def Net(args):
latent_descr = 50 * [('c', 20)]
feature_descr = [('c', 20)]
latent_size = 50
latent_input_size = 50
enc = gentrl.RNNEncoder(latent_size)
dec = gentrl.DilConvDecoder(latent_input_size, args)
model = gentrl.DIS_GENTRL(enc,
dec,
latent_descr,
feature_descr,
beta=0.001)
return model
from moses.metrics.utils import get_n_rings, get_mol
def get_num_rings_6(mol):
r = mol.GetRingInfo()
return len([x for x in r.AtomRings() if len(x) > 6])
def penalized_logP(mol_or_smiles, masked=False, default=-5):
mol = get_mol(mol_or_smiles)
if mol is None:
return default
reward = logP(mol) - SA(mol) - get_num_rings_6(mol)
if masked and not mol_passes_filters(mol):
return default
return reward
df['plogP'] = df['SMILES'].apply(penalized_logP)
df.to_csv('train_plogp_plogpm.csv', index=None)
enc = gentrl.RNNEncoder(latent_size=50)
dec = gentrl.DilConvDecoder(latent_input_size=50)
model = gentrl.GENTRL(enc, dec, 50 * [('c', 20)], [('c', 20)], beta=0.001)
#model.cuda();
md = gentrl.MolecularDataset(sources=[
{'path':'train_plogp_plogpm.csv',
'smiles': 'SMILES',
'prob': 1,
'plogP' : 'plogP',
}],
props=['plogP'])
from torch.utils.data import DataLoader
train_loader = DataLoader(md, batch_size=50, shuffle=True, num_workers=1, drop_last=True)
model.train_as_vaelp(train_loader, lr=1e-4)
! mkdir -p saved_gentrl
model.save('./saved_gentrl/')
#train-RL https://github.com/insilicomedicine/GENTRL/blob/master/examples/train_rl.ipynb
import torch
from moses.metrics.utils import get_mol
import pandas as pd
import pickle
import moses
from moses.utils import CharVocab
from rdkit import RDLogger
RDLogger.DisableLog('rdApp.*')
# Load vocab
dataset_path = "../data/moses_qed_props.csv.gz"
df = pd.read_csv(dataset_path, compression="gzip")
vocab = CharVocab.from_data(df['SMILES'])
enc = gentrl.RNNEncoder(vocab, latent_size=50)
dec = gentrl.DilConvDecoder(vocab, latent_input_size=50, split_len=100)
model = gentrl.GENTRL(enc, dec, 50 * [('c', 20)], [('c', 20)], beta=0.001)
model.cuda()
torch.cuda.set_device(0)
moses_qed_props_model_path = "../models/moses/"
model.load(moses_qed_props_model_path)
model.cuda()
import random
from rdkit import RDLogger
RDLogger.DisableLog('rdApp.*')
generated = []
verbose_lim = 10000