def fcd_statistics(SMILES, n_jobs, gpu, out, canonicalize=False):
"""
Pre-compute FCD statistics for a list of SMILES.
:param SMILES: List SMILES
:param n_jobs: Number of jobs for processing SMILES
:param out: Path to output file
:return:
"""
fcd = FCD(device=f'cuda:{gpu}', n_jobs=n_jobs, canonize=False)
if canonicalize:
can_SMILES = []
for smi in SMILES:
if Chem.MolFromSmiles(smi):
can_SMILES.append(Chem.MolToSmiles(Chem.MolFromSmiles(smi)))
results = fcd.precalc(can_SMILES)
else:
results = fcd.precalc(SMILES)
if not os.path.exists(os.path.dirname(out)):
os.makedirs(os.path.dirname(out))
with open(out, 'wb') as f:
pkl.dump(results, f)
def test_zero_molecule(self):
fcd = FCD()
output_pytorch = fcd([], [])
self.assertNotEqual(
output_pytorch,
output_pytorch,
msg=("FCD should return np.nan on invalid situations"))
def test_multiprocess(self):
fcd = FCD(n_jobs=2)
output_pytorch = fcd(self.set1, self.set2)
diff = abs(self.output_keras - output_pytorch)
self.assertAlmostEqual(
output_pytorch,
self.output_keras,
places=4,
msg=("Outputs differ. keras={},".format(self.output_keras) +
"torch={}. diff is {}".format(output_pytorch, diff)))
res = res.append(
pd.DataFrame({
'input_file':
sampled_file,
'outcome': [
'KL divergence, hydrogen donors',
'KL divergence, hydrogen acceptors',
'Jensen-Shannon distance, hydrogen donors',
'Jensen-Shannon distance, hydrogen acceptors',
'Wasserstein distance, hydrogen donors',
'Wasserstein distance, hydrogen acceptors'
],
'value': [
kl_donors, kl_acceptors, jsd_donors, jsd_acceptors,
emd_donors, emd_acceptors
]
}))
## outcome 20: Frechet ChemNet distance
fcd = FCD(canonize=False)
fcd_calc = fcd(gen_canonical, org_canonical)
res = res.append(
pd.DataFrame({
'input_file': sampled_file,
'outcome': 'Frechet ChemNet distance',
'value': [fcd_calc]
}))
# write output
res.to_csv(output_file, index=False, compression='gzip')