RDKit ligand

Setup

Conformer

source

rdkit_conformer

 rdkit_conformer (SMILES, output=None, method='ETKDG', visualize=True,
                  seed=3)

Gemerate 3D conformers from SMILES

Type Default Details
SMILES SMILES string
output NoneType None file “.sdf” to be saved
method str ETKDG Optimization method, can be ‘UFF’, ‘MMFF’ or ‘ETKDGv3’
visualize bool True whether or not to visualize the compound
seed int 3 randomness of the 3D conformation 
rdkit_conformer('CC1=C(C=CC(=C1)NC2=NC=NC3=CN=C(N=C32)N4CCC(CC4)NC(=O)C=C)OC5=CC6=C(C=C5)N(C=N6)C')

source

tanimoto

 tanimoto (df, smiles_col='SMILES', id_col='ID', target_col=None,
           radius=2)

Calculates the Tanimoto similarity scores between all pairs of molecules in a pandas DataFrame.

Type Default Details
df df with SMILES and ID columns
smiles_col str SMILES colname of SMILES
id_col str ID colname of compound ID
target_col NoneType None colname of compound values (e.g., IC50)
radius int 2 radius of the Morgan fingerprint. 
df = Kras.get_mirati_g12d_raw()[['ID','SMILES','IC50']]
df = df.dropna(subset= 'IC50').reset_index(drop=True)
df
ID SMILES IC50
0 US_1 CN1CCC[C@H]1COc1nc(N2CC3CCC(C2)N3)c2cnc(cc2n1)... 124.7
1 US_2 CN1CCC[C@H]1COc1nc(N2CC3CCC(C2)N3)c2cnc(c(F)c2... 2.7
... ... ... ...
703 paper_37 FC1=C(C2=C(C(Cl)=CC=C3)C3=CC(O)=C2)N=CC4=C1N=C... 2.0
704 paper_38 FC1=C(C2=C(C(C#C)=CC=C3)C3=CC(O)=C2)N=CC4=C1N=... 2.0

705 rows × 3 columns

result = tanimoto(df.head(), target_col = 'IC50')
result
ID1 ID2 SMILES1 SMILES2 SimilarityScore Target1 Target2
0 US_3 US_4 Cn1ccnc1CCOc1nc(N2CC3CCC(C2)N3)c2cnc(c(F)c2n1)... Oc1cc(-c2ncc3c(nc(OCCc4ccccn4)nc3c2F)N2CC3CCC(... 0.779221 9.5 496.2
1 US_1 US_2 CN1CCC[C@H]1COc1nc(N2CC3CCC(C2)N3)c2cnc(cc2n1)... CN1CCC[C@H]1COc1nc(N2CC3CCC(C2)N3)c2cnc(c(F)c2... 0.743590 124.7 2.7
... ... ... ... ... ... ... ...
8 US_1 US_3 CN1CCC[C@H]1COc1nc(N2CC3CCC(C2)N3)c2cnc(cc2n1)... Cn1ccnc1CCOc1nc(N2CC3CCC(C2)N3)c2cnc(c(F)c2n1)... 0.505495 124.7 9.5
9 US_1 US_4 CN1CCC[C@H]1COc1nc(N2CC3CCC(C2)N3)c2cnc(cc2n1)... Oc1cc(-c2ncc3c(nc(OCCc4ccccn4)nc3c2F)N2CC3CCC(... 0.488889 124.7 496.2

10 rows × 7 columns

Rdkit feature

source

get_rdkit

 get_rdkit (SMILES:str)

Extract chemical features from SMILES Reference: https://greglandrum.github.io/rdkit-blog/posts/2022-12-23-descriptor-tutorial.html

source

get_rdkit_3d

 get_rdkit_3d (SMILES:str)

Extract 3d features from SMILES

source

get_rdkit_all

 get_rdkit_all (SMILES:str)

Extract chemical features and 3d features from SMILES

source

remove_hi_corr

 remove_hi_corr (df:pandas.core.frame.DataFrame, thr=0.99)

Remove highly correlated features in a dataframe given a pearson threshold

Type Default Details
df DataFrame
thr float 0.99 threshold

source

preprocess

 preprocess (df:pandas.core.frame.DataFrame, thr=0.99)

Remove features with no variance, and highly correlated features based on threshold.

source

get_rdkit_df

 get_rdkit_df (df:pandas.core.frame.DataFrame, include_3d=False,
               col='SMILES', postprocess=False, chunksize=128)

Extract rdkit features (optionally in parallel with 3D) from SMILES in a df

Type Default Details
df DataFrame
include_3d bool False
col str SMILES
postprocess bool False
chunksize int 128 for parallel process_map 
df=Collins.get_antibiotics_2k().head(300)
df.head()
name SMILES inhibition activity
0 CEFPIRAMIDE Cc1cc(O)c(C(=O)NC(C(=O)NC2C(=O)N3C(C(=O)O)=C(C... 0.041572 1
1 GEMIFLOXACIN MESYLATE CON=C1CN(c2nc3c(cc2F)c(=O)c(C(=O)O)cn3C2CC2)CC... 0.041876 1
2 POLYMYXIN B SULFATE CCC(C)CCCCC(=O)NC(CCN)C(=O)NC(C(=O)NC(CCN)C(=O... 0.041916 1
3 PRAXADINE HYDROCHLORIDE Cl.N=C(N)n1cccn1 0.041964 1
4 CHLORHEXIDINE DIHYDROCHLORIDE Cl.Cl.N=C(NCCCCCCNC(=N)NC(=N)Nc1ccc(Cl)cc1)NC(... 0.042295 1 
# %%time
# feat_raw=get_rdkit_df(df)
# feat_raw
CPU times: user 46.6 ms, sys: 65.7 ms, total: 112 ms
Wall time: 2.63 s
MaxAbsEStateIndex MaxEStateIndex MinAbsEStateIndex MinEStateIndex qed SPS MolWt HeavyAtomMolWt ExactMolWt NumValenceElectrons NumRadicalElectrons MaxPartialCharge MinPartialCharge MaxAbsPartialCharge MinAbsPartialCharge FpDensityMorgan1 FpDensityMorgan2 FpDensityMorgan3 BCUT2D_MWHI BCUT2D_MWLOW BCUT2D_CHGHI BCUT2D_CHGLO BCUT2D_LOGPHI BCUT2D_LOGPLOW BCUT2D_MRHI BCUT2D_MRLOW AvgIpc BalabanJ BertzCT Chi0 Chi0n Chi0v Chi1 Chi1n Chi1v Chi2n Chi2v Chi3n Chi3v Chi4n Chi4v HallKierAlpha Ipc Kappa1 Kappa2 Kappa3 LabuteASA PEOE_VSA1 PEOE_VSA10 PEOE_VSA11 ... fr_benzodiazepine fr_bicyclic fr_diazo fr_dihydropyridine fr_epoxide fr_ester fr_ether fr_furan fr_guanido fr_halogen fr_hdrzine fr_hdrzone fr_imidazole fr_imide fr_isocyan fr_isothiocyan fr_ketone fr_ketone_Topliss fr_lactam fr_lactone fr_methoxy fr_morpholine fr_nitrile fr_nitro fr_nitro_arom fr_nitro_arom_nonortho fr_nitroso fr_oxazole fr_oxime fr_para_hydroxylation fr_phenol fr_phenol_noOrthoHbond fr_phos_acid fr_phos_ester fr_piperdine fr_piperzine fr_priamide fr_prisulfonamd fr_pyridine fr_quatN fr_sulfide fr_sulfonamd fr_sulfone fr_term_acetylene fr_tetrazole fr_thiazole fr_thiocyan fr_thiophene fr_unbrch_alkane fr_urea
0 13.503995 13.503995 0.064423 -1.323430 0.162954 18.619048 612.650 588.458 612.120937 218 0 0.352159 -0.507967 0.507967 0.352159 1.214286 1.952381 2.595238 32.167943 10.004034 2.541608 -2.492828 2.413335 -2.707792 8.005500 -0.150504 3.369454 1.339505e+00 1610.181663 30.128663 22.084778 23.717771 20.044622 12.357347 14.391701 9.367961 11.415452 6.531851 8.968356 4.422333 6.757255 -4.64 2.012862e+09 28.873109 11.690583 5.597490 245.926226 25.953159 34.653617 0.000000 ... 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 1 0 2 0 0 0 1 0 0 0 0 0
1 14.874521 14.874521 0.033885 -3.666667 0.398220 19.484848 485.494 461.302 485.138047 180 0 0.340725 -0.477497 0.477497 0.340725 1.424242 2.060606 2.575758 32.239615 10.092419 2.328029 -2.190189 2.370431 -2.256089 7.845502 0.069447 3.471950 5.549999e-07 1265.508949 24.499271 17.933021 18.749518 15.439091 9.902198 11.750510 7.791830 9.257870 5.167537 5.167537 3.595955 3.595955 -3.10 1.783676e+07 24.540619 9.407810 5.638555 187.744947 25.144793 18.320426 11.635084 ... 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
298 11.161773 11.161773 0.278802 -0.471329 0.463380 10.423077 354.358 336.214 354.110338 134 0 0.307526 -0.426733 0.426733 0.307526 0.615385 1.038462 1.500000 16.546477 10.114910 2.058887 -2.069790 2.248839 -2.041320 5.751292 -0.132566 2.519358 2.225734e+00 814.855170 19.104084 14.645642 14.645642 12.312571 7.801219 7.801219 5.421051 5.421051 3.035442 3.035442 2.145140 2.145140 -3.41 4.201281e+05 18.921939 8.617804 6.238343 150.543633 14.210589 17.248535 0.000000 ... 0 0 0 0 0 3 3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
299 6.134942 6.134942 0.000000 0.000000 0.709648 16.421053 412.597 391.429 409.976001 108 0 0.050449 -0.397550 0.397550 0.050449 1.210526 1.842105 2.473684 79.919764 9.939734 2.186581 -2.311907 2.301302 -2.326367 9.108394 0.184585 2.360656 1.804999e-06 420.152146 13.120956 11.163433 15.151922 8.575387 6.513030 8.099027 5.167360 6.937372 3.797772 4.911697 2.688476 4.002186 0.23 1.225652e+04 17.282002 7.866121 4.622779 138.333022 5.733667 0.000000 0.000000 ... 0 0 0 0 0 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

300 rows × 210 columns

# feat = get_rdkit_df(df,postprocess=True)
# feat

Morgan fingerprints (ECFP)

Reference:

source

get_fp

 get_fp (SMILES, name='ecfp', ELEMENTS_PER_WORKER=1000000)

Super fast method to get molecule fingerprints using scikit-fingerprints

# %%time
# fp = get_fp(df.SMILES, name='ecfp')
# fp
0it [00:00, ?it/s]
CPU times: user 21.2 ms, sys: 78 ms, total: 99.2 ms
Wall time: 2.07 s
array([[0, 1, 1, ..., 0, 0, 0],
       [0, 0, 0, ..., 0, 0, 0],
       [0, 1, 0, ..., 0, 0, 0],
       ...,
       [0, 0, 0, ..., 0, 0, 0],
       [0, 0, 0, ..., 0, 0, 0],
       [0, 0, 1, ..., 0, 0, 0]], dtype=uint8)

Use np.packbits to save space

Save space when dealing with super big dataset

Packbits will transform binary to decimals, if it is [1, 0, 1, 1, 0, 0, 0, 0, 1, 1], then it will take 8 bits in the sequence, [10110000] and [11000000], for the latter, if not enough for 8 bits length, it will add 0.

To transform binary to decimals, take [10110000] for example, calculate through 12^7 + 02^6 + 12^5 + 12^4 + 02^3 + 02^2 + 02^1 + 02^0=176

source

compress_fp

 compress_fp (array)

Compress binary finterprints using np.packbits

# compress_fp(fp)
array([[100,   0,   0, ...,  16,   0,   0],
       [  0, 128,   0, ...,   0,   0,   0],
       [ 64,   0,   0, ...,   0,  16,   0],
       ...,
       [  0, 128,   0, ...,   0,   0,   0],
       [  0,   0,   0, ...,   0,  64,   0],
       [ 32,   0,   0, ...,   0,   0,   0]], dtype=uint8)

Tanimoto

Reference: https://www.kaggle.com/code/towardsentropy/fingerprint-tips-and-tricks/notebook

Slow version:

Method 1

from scipy.spatial import distance
# %%time
# tanimotos_scipy = 1-distance.cdist(fp, fp, metric='jaccard')
# # 30.4s

A little faster with parallel

from sklearn.metrics import pairwise_distances
# %%time
# tanimotos_sklearn = 1 - pairwise_distances(fp, metric='jaccard', n_jobs=-1)
# # 13.1s

Method 2

from rdkit.Chem import DataStructs
# %%time
# tanimotos_rdkit = np.array([DataStructs.BulkTanimotoSimilarity(i, fp) for i in fp])

Method 3

def tanimoto_bitwise(fps_np):
    intersection = (fps_np[:,None] & fps_np[None]).sum(-1)
    union = (fps_np[:,None] | fps_np[None]).sum(-1)
    return intersection / union
# tanimoto_bitwise(fp)

Numba tanimoto

source

tanimoto_numba

 tanimoto_numba (fps)

Get a NxN matrix of tanimoto similarity among N compounds.

# %%time
# tanimoto_numba(fp)
CPU times: user 2.02 s, sys: 319 ms, total: 2.34 s
Wall time: 2.4 s
array([[1.        , 0.11695907, 0.12676056, ..., 0.04918033, 0.06617647,
        0.06206897],
       [0.11695907, 1.        , 0.13020833, ..., 0.09375   , 0.0990991 ,
        0.09166667],
       [0.12676056, 0.13020833, 1.        , ..., 0.06206897, 0.05555556,
        0.07142857],
       ...,
       [0.04918033, 0.09375   , 0.06206897, ..., 1.        , 0.03508772,
        0.0625    ],
       [0.06617647, 0.0990991 , 0.05555556, ..., 0.03508772, 1.        ,
        0.0625    ],
       [0.06206897, 0.09166667, 0.07142857, ..., 0.0625    , 0.0625    ,
        1.        ]], dtype=float32)

Group same compounds

Utilize hash sha256 to encode morgan fp and find same molecule

source

hash_fp

 hash_fp (fp_row)

Hash a binary fingerprint row using SHA256

# hash_fp(fp[0])
'8deb65ba1323d46ca59ebf1206a497761ed2b83a518aa7227a965d5426e25aea'

source

get_same_mol_group

 get_same_mol_group (df, smi_col='SMILES')

Assign a group number to the same compounds by utilizing hash sha256 to encode morgan fp and find same molecule.

# %%time
# df2 = get_same_mol_group(df)
# df2.group.value_counts()
0it [00:00, ?it/s]
CPU times: user 26.4 ms, sys: 124 ms, total: 151 ms
Wall time: 237 ms
group
0      1
206    1
      ..
96     1
299    1
Name: count, Length: 300, dtype: int64

End