns:
tuple of tuples. Graph representation of nodes
"""
protein_pdb_file = os.path.join(data_dir, pdbid,
"{}_protein.pdb".format(pdbid))
ligand_pdb_file = os.path.join(data_dir, pdbid,
"{}_ligand.pdb".format(pdbid))
if not os.path.exists(protein_pdb_file) or \
not os.path.exists(ligand_pdb_file):
raise IOError(".pdb file not found in {}".format(
os.path.join(data_dir, pdbid)))
# combining protein pdb file and ligand pdb file to one pdb file
protein_traj = md.load(protein_pdb_file)
ligand_traj = md.load(ligand_pdb_file)
complex_traj = combine_mdtraj(md.load(protein_pdb_file),
md.load(ligand_pdb_file))
tempdir = tempfile.mkdtemp()
complex_traj.save(os.path.join(tempdir, 'complex.pdb'))
protein = rdmolfiles.MolFromPDBFile(protein_pdb_file)
ligand = rdmolfiles.MolFromPDBFile(ligand_pdb_file)
compl = AllChem.MolFromPDBFile(os.path.join(tempdir, 'complex.pdb'))
return (build_graph_from_molecule(protein),
build_graph_from_molecule(ligand),
build_graph_from_molecule(compl))
def pdb2graph(pdbid, data_dir='./data/pdbbind/v2018'):
"""
pdbid: str. protein code from PDBBind
graph_type: str. One of ['protein', 'ligand', ]
"""
protein_pdb_file = os.path.join(
data_dir, pdbid, "{}_protein.pdb".format(pdbid))
ligand_pdb_file = os.path.join(
data_dir, pdbid, "{}_ligand.pdb".format(pdbid))
if not os.path.exists(protein_pdb_file) or \
not os.path.exists(ligand_pdb_file):
raise IOError(".pdb file not found in {}".format(
os.path.join(data_dir, pdbid)))
# combining protein pdb file and ligand pdb file to one pdb file
protein_traj = md.load(protein_pdb_file)
ligand_traj = md.load(ligand_pdb_file)
complex_traj = combine_mdtraj(protein_traj, ligand_traj)
tempdir = tempfile.mkdtemp()
complex_traj.save(os.path.join(tempdir, 'complex.pdb'))
protein = rdmolfiles.MolFromPDBFile(protein_pdb_file)
ligand = rdmolfiles.MolFromPDBFile(ligand_pdb_file)
compl = rdmolfiles.MolFromPDBFile(os.path.join(tempdir, 'complex.pdb'))
return (build_graph_from_molecule(protein),
build_graph_from_molecule(ligand),
build_graph_from_molecule(compl))
def read_lig_pdb(lig_name): mol_path = "../ligands/"+lig_name+".pdb" mol = rdmolfiles.MolFromPDBFile( mol_path, sanitize=False ) return mol
def print_fragments_from_molecules(filename):
try:
mol = rdmolfiles.MolFromPDBFile(filename)
for _, smi in generate_fragments(mol):
print(smi)
except:
print("Failed", filename)
def retrieveMoleculePDB(ligand_path): """ Returns RDKit molecule objects for requested path PDB file. -- args ligand_path (str): path leading to molecule pdb file -- returns RDKit molecule object """ mol = rdmolfiles.MolFromPDBFile( ligand_path, sanitize=True ) return mol
def load_dataset(self):
# Load files
x, c, y = [], [], []
if self.dataset is None:
tx, tc = [], []
for i, target_n in enumerate(next(os.walk(self.path))[1]):
try:
target = rdmolfiles.MolFromPDBFile(self.path + target_n +
'/receptor.pdb')
targetc = target.GetConformer().GetPositions()
except AttributeError:
print("Failed to load target: " + target_n)
continue
for file, value in (("actives", True), ("decoys", False)):
with open(self.path + target_n + '/' + file + '_final.ism',
'r') as f:
lines = f.readlines()
smiless = [x.strip().split(" ")[0] for x in lines]
out = Parallel(n_jobs=4)(
delayed(Dataset._load_molecule)(smile)
for smile in smiless)
out = filter(None, out)
x_, c_ = zip(*out)
x.extend(x_)
c.extend(c_)
tx.extend([target] * len(x_))
tc.extend([targetc] * len(x_))
y.extend([value] * len(x_))
self.target_size = max([t.GetNumAtoms() for t in tx])
self.molecule_size = max([m.GetNumAtoms() for m in x])
else:
for file, value in (("actives", True), ("decoys", False)):
with open(self.path + file + '_final.ism', 'r') as f:
lines = f.readlines()
smiless = [x.strip().split(" ")[0] for x in lines]
out = Parallel(n_jobs=4)(
delayed(Dataset._load_molecule)(smile)
for smile in smiless)
out = filter(None, out)
x_, c_ = zip(*out)
x.extend(x_)
c.extend(c_)
y.extend([value] * len(x_))
self.molecule_size = max([m.GetNumAtoms() for m in x])
self.mols, self.coords, self.target = np.array(x), np.array(
c), np.array(y)
pkl_dict = {
'mols': self.mols,
'coords': self.coords,
'target': self.target
}
pkl.dump(pkl_dict, open(self.path + "dataset.pkl", "wb"))
# Shuffle data
idx = np.random.permutation(len(self.mols))
self.mols, self.coords, self.target = self.mols[idx], self.coords[
idx], self.target[idx]
# Split data
spl1 = int(len(self.mols) * 0.2)
spl2 = int(len(self.mols) * 0.1)
self.x = {
"train": self.mols[spl1:],
"valid": self.mols[spl2:spl1],
"test": self.mols[:spl2]
}
self.c = {
"train": self.coords[spl1:],
"valid": self.coords[spl2:spl1],
"test": self.coords[:spl2]
}
self.y = {
"train": self.target[spl1:],
"valid": self.target[spl2:spl1],
"test": self.target[:spl2]
}
if self.dataset is None:
self.targets, self.t_coords = np.array(tx), np.array(tc)
pkl_dict = {
'mols': np.array(x),
'coords': np.array(c),
'target': np.array(y),
'targets': self.targets,
't_coords': self.t_coords
}
pkl.dump(pkl_dict, open(self.path + "dataset.pkl", "wb"))
# Shuffle data
self.targets, self.t_coords = self.targets[idx], self.t_coords[idx]
# Split data
spl1 = int(len(self.mols) * 0.2)
spl2 = int(len(self.mols) * 0.1)
self.tx = {
"train": self.targets[spl1:],
"valid": self.targets[spl2:spl1],
"test": self.targets[:spl2]
}
self.tc = {
"train": self.t_coords[spl1:],
"valid": self.t_coords[spl2:spl1],
"test": self.t_coords[:spl2]
}
def mutate(
self, pdbid: str,
replace_with: Dict[int,
Optional[str]]) -> Union[List[str], rdchem.Mol]:
"""Modify amino acid residues at the defined positions.
If the locations indexes exceed the amount in data, then they
will be ignored and a warning will be announced.
Params:
-------
pdbid: str
PDB ID associated with the structure.
replace_with: dict
The index location(s) within the full protein to replace
certain residue(s) with. If a residue associated with a
index location is None, then the modified residue is chosen
randomly. If a certain index exceeds the number of available
residues in the protein, then those enteries are simply
ignored and the user is notified.
Returns:
--------
protein: list of str or rdkit.Chem.rdchem.Mol
Modified protein with residues. If fmt="primary", then list
of string (peptide names) is returned. If fmt="tertiary",
then 3D molecule structure is returned.
"""
# Load PDB structure (download, if necessary)
pdb_dir = maybe_create_dir(os.path.join(self.rootdir, pdbid))
pdb_file = os.path.join(pdb_dir, f"{pdbid}.pdb")
if not os.path.exists(pdb_file):
is_successful = cmd.fetch(pdbid,
name=pdbid,
state=1,
type="pdb",
path=pdb_dir)
if is_successful == -1:
raise DownloadError(f"Unable to download '{pdbid}'.")
else:
cmd.load(pdb_file, object=pdbid, state=1, format="pdb")
# Get all residue names, see: https://pymolwiki.org/index.php/List_Selection
resnames_dict = {"names": []}
cmd.iterate("(name ca)", "names.append(resn)", space=resnames_dict)
residue_names = resnames_dict["names"]
num_residues = len(residue_names)
# Cleanup idxs: remove indicies that exceed number of available residues
nonvalid_idxs = [
idx for idx in replace_with.keys() if idx > num_residues
]
for idx in nonvalid_idxs:
print(
f"OutOfRange: Removing idx {idx} (only {num_residues} residues)."
)
replace_with.pop(idx)
# Randomly choose an amino acid (AA) to replace residue, if None is provided.
# Additionally, format string such that it is a valid 3 letter amino acid.
for idx, residue in replace_with.items():
if residue is None:
replace_with[idx] = np.random.choice(aa3)
elif is_aa(residue):
residue = residue.upper()
if len(residue) == 1:
replace_with[idx] = one_to_three.get(residue)
elif len(residue) == 3:
replace_with[idx] = residue
else:
raise ValueError(
f"Invalid residue '{residue}'. Choose one from "
f"the following {aa1+aa3}.")
# Determine save filepath name
modified_res_str = ":".join(
[f"{k}{three_to_one.get(v)}" for k, v in replace_with.items()])
filename = f"{self.fmt}_{modified_res_str}"
filename += ".pdb" if self.fmt == "tertiary" else ".json"
save_filepath = os.path.join(self.rootdir, pdbid, filename)
# Replace primary structure, i.e. residue names (str)
if self.fmt == "primary":
# Load data from cache, if it exists
protein = None
if os.path.exists(save_filepath):
with open(save_filepath) as json_file:
protein = json.load(json_file)
if protein is None:
for idx, residue in replace_with.items():
residue_names[
idx - 1] = residue # since PDB starts with idx of 1
protein = [three_to_one.get(name) for name in residue_names]
# Save sequence temporarily
_ = maybe_create_dir(save_filepath)
with open(save_filepath, "w") as outfile:
json.dump(protein, outfile)
# Replace tertiary structure, i.e. residue's 3D coordinates
elif self.fmt == "tertiary":
if not os.path.exists(save_filepath):
# Split states so that we can optimize only on specific state(s).
# NOTE: Might be useful to choose lowest energy state to mutate,
# OR mutate rotamers for all positions, then choose one with
# lowest energy.
cmd.split_states(object=pdbid)
# Delete all other objects other than one we want to mutate
# NOTE: For now, keep only first object. This might change
# depending on which state needs to be kept.
objs = cmd.get_object_list() # aka states
keep_objs = [pdbid + "_0001"]
for obj in objs:
if obj not in keep_objs:
cmd.delete(obj)
assert keep_objs == cmd.get_object_list()
# Mutate residues
cmd.wizard("mutagenesis")
wizard: Mutagenesis = cmd.get_wizard()
for idx, res in replace_with.items():
selection = "{0:s}//A/{1:d}/".format(keep_objs[0], idx)
wizard.do_select(
selection) # select which residue index to replace
wizard.set_mode(
res) # choose name of residue to replace with
wizard.do_state(
1
) # select rotamer with least strain (aka conflicts w/ other atoms)
wizard.apply() # apply point mutation
cmd.set_wizard(None) # close wizard
# Save PDB temporarily
_ = maybe_create_dir(save_filepath)
cmd.save(save_filepath, selection=pdbid, format="pdb")
cmd.delete("all") # remove all objects, clears workspace
# Load + choose model/structure with lowest energy
# NOTE: If sanitize=True, the function checks if Mol has the correct
# hybridization/valance structure (aka is it chemically reasonable).
# When converting from the PDB block, this sometimes results in
# improper parsing. Instead, for now, we just check if the Mol is
# syntactically valid (i.e. all rings/branches closed, no illegal
# atom types, etc).
protein = rdmolfiles.MolFromPDBFile(save_filepath,
sanitize=False,
removeHs=False)
if protein.GetNumConformers() > 1:
protein = _get_conformer(protein, conformer="min", algo="MMFF")
else:
raise NotImplementedError
# Remove file, if not needed
if not self.cache:
os.remove(save_filepath)
return protein