Molecular design

In [2]:

import os
import sys

os.environ["TOKENIZERS_PARALLELISM"] = "false"

import safe as sf
import datamol as dm
from safe.sample import logger
# clean logger a bit
logger.configure(handlers=[{"sink": sys.stdout, "level": "INFO"}])

import os import sys os.environ["TOKENIZERS_PARALLELISM"] = "false" import safe as sf import datamol as dm from safe.sample import logger # clean logger a bit logger.configure(handlers=[{"sink": sys.stdout, "level": "INFO"}])

Out[2]:

[1]

Load the default pretrained Safe model.

We will use this unique model for all the downstream molecular design tasks.

In [3]:

designer = sf.SAFEDesign.load_default(verbose=True)

designer.model

designer = sf.SAFEDesign.load_default(verbose=True) designer.model

/Users/emmanuel.noutahi/miniconda3/envs/safe/lib/python3.12/site-packages/huggingface_hub/file_download.py:1150: FutureWarning: `resume_download` is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use `force_download=True`.
  warnings.warn(

Out[3]:

SAFEDoubleHeadsModel(
  (transformer): GPT2Model(
    (wte): Embedding(1880, 768)
    (wpe): Embedding(1024, 768)
    (drop): Dropout(p=0.1, inplace=False)
    (h): ModuleList(
      (0-11): 12 x GPT2Block(
        (ln_1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2SdpaAttention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
    )
    (ln_f): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
  )
  (lm_head): Linear(in_features=768, out_features=1880, bias=False)
  (multiple_choice_head): PropertyHead(
    (summary): Linear(in_features=768, out_features=64, bias=True)
    (activation): ReLU()
    (out): Linear(in_features=64, out_features=1, bias=True)
  )
)

Let's start with the below molecule.

In [4]:

candidate_smiles = "O=C(C#CCN1CCCCC1)Nc1ccc2ncnc(Nc3cccc(Br)c3)c2c1"
candidate_mol = dm.to_mol(candidate_smiles)

dm.to_image(candidate_mol)

candidate_smiles = "O=C(C#CCN1CCCCC1)Nc1ccc2ncnc(Nc3cccc(Br)c3)c2c1" candidate_mol = dm.to_mol(candidate_smiles) dm.to_image(candidate_mol)

Out[4]:

De novo generation¶

Generation of novel molecules without any constraints.

In [5]:

generated_smiles = designer.de_novo_generation(sanitize=True, n_samples_per_trial=12)

generated_smiles[:5]

generated_smiles = designer.de_novo_generation(sanitize=True, n_samples_per_trial=12) generated_smiles[:5]

  0%|          | 0/1 [00:00<?, ?it/s]

/Users/emmanuel.noutahi/miniconda3/envs/safe/lib/python3.12/site-packages/transformers/tokenization_utils_base.py:1601: FutureWarning: `clean_up_tokenization_spaces` was not set. It will be set to `True` by default. This behavior will be depracted in transformers v4.45, and will be then set to `False` by default. For more details check this issue: https://github.com/huggingface/transformers/issues/31884
  warnings.warn(
/Users/emmanuel.noutahi/miniconda3/envs/safe/lib/python3.12/site-packages/transformers/generation/configuration_utils.py:615: UserWarning: `num_beams` is set to 1. However, `early_stopping` is set to `True` -- this flag is only used in beam-based generation modes. You should set `num_beams>1` or unset `early_stopping`.
  warnings.warn(

2024-09-12 14:35:07.152 | INFO     | safe.sample:de_novo_generation:760 - After sanitization, 12 / 12 (100.00 %) generated molecules are valid !

Out[5]:

['CC(C)(C)OC(=O)N1CC[C@@]2(C(=O)N3C[C@H]4COC[C@@H](C3)[C@H]4NC(=O)C3(C)COC3)C[C@@H]2C1',
 'C[C@@H](N=C(O)Cc1ccc([N+](=O)[O-])cc1)c1ccccn1',
 'N#Cc1ccc(N2C[C@H]3CC[C@@H](C2)N3C(=O)CCc2cn[nH]n2)nc1',
 'CC(C)c1occc1C(=O)N1CC[C@@H](CNc2cnc(C#N)cn2)C1',
 'C=C(Cl)CN(Cc1ccco1)C(=O)NCc1ccccc1CNC(=O)OC(C)(C)C']

In [6]:

dm.to_image(generated_smiles[:12], mol_size=(350, 200))

dm.to_image(generated_smiles[:12], mol_size=(350, 200))

Out[6]:

Scaffold Decoration¶

For scaffold decoration, we wish to generate new molecules that would contain a given scaffold as core. Usually, the attachment point on the scaffold should dictate where the new vectors will be added.

In [7]:

scaffold = "[*]N-c1ccc2ncnc(-N[*])c2c1"

dm.to_image(scaffold)

scaffold = "[*]N-c1ccc2ncnc(-N[*])c2c1" dm.to_image(scaffold)

Out[7]:

In [8]:

generated_smiles = designer.scaffold_decoration(
    scaffold=scaffold,
    n_samples_per_trial=12,
    n_trials=2,
    sanitize=True,
    do_not_fragment_further=True,
)

generated_mols = [dm.to_mol(x) for x in generated_smiles]

generated_smiles = designer.scaffold_decoration( scaffold=scaffold, n_samples_per_trial=12, n_trials=2, sanitize=True, do_not_fragment_further=True, ) generated_mols = [dm.to_mol(x) for x in generated_smiles]

  0%|          | 0/2 [00:00<?, ?it/s]

2024-09-12 14:35:21.081 | INFO     | safe.sample:scaffold_decoration:635 - After sanitization, 23 / 24 (95.83 %)  generated molecules are valid !

In [9]:

dm.viz.lasso_highlight_image(generated_mols[:12], dm.from_smarts(scaffold), mol_size=(350, 200), color_list=["#ff80b5"], scale_padding=0.1)

dm.viz.lasso_highlight_image(generated_mols[:12], dm.from_smarts(scaffold), mol_size=(350, 200), color_list=["#ff80b5"], scale_padding=0.1)

Out[9]:

Pattern decoration¶

For pattern decoration, we wish to generate new molecules that would contain a given molecular pattern expressed as a SMARTS. SMARTS are different from scaffolds as they allow allows "rules" for matching several atoms and bonds simultaneously. In SAFE, the pattern decoration only supports a subset of the SMARTS specification. We recommend filtering generated compounds after to preserve your specifc input patterns. Alternatively, you can use the sanitize input.

In [10]:

patterns = "FC(F)(F)C(*)[N,#8]C(=O)c1[c,n]cc[c,n]c1"
dm.to_image(dm.from_smarts(patterns))

patterns = "FC(F)(F)C(*)[N,#8]C(=O)c1[c,n]cc[c,n]c1" dm.to_image(dm.from_smarts(patterns))

Out[10]:

In [11]:

generated_smiles = designer.pattern_decoration(patterns,
    n_samples_per_trial=50,
    n_trials=1,
    sanitize=True,
    n_scaff_samples=10,
    n_scaff_random=5,
    do_not_fragment_further=True,
)
    
generated_mols = [dm.to_mol(x) for x in generated_smiles]
dm.viz.lasso_highlight_image(generated_mols[:12], dm.from_smarts(patterns), mol_size=(350, 200), color_list=["#ff80b5"], scale_padding=0.1)

generated_smiles = designer.pattern_decoration(patterns, n_samples_per_trial=50, n_trials=1, sanitize=True, n_scaff_samples=10, n_scaff_random=5, do_not_fragment_further=True, ) generated_mols = [dm.to_mol(x) for x in generated_smiles] dm.viz.lasso_highlight_image(generated_mols[:12], dm.from_smarts(patterns), mol_size=(350, 200), color_list=["#ff80b5"], scale_padding=0.1)

  0%|          | 0/1 [00:00<?, ?it/s]

  0%|          | 0/1 [00:00<?, ?it/s]

  0%|          | 0/1 [00:00<?, ?it/s]

  0%|          | 0/1 [00:00<?, ?it/s]

  0%|          | 0/1 [00:00<?, ?it/s]

  0%|          | 0/1 [00:00<?, ?it/s]

  0%|          | 0/1 [00:00<?, ?it/s]

  0%|          | 0/1 [00:00<?, ?it/s]

  0%|          | 0/1 [00:00<?, ?it/s]

  0%|          | 0/1 [00:00<?, ?it/s]

  0%|          | 0/1 [00:00<?, ?it/s]

  0%|          | 0/1 [00:00<?, ?it/s]

  0%|          | 0/1 [00:00<?, ?it/s]

  0%|          | 0/1 [00:00<?, ?it/s]

2024-09-12 14:35:42.532 | INFO     | safe.sample:pattern_decoration:719 - After sanitization, 27 / 50 (54.00%) generated molecules are valid!

Out[11]:

Super structure generation¶

In super structure generation, we just want to generate superstructure of a molecular subgraph

In [12]:

superstructure = "c1ccc2ncncc2c1"

dm.to_image(superstructure)

superstructure = "c1ccc2ncncc2c1" dm.to_image(superstructure)

Out[12]:

In [13]:

generated_smiles = designer.super_structure(
    core=superstructure,
    n_samples_per_trial=12,
    n_trials=1,
    sanitize=True,
    do_not_fragment_further=False,
    attachment_point_depth=3,
)

generated_smiles

generated_smiles = designer.super_structure( core=superstructure, n_samples_per_trial=12, n_trials=1, sanitize=True, do_not_fragment_further=False, attachment_point_depth=3, ) generated_smiles

  0%|          | 0/1 [00:00<?, ?it/s]

2024-09-12 14:35:44.980 | INFO     | safe.sample:super_structure:589 - After sanitization, 12 / 12 (100.00 %)  generated molecules are valid !

Out[13]:

['NC=Nc1nc(C(N)=O)c2cc([N+](=O)[O-])ccc2n1',
 'N#CCCNc1ccc2nc(C(N)=O)nc(C(N)=O)c2c1',
 'FC(F)(F)c1ccc2nc(C(F)(F)F)nc(C(F)(F)F)c2c1',
 'FC(F)COCCNc1ccc2nc(C(F)(F)F)nc(C(F)(F)F)c2c1',
 'NC=Nc1nc(/C=C/N)c2cc(S(N)(=O)=O)ccc2n1',
 'O=[N+]([O-])c1ccc2nc(C(F)F)nc(N=S)c2c1',
 'NC(=S)c1ccc2ncnc(N=CCC(F)(F)F)c2c1',
 'N#CNc1nc(C(F)(F)C(F)(F)F)c2cc(C(F)(F)F)ccc2n1',
 'OCc1nc(NCCCC(F)(F)F)nc2ccc(C(F)(F)F)cc12',
 'O[C@H](CCNc1nc(C(F)(F)F)nc2ccc(C(F)(F)F)cc12)C(F)(F)F',
 'O=C([O-])C(F)(F)C(F)(F)C(=O)Nc1nc(CO)c2cc(C(F)(F)F)ccc2n1',
 'N#C[C@@H](Nc1nc(C(F)(F)F)c2cc(C(=O)[O-])ccc2n1)C(N)=O']

In [14]:

dm.to_image(generated_smiles[:12], mol_size=(350, 200))

dm.to_image(generated_smiles[:12], mol_size=(350, 200))

Out[14]:

Motif Extension¶

In motif extension, we are interested in generating a molecule containing a given motif as starting point.

In [15]:

motif = "[*]-N1CCCCC1"

dm.to_image(motif)

motif = "[*]-N1CCCCC1" dm.to_image(motif)

Out[15]:

In [16]:

# let's make some long sequence
generated_smiles = designer.motif_extension(
    motif=motif,
    n_samples_per_trial=12,
    n_trials=1,
    sanitize=True,
    do_not_fragment_further=False,
    min_length=25,
    max_length=80,
)

generated_smiles

# let's make some long sequence generated_smiles = designer.motif_extension( motif=motif, n_samples_per_trial=12, n_trials=1, sanitize=True, do_not_fragment_further=False, min_length=25, max_length=80, ) generated_smiles

  0%|          | 0/1 [00:00<?, ?it/s]

2024-09-12 14:35:47.217 | INFO     | safe.sample:scaffold_decoration:635 - After sanitization, 11 / 12 (91.67 %)  generated molecules are valid !

Out[16]:

['O=S(=O)([O-])CCCCCCCCCN1CCCCC1.[Na+]',
 'O=C(c1n[nH]nc1N1CCCCC1)N1CCCCC1',
 'C1CCN(CC2CCCNCC2)CC1',
 'C1CCN([C@H]2CNCCO2)CC1.Cl.Cl.Cl',
 'C1CCN(C2CCC2)CC1.N=c1ncc(C(F)(F)F)n[nH]1',
 'Cl.N=C(O)C(N)CN1CCCCC1',
 'O=C(NCCSC(F)(F)F)[C@@H](CCC(F)(F)F)N1CCCCC1',
 'C1CCN(CCNC2CCCNCC2)CC1',
 'O=C(NCCN(CC(=O)N1CCCCC1)CC(F)F)C(F)F',
 'CC(CF)(CF)NS(=O)(=O)CCCS(=O)(=O)N1CCCCC1',
 'O=C1CCCCN1CCN1CCCCC1']

In [17]:

dm.to_image(generated_smiles[:12], mol_size=(350, 200))

dm.to_image(generated_smiles[:12], mol_size=(350, 200))

Out[17]:

Scaffold Morphing¶

In scaffold morphing, we wish to replace a scaffold by another one in a molecule. The process requires as input that the user provides either the side chains or the input molecules and the core

In [18]:

side_chains = "[1*]C(=O)C#CCN1CCCCC1.[2*]c1cccc(Br)c1"

dm.to_image(side_chains)

side_chains = "[1*]C(=O)C#CCN1CCCCC1.[2*]c1cccc(Br)c1" dm.to_image(side_chains)

Out[18]:

In [19]:

generated_smiles = designer.scaffold_morphing(
    side_chains=side_chains,
    n_samples_per_trial=12,
    n_trials=1,
    sanitize=True,
    do_not_fragment_further=False,
    random_seed=100,
)

dm.to_image(generated_smiles[:12], mol_size=(350, 200))

generated_smiles = designer.scaffold_morphing( side_chains=side_chains, n_samples_per_trial=12, n_trials=1, sanitize=True, do_not_fragment_further=False, random_seed=100, ) dm.to_image(generated_smiles[:12], mol_size=(350, 200))

  0%|          | 0/1 [00:00<?, ?it/s]

2024-09-12 14:35:52.033 | INFO     | safe.sample:_fragment_linking:490 - After sanitization, 8 / 12 (66.67 %)  generated molecules are valid !

Out[19]:

Linker generation¶

Linker generation is mostly the same thing as scaffold morphing ...

In [20]:

linker_generation = ["[*]-N1CCCCC1", "Brc1cccc(Nc2ncnc3ccc(-[*])cc23)c1"]

dm.to_image(linker_generation)

linker_generation = ["[*]-N1CCCCC1", "Brc1cccc(Nc2ncnc3ccc(-[*])cc23)c1"] dm.to_image(linker_generation)

Out[20]:

In [21]:

generated_smiles = designer.linker_generation(
    *linker_generation,
    n_samples_per_trial=12,
    n_trials=1,
    sanitize=True,
    do_not_fragment_further=False,
    random_seed=100,
)

generated_smiles

generated_smiles = designer.linker_generation( *linker_generation, n_samples_per_trial=12, n_trials=1, sanitize=True, do_not_fragment_further=False, random_seed=100, ) generated_smiles

  0%|          | 0/1 [00:00<?, ?it/s]

2024-09-12 14:35:56.575 | INFO     | safe.sample:_fragment_linking:490 - After sanitization, 12 / 12 (100.00 %)  generated molecules are valid !

Out[21]:

['Brc1cccc(Nc2ncnc3ccc(-c4ccc5nc6c(-n7cccn7)ccc(-c7ccc(OCCCN8CCCCC8)cc7)c6nc5c4)cc23)c1',
 'Brc1cccc(Nc2ncnc3ccc(CCCOc4ccc(-c5ccc(-n6cccn6)c6nc7ccc(N8CCCCC8)cc7nc56)cc4)cc23)c1',
 'Brc1cccc(Nc2ncnc3ccc(-c4ccc5c(c4)c4c6ccccc6n(-c6cccc(N7CCCCC7)c6)c4n5-c4ccccc4)cc23)c1',
 'Brc1cccc(Nc2ncnc3ccc(-c4cccc(-n5c6ccccc6c6c7cc(N8CCCCC8)ccc7n(-c7ccccc7)c65)c4)cc23)c1',
 'O=C(CN1CC=C(CCCN=c2[nH]n(C(=O)CN3CCCCC3)c3ncccc23)CC1)n1nc(-c2ccc3ncnc(Nc4cccc(Br)c4)c3c2)c2cccnc21',
 'O=C(CN1CC=C(CCCN=c2[nH]n(C(=O)Cc3ccc4ncnc(Nc5cccc(Br)c5)c4c3)c3ncccc23)CC1)n1nc(N2CCCCC2)c2cccnc21',
 'Brc1cccc(Nc2ncnc3ccc(Cc4ccc(-c5ccc(N6CCCCC6)cc5)cc4)cc23)c1',
 'Brc1cccc(Nc2ncnc3ccc(-c4ccc(-c5ccc(CN6CCCCC6)cc5)cc4)cc23)c1',
 'O=C(NC1NNC2C1=CC=CC2CN1CCCCC1)c1ccc(-c2ccc3ncnc(Nc4cccc(Br)c4)c3c2)s1',
 'O=C(NC1NNC2C1=CC=CC2Cc1ccc2ncnc(Nc3cccc(Br)c3)c2c1)c1ccc(N2CCCCC2)s1',
 'Brc1cccc(Nc2ncnc3ccc([Si](c4ccccc4)(c4ccccc4)c4cccc(N5CCCCC5)n4)cc23)c1',
 'Brc1cccc(Nc2ncnc3ccc(-c4cccc([Si](c5ccccc5)(c5ccccc5)N5CCCCC5)n4)cc23)c1']

In [22]:

dm.to_image(generated_smiles[:12], mol_size=(350, 200))

dm.to_image(generated_smiles[:12], mol_size=(350, 200))

Out[22]:

---

The End !