Generative model is very hot not only in computer vision, natural language processing but also chemoinformatics.
As you know, recent version of deep learning based compound generator works very well but it is required huge computer resources for building the model. And also SMILES based approach sometime generates invalid molecules.
Recently I read very interesting article reported Alan Aspuru-Guzik who is the pioneer of chemical VAE. You can read the article from chemRxiv.
Beyond Generative Models: Superfast Traversal, Optimization, Novelty,
Exploration and Discovery (STONED) Algorithm for Molecules using
SELFIES
As the name describes that STONED uses SELFIES which is new compound representation method which is developed by their group. SELFIES can be install via pip. Just type $pip install selfies.
SELFIES means Self-Referencing Embedded Strings.
So it is a good tool for compound representation.
There are three key methods in STONED, 1) reorder string (randomize SMILES), 2) convert to SELFIES 3) perform random mutations which are addition, deletion and replacement. By using this mutation method, deep learning is not required for compound generation.
Rediscovery with Genetic Algorithm is the one of interesting topic for me.
Rediscovery means that generates target molecule from randomized strings. Fortunately all code is disclosed on github.
https://github.com/aspuru-guzik-group/stoned-selfies
So it is easy to run STONED on your PC.
$ git clone https://github.com/aspuru-guzik-group/stoned-selfies.git
$ cd stoned-selfies
STONED requires selfies, rdkit and numpy. If these packages are available on your PC, STONED will run without error.
I modified GA_rediscover.py to more general. Put out the configuration as yml. My code (modified version) is below.
flex_GA_rediscover.py
"""
Created on Sat May 23 18:17:31 2020
celebx = 'CC1=CC=C(C=C1)C2=CC(=NN2C3=CC=C(C=C3)S(=O)(=O)N)C(F)(F)F'
tiotixene = 'CN1CCN(CC1)CCC=C2C3=CC=CC=C3SC4=C2C=C(C=C4)S(=O)(=O)N(C)C'
Troglitazone = 'CC1=C(C2=C(CCC(O2)(C)COC3=CC=C(C=C3)CC4C(=O)NC(=O)S4)C(=C1O)C)C'
@author: akshat
"""
import selfies
import numpy as np
import random
from rdkit.Chem import MolFromSmiles as smi2mol
from rdkit.Chem import MolToSmiles as mol2smi
from rdkit import Chem
from rdkit.Chem import AllChem
from rdkit.DataStructs.cDataStructs import TanimotoSimilarity
from selfies import encoder, decoder
from rdkit import RDLogger
import yaml
RDLogger.DisableLog('rdApp.*')
def get_ECFP4(mol):
''' Return rdkit ECFP4 fingerprint object for mol
Parameters:
mol (rdkit.Chem.rdchem.Mol) : RdKit mol object
Returns:
rdkit ECFP4 fingerprint object for mol
'''
return AllChem.GetMorganFingerprint(mol, 2)
def sanitize_smiles(smi):
'''Return a canonical smile representation of smi
Parameters:
smi (string) : smile string to be canonicalized
Returns:
mol (rdkit.Chem.rdchem.Mol) : RdKit mol object (None if invalid smile string smi)
smi_canon (string) : Canonicalized smile representation of smi (None if invalid smile string smi)
conversion_successful (bool): True/False to indicate if conversion was successful
'''
try:
mol = smi2mol(smi, sanitize=True)
smi_canon = mol2smi(mol, isomericSmiles=False, canonical=True)
return (mol, smi_canon, True)
except:
return (None, None, False)
def mutate_selfie(selfie, max_molecules_len, write_fail_cases=False):
'''Return a mutated selfie string (only one mutation on slefie is performed)
Mutations are done until a valid molecule is obtained
Rules of mutation: With a 50% propbabily, either:
1. Add a random SELFIE character in the string
2. Replace a random SELFIE character with another
Parameters:
selfie (string) : SELFIE string to be mutated
max_molecules_len (int) : Mutations of SELFIE string are allowed up to this length
write_fail_cases (bool) : If true, failed mutations are recorded in "selfie_failure_cases.txt"
Returns:
selfie_mutated (string) : Mutated SELFIE string
smiles_canon (string) : canonical smile of mutated SELFIE string
'''
valid=False
fail_counter = 0
chars_selfie = get_selfie_chars(selfie)
while not valid:
fail_counter += 1
alphabet = list(selfies.get_semantic_robust_alphabet()) # 34 SELFIE characters
choice_ls = [1, 2] # 1=Insert; 2=Replace; 3=Delete
random_choice = np.random.choice(choice_ls, 1)[0]
# Insert a character in a Random Location
if random_choice == 1:
random_index = np.random.randint(len(chars_selfie)+1)
random_character = np.random.choice(alphabet, size=1)[0]
selfie_mutated_chars = chars_selfie[:random_index] + [random_character] + chars_selfie[random_index:]
# Replace a random character
elif random_choice == 2:
random_index = np.random.randint(len(chars_selfie))
random_character = np.random.choice(alphabet, size=1)[0]
if random_index == 0:
selfie_mutated_chars = [random_character] + chars_selfie[random_index+1:]
else:
selfie_mutated_chars = chars_selfie[:random_index] + [random_character] + chars_selfie[random_index+1:]
# Delete a random character
elif random_choice == 3:
random_index = np.random.randint(len(chars_selfie))
if random_index == 0:
selfie_mutated_chars = chars_selfie[random_index+1:]
else:
selfie_mutated_chars = chars_selfie[:random_index] + chars_selfie[random_index+1:]
else:
raise Exception('Invalid Operation trying to be performed')
selfie_mutated = "".join(x for x in selfie_mutated_chars)
sf = "".join(x for x in chars_selfie)
try:
smiles = decoder(selfie_mutated)
mol, smiles_canon, done = sanitize_smiles(smiles)
if len(selfie_mutated_chars) > max_molecules_len or smiles_canon=="":
done = False
if done:
valid = True
else:
valid = False
except:
valid=False
if fail_counter > 1 and write_fail_cases == True:
f = open("selfie_failure_cases.txt", "a+")
f.write('Tried to mutate SELFIE: '+str(sf)+' To Obtain: '+str(selfie_mutated) + '\n')
f.close()
return (selfie_mutated, smiles_canon)
def get_selfie_chars(selfie):
'''Obtain a list of all selfie characters in string selfie
Parameters:
selfie (string) : A selfie string - representing a molecule
Example:
>>> get_selfie_chars('[C][=C][C][=C][C][=C][Ring1][Branch1_1]')
['[C]', '[=C]', '[C]', '[=C]', '[C]', '[=C]', '[Ring1]', '[Branch1_1]']
Returns:
chars_selfie: list of selfie characters present in molecule selfie
'''
chars_selfie = [] # A list of all SELFIE sybols from string selfie
while selfie != '':
chars_selfie.append(selfie[selfie.find('['): selfie.find(']')+1])
selfie = selfie[selfie.find(']')+1:]
return chars_selfie
def get_reward(selfie_A_chars, selfie_B_chars):
'''Return the levenshtein similarity between the selfies characters in 'selfie_A_chars' & 'selfie_B_chars'
Parameters:
selfie_A_chars (list) : list of characters of a single SELFIES
selfie_B_chars (list) : list of characters of a single SELFIES
Returns:
reward (int): Levenshtein similarity between the two SELFIES
'''
reward = 0
iter_num = max(len(selfie_A_chars), len(selfie_B_chars)) # Larger of the selfie chars to iterate over
for i in range(iter_num):
if i+1 > len(selfie_A_chars) or i+1 > len(selfie_B_chars):
return reward
if selfie_A_chars[i] == selfie_B_chars[i]:
reward += 1
return reward
# Executable code for EXPERIMENT C (Three different choices):
# read params from yaml
cfg = yaml.load(open('ga_conf.yml', 'r'), yaml.SafeLoader)
N = cfg['params']['N'] # Number of runs
simlr_path_collect = cfg['params']['simlr_path_collect']
svg_file_name = cfg['params']['svg_file_name']
starting_mol_name = cfg['params']['starting_mol_name']
data_file_name = cfg['params']['data_file_name']
starting_smile = cfg['params']['starting_smile']
show_gen_out = cfg['params']['show_gen_out']
len_random_struct = len(get_selfie_chars(encoder(starting_smile))) # Length of the starting SELFIE structure
for i in range(N):
print('Run number: ', i)
with open(data_file_name, 'a') as myfile:
myfile.write('RUN {} \n'.format(i))
# celebx = 'CC1=CC=C(C=C1)C2=CC(=NN2C3=CC=C(C=C3)S(=O)(=O)N)C(F)(F)F'
starting_selfie = encoder(starting_smile)
starting_selfie_chars = get_selfie_chars(starting_selfie)
max_molecules_len = len(starting_selfie_chars)
# Random selfie initiation:
alphabet = list(selfies.get_semantic_robust_alphabet()) # 34 SELFIE characters
selfie = ''
for i in range(random.randint(1, len_random_struct)): # max_molecules_len = max random selfie string length
selfie = selfie + np.random.choice(alphabet, size=1)[0]
starting_selfie = [selfie]
print('Starting SELFIE: ', starting_selfie)
generation_size = 500
num_generations = 10000
save_best = []
simlr_path = []
reward_path = []
# Initial set of molecules
population = np.random.choice(starting_selfie, size=500).tolist() # All molecules are in SELFIES representation
for gen_ in range(num_generations):
# Calculate fitness for all of them
fitness = [get_reward(starting_selfie_chars, get_selfie_chars(x)) for x in population]
fitness = [float(x)/float(max_molecules_len) for x in fitness] # Between 0 and 1
# Keep the best member & mutate the rest
# Step 1: Keep the best molecule
best_idx = np.argmax(fitness)
best_selfie = population[best_idx]
# Diplay some Outputs:
if show_gen_out:
print('Generation: {}/{}'.format(gen_, num_generations))
print(' Top fitness: ', fitness[best_idx])
print(' Top SELFIE: ', best_selfie)
with open(data_file_name, 'a') as myfile:
myfile.write(' SELFIE: {} FITNESS: {} \n'.format(best_selfie, fitness[best_idx]))
# Maybe also print the tanimoto score:
mol = Chem.MolFromSmiles(decoder(best_selfie))
target = Chem.MolFromSmiles(starting_smile)
fp_mol = get_ECFP4(mol)
fp_target = get_ECFP4(target)
score = TanimotoSimilarity(fp_mol, fp_target)
simlr_path.append(score)
reward_path.append(fitness[best_idx])
save_best.append(best_selfie)
# Step 2: Get mutated selfies
new_population = []
for i in range(generation_size-1):
# selfie_mutated, _ = mutate_selfie(best_selfie, max_molecules_len, write_fail_cases=True)
selfie_mutated, _ = mutate_selfie(best_selfie, len_random_struct, write_fail_cases=True) # 100 == max_mol_len allowen in mutation
new_population.append(selfie_mutated)
new_population.append(best_selfie)
# Define new population for the next generation
population = new_population[:]
if score >= 1:
print('Limit reached')
simlr_path_collect.append(simlr_path)
break
import matplotlib.pyplot as plt
x = [i+1 for i in range(max([len(x) for x in simlr_path_collect]))]
plt.style.use(u'classic')
plt.plot(x, [1.2 for _ in range(len(x))], marker='', color='white', linewidth=4) # axis line
plt.plot(x, [1 for _ in range(len(x))], '--', color='orange', linewidth=2.5, label='Rediscovery') # Highlight line
colors = plt.cm.Blues
profiles = 20
color_shift = 0.4
color_values = [ni/profiles + color_shift for ni in range(profiles)]
for ni in range(len(color_values)):
if color_values[ni] < 0.2:
color_values[ni] -= 1
cm = [colors(x) for x in color_values]
for i,simlr_path in enumerate(simlr_path_collect):
plt.plot([i+1 for i in range(len(simlr_path))], simlr_path, marker='', color=cm[i], linewidth=2.5, alpha=0.5)
plt.title('Rediscovering '+starting_mol_name, fontsize=20, fontweight=0, color='black', loc='left')
plt.xlabel('Generation')
plt.ylabel('ECPF4 Similarity')
plt.savefig('Celecoxib_run.png', dpi=196, bbox_inches='tight')
plt.show()
#ga_conf.yml
params:
N : 20 # Number of runs
simlr_path_collect : []
svg_file_name : 'Troglitazone_run.svg'
starting_mol_name : 'Troglitazone'
data_file_name : '20_runs_data_Troglitazone.txt'
starting_smile : 'CC1=C(C2=C(CCC(O2)(C)COC3=CC=C(C=C3)CC4C(=O)NC(=O)S4)C(=C1O)C)C'
show_gen_out : False
To generate molecule with above config file, just type ‘python flex_GA_rediscovery.py
After running the code, I could get image and compounds as selfies in txt file.
OK, let’s check it. Image file is below. After 50 generation, highly similar compounds are generated.
Next, I check structures which are generated STONED.
As the notebook shows that after running several epochs, target similar compounds are generated. In the original repo some very useful examples are available.
It took short time to conduct GA_based compound generation and it is important that only target molecule is required to run the process.
In summary STONED seems very robust, works fast and doesn’t require huge training data but generates many diverse molecules.
If you have an interest the code, please read original article and check the code.
