UMAP Visualization of SARS-CoV-2 Data in ChEMBL

Uniform Manifold Approximation and Projection (UMAP) uses the nearest neighbour graph to perform dimesion reduction for visualization. I have been recently using UMAP to visualize the chemical space coverage of compounds. This piece of code computes UMAP for SARS-CoV-2 Database in ChEMBL consisting of 6900 compounds (downloded in November 28, 2020) as an example. Please modify the path used to suit your own directory when using the same csv file, which could be downloaded from https://www.ebi.ac.uk/chembl/g/#browse/compounds/filter/_metadata.compound_records.src_id%3A52.

#Import modules
import pandas as pd
import numpy as np
from tqdm import tqdm
from rdkit import Chem, DataStructs
from rdkit.Chem import AllChem
import seaborn as sns
import umap

#Read csv file from SARS-Cov-2 Database in ChEMBL (Downloaded in Nov. 28 2020)
#Downloaded from https://www.ebi.ac.uk/chembl/g/#browse/compounds/filter/_metadata.compound_records.src_id%3A52
df = pd.read_csv("./SARS_CoV_2_ChEMBL_Nov_28.csv", sep=';', engine='python')
df

	ChEMBL ID	Name	Synonyms	Type	Max Phase	Molecular Weight	Targets	Bioactivities	AlogP	PSA	...	Structure Type	Inorganic Flag	Heavy Atoms	HBA Lipinski	HBD Lipinski	#RO5 Violations (Lipinski)	Molecular Weight (Monoisotopic)	Molecular Species	Molecular Formula	Smiles
0	CHEMBL2087742	NaN	NaN	Small molecule	0	402.42	16	33	4.31	59.29	...	MOL	-1	29	5	1	0	402.1667	NEUTRAL	C21H21F3N4O	Cc1c(F)cc(C(=O)NC2CC2)cc1-c1c(F)cn2c(C(C)(C)C)...
1	CHEMBL4075936	NaN	NaN	Small molecule	0	327.86	2	4	5.32	20.31	...	MOL	-1	23	2	0	1	327.1390	NEUTRAL	C20H22ClNO	O=C(c1ccc(Cl)cc1)N(Cc1ccccc1)C1CCCCC1
2	CHEMBL3348842	NaN	NaN	Small molecule	0	402.67	2	2	4.59	44.29	...	MOL	-1	29	3	3	0	402.3610	BASE	C26H46N2O	CN[C@@H](C)[C@H]1[C@H](O)C[C@@]2(C)[C@@H]3CC[C...
3	CHEMBL1257423	NaN	NaN	Small molecule	0	418.38	9	14	3.51	96.37	...	MOL	-1	30	7	3	0	418.1253	NEUTRAL	C20H17F3N4O3	O=C(NCCO)c1cccc(-c2cc(Nc3ccc(OC(F)(F)F)cc3)ncn...
4	CHEMBL2182052	NaN	NaN	Small molecule	0	456.50	8	10	6.25	101.66	...	MOL	-1	34	7	2	1	456.1685	ACID	C27H24N2O5	Cc1noc(-c2ccc(-c3ccc(CC(=O)O)cc3)cc2)c1NC(=O)O...
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
6895	CHEMBL751	BROMFENAC SODIUM	AHR-10282B|BROMDAY|BROMFENAC MONOSODIUM SALT|B...	Small molecule	4	356.15	5	13	2.89	80.39	...	MOL	0	20	4	3	0	333.0001	ACID	C15H11BrNNaO3	Nc1c(CC(=O)[O-])cccc1C(=O)c1ccc(Br)cc1.[Na+]
6896	CHEMBL1236682	REFAMETINIB	BAY 869766|BAY 8697661|BAY-86-9766|BAY-869766|...	Small molecule	2	572.34	282	285	3.48	107.89	...	MOL	0	31	7	4	1	572.0090	NEUTRAL	C19H20F3IN2O5S	COc1cc(F)c(F)c(Nc2ccc(I)cc2F)c1NS(=O)(=O)C1(C[...
6897	CHEMBL17350	TRAXOPRODIL	CP-101,606|CP-101606|TRAXOPRODIL	Small molecule	2	327.42	22	47	2.80	63.93	...	MOL	0	24	4	3	0	327.1834	BASE	C20H25NO3	C[C@@H]([C@@H](O)c1ccc(O)cc1)N1CCC(O)(c2ccccc2...
6898	CHEMBL210651	APRICITABINE	APRICITABINE	Small molecule	3	229.26	5	65	-0.59	90.37	...	MOL	0	15	6	3	0	229.0521	NEUTRAL	C8H11N3O3S	Nc1ccn([C@H]2CO[C@@H](CO)S2)c(=O)n1
6899	CHEMBL572	NITROFURANTOIN	BERKFURIN|CEDURAN|Dantafur|FURADANTIN|FURALAN|...	Small molecule	4	238.16	207	3101	0.07	118.05	...	MOL	0	17	9	1	0	238.0338	NEUTRAL	C8H6N4O5	O=C1CN(/N=C/c2ccc([N+](=O)[O-])o2)C(=O)N1

6900 rows × 31 columns

#Remove nan from 'Smiles' column in df 
SARS_CoV_2_smiles_filter = [x for x in df.Smiles if str(x) != 'nan']

#Define methods for fingerprints
#Computed using 2048 bits with radius 2

#Convert list of SMILES to list of fingerprints
def fingerprint_list_from_smiles_list(smiles_list, n_bits=2048):
    fingerprint_list = []
    for smiles in tqdm(smiles_list):
        mol = Chem.MolFromSmiles(smiles)
        fingerprint_list.append(fingerprint_as_array(mol, n_bits))
    return fingerprint_list

def fingerprint_as_array(mol, n_bits=2048):
    fingerprint = AllChem.GetMorganFingerprintAsBitVect(mol, 2, nBits=n_bits)
    array = np.zeros((1,), np.int)
    DataStructs.ConvertToNumpyArray(fingerprint, array)
    return array

#Convert list of SMILES for SARS_CoV_2 to fingerprint
fingerprint_SARS_CoV_2_smiles_filter = fingerprint_list_from_smiles_list(SARS_CoV_2_smiles_filter)

100%|██████████| 6843/6843 [00:04<00:00, 1503.90it/s]

#Convert list of array to a single array using numpy
fingerprint_array_SARS_CoV_2_smiles_filter = np.array(fingerprint_SARS_CoV_2_smiles_filter)

#Compute UMAP on the SARS_CoV_2 database 
umap = umap.UMAP()

umap_fingerprint_array_SARS_CoV_2_smiles_filter = umap.fit_transform(fingerprint_array_SARS_CoV_2_smiles_filter)

#Place the UMAP result into a pandas dataframe for plotting
umap_fingerprint_array_SARS_CoV_2_smiles_filter_fig = pd.DataFrame(umap_fingerprint_array_SARS_CoV_2_smiles_filter,columns=["X","Y"])

#Set the figure for seaborn
sns.set(rc={'figure.figsize': (10, 10)})
sns.set(font_scale=1.5)
sns.set_style('whitegrid')

#Visualize the UMAP using seaborn
fig = sns.jointplot(data=umap_fingerprint_array_SARS_CoV_2_smiles_filter_fig,x="X",y="Y", kind="hex", height=10, ratio=5, color="blue")