def toStringRep(v):
"""Generate a string representation of a molvector
:param v: molvector
"""
p = GetPeriodicTable()
chunksize = atom_size + bond_chunk_size
nchunks = len(v) // chunksize
m = RWMol()
out = []
for i in range(nchunks):
start = i * (atom_size + bond_chunk_size)
el, c, h, b1, o1, b2, o2, b3, o3, b4, o4 = v[start:start + chunksize]
el = ("%2s" % p.GetElementSymbol(el)).replace(" ", "_")
out.append(el)
assert c < 10
out.append(str(c))
assert h < 10
out.append(str(h))
for btype, o in ((b1, o1), (b2, o2), (b3, o3), (b4, o4)):
out.append(bond_symbols[btype])
out.append("%04d" % o)
return "".join(out)
help='Path to molecules to use as seeds')
# Parse the arguments
args = arg_parser.parse_args()
run_params = args.__dict__
# Get the list of elements
# We want those where SMILES supports implicit valences
mpnn_dir = os.path.join('notebooks', 'mpnn-training')
with open(os.path.join(mpnn_dir, 'atom_types.json')) as fp:
atom_types = json.load(fp)
with open(os.path.join(mpnn_dir, 'bond_types.json')) as fp:
bond_types = json.load(fp)
pt = GetPeriodicTable()
if len(args.elements) == 0:
elements = [pt.GetElementSymbol(i) for i in atom_types]
else:
elements = args.elements
elements = [e for e in elements if MolFromSmiles(e) is not None]
logger.info(f'Using {len(elements)} elements: {elements}')
# Prepare the one-shot model. We the molecules to compare against and the comparison model
with open(os.path.join('seed-molecules', 'top_100_pIC50.json')) as fp:
comparison_mols = [convert_smiles_to_nx(s) for s in json.load(fp)]
oneshot_dir = 'similarity'
oneshot_model = load_model(os.path.join(oneshot_dir, 'oneshot_model.h5'),
custom_objects=custom_objects)
with open(os.path.join(oneshot_dir, 'atom_types.json')) as fp:
os_atom_types = json.load(fp)
with open(os.path.join(oneshot_dir, 'bond_types.json')) as fp:
os_bond_types = json.load(fp)