def check_drigoni(download_path):
def conv(smile):
return Chem.MolToSmiles(Chem.MolFromSmiles(smile))
print('reading data...')
raw_data = train_valid_split(download_path)
data = [conv(data_item['smiles']) for data_item in raw_data['valid']]
data.extend([conv(data_item['smiles']) for data_item in raw_data['train']])
print('reading dataset')
dataset = readStr_qm9()
print('len dataset:' + str(len(dataset)) + " len data: " + str(len(data)))
count = 0
saw = 0
for i, data_item in enumerate(data):
if data_item not in dataset:
count +=1
saw += 1
print('end: ' + str(count) + " saw = " + str(saw))
count = 0
saw = 0
for i, data_item in enumerate(dataset):
if data_item not in data:
count +=1
saw += 1
print('end: ' + str(count) + " saw = " + str(saw))
def createSubSetQM9(n=5000):
from read_dataset import readStr_qm9
D = readStr_qm9()
np.random.shuffle(D)
subSet = D[:n]
f = open("qm9_sub" + str(n) + ".smi", "w")
for l in subSet:
f.write(l)
f.write("\n")
f.close()
def main():
decoded_file = GRAMMAR_WEIGHTS.split(".")[0] + "_decRes.txt"
priors_file = GRAMMAR_WEIGHTS.split(".")[0] + "_priorsRes.txt"
generation_file = GRAMMAR_WEIGHTS.split(".")[0] + "_generationRes.txt"
grammar_model = molecule_vae.Qm9GrammarModel(GRAMMAR_WEIGHTS)
XTE = readStr_qm9()
XTE = XTE[0:5000]
# rember to comment and uncomment the line in the #moelcule_vae file
decoded_result = reconstruction(grammar_model, XTE)
save_decoded_results(XTE, decoded_result, decoded_file)
# decoded_priors = prior(grammar_model)
# save_decoded_priors(decoded_priors, priors_file)
decoded_generation = generation(grammar_model)
save_decoded_priors(decoded_generation, generation_file)
def main():
from att_model_proxy import AttMolProxy as ProxyModel
from att_model_proxy import cmd_args
# takes the model and calculate the decode results
model = ProxyModel()
# update where to save
decoded_file = cmd_args.save_dir + '/decoded_results.txt'
# reading smiles test set
if cmd_args.smiles_file == 'qm9':
smiles_list = readStr_qm9()
elif cmd_args.smiles_file == 'zinc':
smiles_list = read_zinc()
XTE = smiles_list[0:nb_smiles]
decoded_result = reconstruct(model, XTE)
decoded_result = np.array(decoded_result)
save_decoded_results(XTE, decoded_result, decoded_file)
def train_valid_split_drigoni():
print('reading dataset')
dataset = readStr_qm9()[5000:]
valid_idx = np.random.randint(0, high=len(dataset), size=round(len(dataset)*0.1))
raw_data = {'train': [], 'valid': []} # save the train, valid dataset.
file_count=0
for i, smiles in enumerate(dataset):
val = QED.qed(Chem.MolFromSmiles(smiles))
if i not in valid_idx:
raw_data['train'].append({'smiles': smiles, 'QED': val})
else:
raw_data['valid'].append({'smiles': smiles, 'QED': val})
file_count += 1
if file_count % 2000 == 0:
print('finished reading: %d' % file_count, end='\r')
return raw_data
def qm9StrProve():
L = readStr_qm9()
MAX = 120
count = 0
countDot = 0
countAst = 0
nMolMIn9 = 0
nMaxAtoms = 0
nMinAtom = 100
nMol1 = 0
for s in L:
if len(s) > MAX:
count = count + 1
if "." in s:
countDot = countDot + 1
if "*" in s:
countAst = countAst + 1
m = Chem.MolFromSmiles(s)
atom = m.GetNumAtoms()
if atom <= 9:
nMolMIn9 = nMolMIn9 + 1
if atom > nMaxAtoms:
nMaxAtoms = atom
elif atom < nMinAtom:
nMinAtom = atom
if atom == 1:
nMol1 = nMol1 + 1
print("Numero molecole con num. atomi <= 9: {}".format(nMolMIn9))
print("Numero massimo di atomi: {}".format(nMaxAtoms))
print("Numero minimo di atomi: {}".format(nMinAtom))
print("Numero molecole con caratteri superiori a 120: {}".format(count))
print("Numero molecole con carattere '.' : {}".format(countDot))
print("Numero molecole con carattere '*' : {}".format(countAst))
print("Numero di molecole lette: {}".format(len(L)))
print(
"Numero di molecole lette formate da un solo atomo: {}".format(nMol1))
print("-------------- FINE ---------------")
all_onehot = np.zeros((len(L), cmd_args.max_decode_steps, DECISION_DIM),
dtype=np.byte)
all_masks = np.zeros((len(L), cmd_args.max_decode_steps, DECISION_DIM),
dtype=np.byte)
for start, b_pair in zip(range(0, len(L), chunk_size), list_binary):
all_onehot[start:start + chunk_size, :, :] = b_pair[0]
all_masks[start:start + chunk_size, :, :] = b_pair[1]
#f_smiles = '.'.join(cmd_args.smiles_file.split('/')[-1].split('.')[0:-1])
f_smiles = cmd_args.smiles_file
out_file = '%s/%s-%d.h5' % (cmd_args.save_dir, f_smiles,
cmd_args.skip_deter)
h5f = h5py.File(out_file, 'w')
h5f.create_dataset('x', data=all_onehot)
h5f.create_dataset('masks', data=all_masks)
h5f.close()
if __name__ == '__main__':
smiles_list = []
if cmd_args.smiles_file == 'qm9':
smiles_list = readStr_qm9()
elif cmd_args.smiles_file == 'zinc':
smiles_list = read_zinc()
train_dataset = smiles_list[5000:]
run_job(train_dataset)
from plot_utils import *
from read_dataset import readStr_qm9, read_zinc
from smile_metrics import MolecularMetrics as mm
from utils import save_scores_bias, load_decoded_results, calc_perc
folder = "bias/"
# take params
name = sys.argv[1]
file = sys.argv[2]
dataset = sys.argv[3]
if dataset == "zinc":
trainingSet = read_zinc()
else:
trainingSet = readStr_qm9()
trainingSet = trainingSet[5000:]
# make folder
try:
os.makedirs(folder + name)
except OSError:
print("Creation of the directory %s failed" % (folder + name))
else:
print("Successfully created the directory %s " % (folder + name))
# READ SMILES
smi = dict()
smi['smiles'], smi['decoded'] = load_decoded_results(file)
smi['valid'] = []
for line in smi['decoded']:
from __future__ import print_function
import nltk
import qm9_grammar
import numpy as np
import h5py
import molecule_vae
import sys, os
sys.path.append('%s/../_utils' % os.path.dirname(os.path.realpath(__file__)))
from read_dataset import readStr_qm9
MAX_LEN = 277
L = readStr_qm9()
NCHARS = len(qm9_grammar.GCFG.productions())
def to_one_hot(smiles):
""" Encode a list of smiles strings to one-hot vectors """
assert type(smiles) == list
prod_map = {}
for ix, prod in enumerate(qm9_grammar.GCFG.productions()):
prod_map[prod] = ix
tokenize = molecule_vae.get_zinc_tokenizer(qm9_grammar.GCFG)
tokens = map(tokenize, smiles)
parser = nltk.ChartParser(qm9_grammar.GCFG)
parse_trees = [parser.parse(t).next() for t in tokens]
productions_seq = [tree.productions() for tree in parse_trees]
indices = [
np.array([prod_map[prod] for prod in entry], dtype=int)
for entry in productions_seq
result_list = Parallel(n_jobs=-1)(delayed(parse_many)(chunk[i:i +
size], grammar)
for i in range(0, len(chunk), size))
return [_1 for _0 in result_list for _1 in _0]
import cPickle as cp
from tqdm import tqdm
if __name__ == '__main__':
save_dir = cmd_args.save_dir
fname = save_dir + (
cmd_args.smiles_file.split('/')[-1]).split('.')[0] + '.cfg_dump'
fout = open(fname, 'wb')
grammar = parser.Grammar(cmd_args.grammar_file)
smiles = []
if cmd_args.smiles_file == 'qm9':
smiles = readStr_qm9()
elif cmd_args.smiles_file == 'zinc':
smiles = read_zinc()
for i in tqdm(range(len(smiles))):
ts = parser.parse(smiles[i], grammar)
assert isinstance(ts, list) and len(ts) == 1
n = AnnotatedTree2MolTree(ts[0])
cp.dump(n, fout, cp.HIGHEST_PROTOCOL)
fout.close()
def main():
torch.manual_seed(0)
lg = rdkit.RDLogger.logger()
lg.setLevel(rdkit.RDLogger.CRITICAL)
parser = OptionParser()
parser.add_option("-t", "--test", dest="test_path")
parser.add_option("-v", "--vocab", dest="vocab_path")
parser.add_option("-m", "--model", dest="model_path")
parser.add_option("-w", "--hidden", dest="hidden_size", default=200)
parser.add_option("-l", "--latent", dest="latent_size", default=56)
parser.add_option("-d", "--depth", dest="depth", default=3)
opts, args = parser.parse_args()
vocab = [x.strip("\r\n ") for x in open(opts.vocab_path)]
vocab = Vocab(vocab)
hidden_size = int(opts.hidden_size)
latent_size = int(opts.latent_size)
depth = int(opts.depth)
model = JTNNVAE(vocab, hidden_size, latent_size, depth)
model.load_state_dict(torch.load(opts.model_path))
model = model.cuda()
dataset_name = opts.test_path
result_file = dataset_name + "_decoded_results.txt"
priors_file = dataset_name + "_decoded_priors.txt"
generation_fie = dataset_name + "_generation.txt"
# read dataset
if dataset_name == "zinc":
XTE = read_zinc()
else:
D = readStr_qm9()
# fix problem about molecule with '.' inside
XTE = []
for mol in D:
if "." not in mol:
XTE.append(mol)
# reconstruction
XTE = XTE[0:5000]
XTE = filter(lambda x: len(x) > 1,
XTE) #needed for removing smiles with only a char.
decoded_result = reconstruction(model, XTE, 20, 1)
save_decoded_results(XTE, decoded_result, result_file)
# prior
# decoded_priors_witherrors = model.sample_prior_eval(True, 1000, 10)
# decoded_priors = []
# for i in decoded_priors_witherrors:
# decoded_priors.append(sanitize(i))
# save_decoded_priors(decoded_priors, priors_file)
# generation
generation_witherrors = model.sample_prior_eval(True, 20000, 1)
generation = []
for i in generation_witherrors:
generation.append(sanitize(i))
save_decoded_priors(generation, generation_fie)