Sourcing chemical structures for in silico discovery

Getting ionizable lipid structures from online catalogs

Online catalogs of chemical structures

The building blocks of an ionizable lipid are fragments of different moieties in the lipid. These fragments are shared in common across combinatorial libraries, providing insight and control into their variations.

In the papers in which they appear, the fragments are typically dictated by synthesis or other resource constraints. For in silico AI/ML, we often want to mine the SMILES and lipids corresponding to known structures, which is a daunting task given the number that have been studied. Scouring the web for catalogs and literature online is the only way to keep up with such structures.

We go through the process of mining two such catalogs – Broadpharm and Medchem – which together cover the vast majority of available fragments and structures.

These cover a large fraction of the available literature, including structures that are not under patent. We will scrape each of these catalogs and combine them into a data frame that is the result of this notebook.

Procedure

In each case, getting the necessary structures from the catalogs is a two-step process:

1. Retrieve structure images from the catalog: This is done by scraping the catalog for the internal IDs of chemical structures. These correspond to structure images in the catalog. Each image is then downloaded from the catalog and saved to a local directory. The implementation below takes a URL and returns a data frame of the structures in the catalog.

import requests, os
from urllib.parse import urlparse

def scrape_url(url):
    """
    Scrapes the HTML content from a given URL and returns it as a string.
    
    Args:
        url (str): The URL to scrape
        
    Returns:
        str: The HTML content of the page
        
    Raises:
        ValueError: If the URL is invalid
        requests.RequestException: If the request fails
    """
    # Validate URL
    parsed_url = urlparse(url)
    if not parsed_url.scheme or not parsed_url.netloc:
        raise ValueError("Invalid URL. Please provide a complete URL including http:// or https://")
    
    # Set a user agent to mimic a browser request
    headers = {
        'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36'
    }
    
    # Make the request
    try:
        response = requests.get(url, headers=headers, timeout=10)
        response.raise_for_status()  # Raise an exception for 4XX/5XX responses
        return response.text
    except requests.exceptions.RequestException as e:
        raise requests.RequestException(f"Failed to retrieve the webpage: {e}")


def retrieve_chemstructure_images(IDs, names, imageID_to_url, name_to_path):
    img_path = {}
    for i in range(len(IDs)):
        if len(names[i]) == 0:
            continue
        cpd_name = names[i]
        image_url = imageID_to_url(IDs[i])
        if '/' in cpd_name:
            cpd_name = cpd_name.replace('/', '-')
        image_path = name_to_path(cpd_name)
        if not os.path.exists(image_path):
            img_data = requests.get(image_url).content
            print(f'Downloading {image_url} to {image_path}')
            with open(image_path, 'wb') as handler:
                handler.write(img_data)
        img_path[cpd_name] = image_path
    return img_path

2. Convert the structure images to SMILES strings: This is done using deep computer vision models that are trained for this specific task. We use the DECIMER transformer model for this purpose.

The SMILES string conversion requires some manual verification of the answers, so it is not yet fully automated. Though further automatic SMILES conversion presents some challenges, there is scope for addressing these as needed.¹

from DECIMER import predict_SMILES as DECIMER_predict_SMILES

def predict_SMILES_from_images(img_path_dict):
    # To run this, DECIMER must be installed, requiring opencv-python and keras-preprocessing
    from DECIMER import predict_SMILES as DECIMER_predict_SMILES
    chem_smiles = {}
    i = 0
    for cpd_name, image_path in img_path_dict.items():
        SMILES_str = DECIMER_predict_SMILES(image_path)
        #SMILES_str = openchemie_predict_SMILES(image_path)
        chem_smiles[cpd_name] = SMILES_str
        i += 1
        print(i, cpd_name)
    return chem_smiles

Examples

A look at Broadpharm’s IL catalog

Broadpharm is one of the largest vendors for ionizable lipids in LNP design and has a detailed and organized catalog with solid coverage of the literature. The structures available there can be inspected from their website.

in_url = "https://broadpharm.com/product-categories/lipid/ionizable-lipid"
in_str = scrape_url(in_url)

Get catalog IDs and corresponding chemical names of all structures

Each structure has a unique catalog ID which is necessary to retrieve it from within the catalog. But we typically want to store it under a different name – often the structure’s trademarked name, or another referent that is useful for looking it up in the literature.

We first retrieve the mapping between the catalog IDs and chemical names – a quick step that only involves some HTML traversals. BeautifulSoup is a standard HTML/XML parser that makes this process easier.

The code here is specific to the format of the Broadpharm catalog, and would change with time and other catalogs. The best way to write this code is therefore adaptively, with a code generation model.

from bs4 import BeautifulSoup
import pandas as pd

# Parse the HTML data using BeautifulSoup
soup = BeautifulSoup(in_str, 'html.parser')


# Initialize an empty list to store the extracted data
compound_data = []

for tr in soup.find_all('tr'):
    td = tr.find_all('td')
    row = [i.text for i in td]
    if len(row) != 6:
        pass  # print(row)
    elif row[1] != '':
        compound = {
            'Product ID': row[0],
            'Name': row[1],
            'Molecular Structure': row[2],
            'Molecular Weight': row[3],
            'Purity': row[4],
            'Pricing': row[5]
        }
        compound_data.append(compound)
        

# Convert the list of dictionaries to a Pandas DataFrame
compound_df = pd.DataFrame(compound_data)

IDs = compound_df['Product ID'].tolist()
names = compound_df['Name'].tolist()

Using this mapping, it is easy to retrieve the images of all structures with their catalog IDs.

file_path_pfx = "../../files/LNP_catalog/"

broadpharm_imageID_to_url = lambda x: f'https://broadpharm.com/web/images/mol_images/{x}.gif'
# If the folder f'{file_path_pfx}medchem_lipids/' does not exist, create it.
if not os.path.exists(f'{file_path_pfx}broadpharm_lipids/'):
    os.makedirs(f'{file_path_pfx}broadpharm_lipids/')

broadpharm_name_to_path = lambda x: f'{file_path_pfx}broadpharm_lipids/{x}.png'
img_path = retrieve_chemstructure_images(IDs, names, broadpharm_imageID_to_url, broadpharm_name_to_path)

Predict SMILES from image paths in dataframe

The next step is to predict the SMILES strings from the image paths in the dataframe, which is where the DECIMER model comes in.

(This could also be done using the openchemie toolkit, from which the image recognition model (Qian et al. 2023) has compared favorably to other methods from the literature (Rajan et al. 2024). )

from datetime import date
dtime = date.today().strftime("%Y-%m-%d")

catalog_name = f'Broadpharm'
chem_smiles = predict_SMILES_from_images(img_path)

new_fname = file_path_pfx + f'{catalog_name}_smiles.tsv'
pd.Series(chem_smiles).to_csv(new_fname, sep='\t', header=False)

Medchem’s IL catalog

Doing the same for MedChem’s catalog illustrates how catalog-specific these workflows get.

in_url = "https://www.medchemexpress.com/search.html?q=ionizable+lipid&type=inhibitors-and-agonists"

# scrape the url manually to yield the string below, because the website appears to not be statically rendered.
from selenium import webdriver
from selenium.webdriver.chrome.options import Options

options = Options()
options.add_argument("--headless")
options.add_argument("--disable-blink-features=AutomationControlled")
options.add_argument(
    "user-agent=Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) "
    "AppleWebKit/537.36 (KHTML, like Gecko) Chrome/124.0.0.0 Safari/537.36"
)

driver = webdriver.Chrome(options=options)

driver.get("https://www.medchemexpress.com/search.html?q=ionizable+lipid&type=inhibitors-and-agonists")
# driver.execute_script("Object.defineProperty(navigator, 'webdriver', {get: () => undefined})")

in_str = driver.page_source
driver.quit()

Get catalog IDs and corresponding chemical names of all structures

Again, BeautifulSoup comes to the rescue, though this code needs to be adapted to the specific structure of the Medchem catalog this time. The code can in general be written by an LLM-based agent.

from bs4 import BeautifulSoup
import requests, os

import pandas as pd


# Parse the HTML data using BeautifulSoup
new_soup = BeautifulSoup(in_str, 'html.parser')

names = []
IDs = []
i = 0
for x in new_soup.find_all('li'):
    if x.dl is not None:
        if len(x.dl.tr.text) > 0:
            names.append(x.dl.tr.text.strip().split('\n')[0])
        else:
            if len(x.dl.tr.a.contents) >= 2:
                names.append(x.dl.tr.a.contents[1].contents[0])
            else:
                names.append(x.dl.tr.a.contents[0].contents[0])
        if len(x.dl.dt.text) > 0:
            IDs.append(x.dl.dt.text)
        else:
            IDs.append(x.dl.dt.a.contents[0])
        i += 1

Armed with this mapping, we now again retrieve the images of all structures.

medchem_imageID_to_url = lambda x: f'https://file.medchemexpress.com/product_pic/{x}.gif'
# If the folder f'{file_path_pfx}medchem_lipids/' does not exist, create it.
if not os.path.exists(f'{file_path_pfx}medchem_lipids/'):
    os.makedirs(f'{file_path_pfx}medchem_lipids/')

medchem_name_to_path = lambda x: f'{file_path_pfx}medchem_lipids/{x}.png'
img_path = retrieve_chemstructure_images(IDs, names, medchem_imageID_to_url, medchem_name_to_path)

Predict SMILES from image paths in dataframe

Again we proceed very similarly to the Broadpharm catalog, using model predictions to convert the images to SMILES strings.

from datetime import date
dtime = date.today().strftime("%Y-%m-%d")

catalog_name = f'Medchem'
chem_smiles = predict_SMILES_from_images(img_path)

new_fname = file_path_pfx + f'{catalog_name}_smiles.tsv'
pd.Series(chem_smiles).to_csv(new_fname, sep='\t', header=False)

Consolidate known lipid structures

We can now merge all the catalogs that have been collected, collapsing duplicate entries appropriately and logging the source(s) of any structure. Doing a quick RDKit canonical-SMILES comparison during this process corrects for the indeterminacy in how the structure is represented (there is more than one valid SMILES for a molecule, e.g. depending on the choice of starting atom). So only unique structures are included in the final database.

import numpy as np, pandas as pd
file_path_pfx = "../../files/LNP_catalog/"
smiles_df_paths = ['Broadpharm_smiles.tsv', 'Medchem_smiles.tsv']
consolidated_smiles = {}
for fname in smiles_df_paths:
    s = file_path_pfx + fname
    cat_name = s.split('_smiles')[0]
    consolidated_smiles[cat_name] = dict(np.array(pd.read_csv(s, sep='\t', header=None)))

from rdkit import Chem

new_df = {
    'Catalog': [],
    'Name': [],
    'SMILES': []
}
for catname in consolidated_smiles.keys():
    thiscat = consolidated_smiles[catname]
    thiscat_list = [x for x in zip(*thiscat.items())]
    new_df['Name'].extend(list(thiscat_list[0]))
    new_df['SMILES'].extend(list(thiscat_list[1]))
    new_df['Catalog'].extend([catname] * len(thiscat))
new_df = pd.DataFrame(new_df)
new_df['SMILES'] = [x.replace('[R5]', 'N=[N+]=[N-]') for x in new_df['SMILES']]
new_df['SMILES'] = [Chem.CanonSmiles(x) for x in new_df['SMILES']]

# We find that the model occasionally mislabels explicitly drawn hydrogens as isotopic, so we fix that.
new_df['SMILES'] = [x.replace('[2H]', '[H]') for x in new_df['SMILES']]
new_df['SMILES'] = [x.replace('[3H]', '[H]') for x in new_df['SMILES']]

# Consolidate so that new_df['SMILES'] are all unique. For any duplicates, combine them by concatenating the contents of each of their other columns.
new_df_combined = new_df.groupby('SMILES').agg(lambda x: ' | '.join(set(x))).reset_index()

This combined set of SMILES can be further analyzed, or used to seed a virtual space.

new_df_combined.to_csv(file_path_pfx + 'known_IL_combined.tsv', sep='\t', header=False, index=False)

Directly sourcing fragments

By far the most comprehensive way to ensure broad virtual screening is to collect fragments, not structures, because an ultralarge virtual space defines combinatorially many structures from its fragments.

To mine such fragments, we turn to a growing body of literature using libraries that are combinatorially defined, where the study itself specifies fragments rather than the entire library of structures. Such studies almost always explicitly enumerate the library, and we can massively amplify their power by incorporating their fragments into ultralarge libraries for virtual screening.

Amine head fragments

We can retrieve a vast variety of amine fragments from non-LNP catalogs, for use in synthesis. We download an .sdf file directly from the Chemspace catalog at this link.

import zipfile, requests


def local_zip_download(zip_path, output_dir):
    new_fname = zip_path.split('/')[-1]
    # Check if new_fname is a zip file
    if not new_fname.endswith('.zip'):
        return ''
    out_path = output_dir + new_fname
    with requests.get(zip_path, stream=True) as r:
        r.raise_for_status()
        with open(out_path, "wb") as f:
            for chunk in r.iter_content(chunk_size=8192):
                f.write(chunk)
    return out_path


def extract_sdf_from_zip(zip_path, output_dir, download=False):
    if download:
        zip_path = local_zip_download(zip_path, output_dir)
    fnames_written = []
    with zipfile.ZipFile(zip_path, 'r') as zip_ref:
        for file_name in zip_ref.namelist():
            if file_name.lower().endswith(".sdf"):
                zip_ref.extract(file_name, output_dir)
                print(f"Extracted: {file_name} to {output_dir}")
                fnames_written.append(file_name)
    return fnames_written

Chemspace stores these structures as a compressed .sdf file.

file_path_pfx = "../../files/LNP_catalog/"
chemspace_file_path = file_path_pfx + 'Chemspace_Amine_Fragments_Set.zip'  # Replace with the desired local filename and extension

from rdkit import Chem
from rdkit.Chem import PandasTools


sdf_dir = file_path_pfx
sdf_fnames = extract_sdf_from_zip(chemspace_file_path, sdf_dir)

chemspace_sdf = PandasTools.LoadSDF(sdf_dir + sdf_fnames[0], removeHs=False)
chemspace_sdf['SMILES'] = [Chem.MolToSmiles(x) for x in chemspace_sdf['ROMol']]
chemspace_sdf['Name'] = chemspace_sdf['CHEMSPACE_ID']
chemspace_sdf

Extracted: Chemspace_Amine_Fragments_Set.sdf to ../../files/LNP_catalog/

	CHEMSPACE_ID	CHEMSPACE_URL	ID	ROMol	SMILES	Name
0	CSSS00007999857	https://chem-space.com/CSSS00007999857			CNC1CCCN(C2Cc3ccccc3C2)C1	CSSS00007999857
1	CSSS00012024712	https://chem-space.com/CSSS00012024712			Oc1cccc2c1CCCN2	CSSS00012024712
2	CSSS02018321032	https://chem-space.com/CSSS02018321032			O=S(=O)(Cc1ccc(F)cc1)N1CC2CCC(C1)N2	CSSS02018321032
3	CSSS00102954942	https://chem-space.com/CSSS00102954942			CN(C)c1nccc2c1CCNC2.Cl	CSSS00102954942
4	CSSS00133055275	https://chem-space.com/CSSS00133055275			Cc1ccc2oc(C(=O)NCC3CCCCN3)cc2c1.Cl	CSSS00133055275
...	...	...	...	...	...	...
18590	CSSS06359898475	https://chem-space.com/CSSS06359898475			Cl.FCCNC1CCOC1	CSSS06359898475
18591	CSSS00021525833	https://chem-space.com/CSSS00021525833			CC(C)(C)CN1CCC(C2CCNCC2)C1	CSSS00021525833
18592	CSSS00015926175	https://chem-space.com/CSSS00015926175			C1CN[C@H]2COC[C@H]2C1	CSSS00015926175
18593	CSSS00000685022	https://chem-space.com/CSSS00000685022			CNCC(=O)Nc1c(C)cc(C)cc1C.Cl	CSSS00000685022
18594	CSSS00027672546	https://chem-space.com/CSSS00027672546			Cc1ccc(C)c(C(=O)NC2CNCCC2C)c1.Cl	CSSS00027672546

18595 rows × 6 columns

Similar databases of amine fragments are available from other, smaller sources.

frags_file_path = 'http://fchgroup.net/files/fragments/FCHGroup_fragment-like_amines.zip'
sdf_fnames = extract_sdf_from_zip(frags_file_path, sdf_dir, download=True)

fch_sdf = PandasTools.LoadSDF(sdf_dir + sdf_fnames[0], removeHs=False)
fch_sdf['SMILES'] = [Chem.MolToSmiles(x) for x in fch_sdf['ROMol']]
fch_sdf['Name'] = [f"FCH_{i}" for i in range(fch_sdf.shape[0])]
fch_sdf

Extracted: FCHGroup_fragment-like_amines.sdf to ../../files/LNP_catalog/

	ID	URL	ROMol	SMILES	Name
0		https://chem-space.com/CSC000156186		Clc1ccc(CNc2cnccn2)cc1	FCH_0
1		https://chem-space.com/CSC000251315		CC(Nc1ncccn1)C1CC2CCC1C2	FCH_1
2		https://chem-space.com/CSC000156714		N#Cc1cccnc1NC1CCCC1	FCH_2
3		https://chem-space.com/CSC116272021		CC(C)(C)NCC(=O)NC(=O)NC1CCCCC1	FCH_3
4		https://chem-space.com/CSC116272401		COc1ccccc1C=CC1NC(=O)c2ccccc2N1	FCH_4
...	...	...	...	...	...
2150		https://chem-space.com/CSC116267824		O=C(NC[C@]12CNC[C@@]1(C(F)(F)F)C2)C1CCCC1	FCH_2150
2151		https://chem-space.com/CSC150710329		NC1CC12CC(NC(=O)c1cccc3c1CCN3)C2	FCH_2151
2152		https://chem-space.com/CSC116268621		Cc1cc(=O)[nH]c(NC2CCCc3c(C)cccc32)n1	FCH_2152
2153		https://chem-space.com/CSC150714667		C[C@H](NC1CCCC(F)(F)CC1)C(=O)N(C)C	FCH_2153
2154		https://chem-space.com/CSC150711165		Cn1cc(CN[C@@]23CCC[C@@H]2C3)ccc1=O	FCH_2154

2155 rows × 5 columns

Amines typically act as nucleophiles in organic chemistry and LNP chemistry, particularly when participating in synthesis. This means that the head fragments that scientists in LNP discovery typically care about are primary and secondary amines.

Primary amines

Primary amines are particularly useful reagents to have, so we devote special attention to their enumeration here for purposes of downstream use in other workflows. First, we gather a set of primary amines that are well studied as head groups in ionizable lipids, from a seminal line of work on short-RNA LNP delivery (Akinc et al. 2008).

import mols2grid

frags_whitehead_amines = ['NCCCCCO', 'CCNCCCNCC', 'CNCCOCCNC', 'CN(C)CCCN', 'CCN(CC)CCCN', 'NCCN(CCO)CCO', 'NCCN', 'NCCNCCO', 'NCCN(CCN)CCN', 'CN(CCN)CCN', 'CCN(CC)CCNCCN', 'N/C=C1\\\\CCCN1', 'NCCNCCNCCNCCNCCN', 'NCCCOCCOCCOCCCN', 'NCCOCCO', 'CC(C)C[C@H](N)CO', 'NCCNCCN1CCN(CCN)CC1', 'NCCN1CCNCC1', 'CC(N)CNCC(C)N', 'CCN(CCCNC)CCCNC', 'CN(C)CCCNCCCN', 'CNCCN(CCNC)CCNC', 'CC(CN)N', 'CN(CCCN)CCCN', 'C(CN)CN1CCN(CCCN)CC1', 'C1=CC(=CC(=C1)CN)CN', 'C1=CC2=C(C=C1N)OCCO2', 'COCCCN', 'C1COC(O1)CCN', 'C1OC(C)(C)OC1CN', 'COCCN', 'CCOCCN', 'COC(OC)CN', 'C1CC(OC1)CN', 'C(CO)CN', 'C[C@@H](O)CN', 'C[C@H](O)CCN', 'CC(CO)(CO)N', 'C(CCO)CN', 'C(COCCO)N', 'C(O)C(C)(C)CN', 'CCC(CO)(CO)N', 'C(CCCO)CCN', 'C1C[C@H](O)CC[C@H]1N', 'N(C)CCNC', 'N(CC)CCNCC', 'N(C(C)C)CCNC(C)C', 'N(CC)CCCNCC', 'N(C)CCOCCOCCNC', 'N1CCCNCC1', 'CCCN(CC)CN', 'C1CCN(C1)CCN', 'CN(C)CCN', 'CC(CN(C)C)N', 'C1CCN(CC1)CCN', 'C1COCCN1CCN', 'C1COCCN1CCCN', 'C1=CN(C=N1)CCCN', 'CNCCN', 'CNCCCN', 'CCCNCCN', 'C(CNCCNCCN)N', 'C(CN)CN', 'C(CCN)CN', 'CCCCNCCN', 'N(CCNCCO)CCO', 'CCCCCCCCCCCCCN', 'C(CN)CNCCN', 'CCCNCCCN', 'C(CCNCCCN)CNCCCN', 'CC(C)N(C(C)C)CCN', 'C(CNCCNCCNCCN)N', 'C(CN)CNCCN', 'C(CNCCN)N', 'CCCCCCCCCCCCN(CCN)CCN', 'C1CC[C@@H]([C@H](C1)N)N', 'C1CC[C@@H]([C@@H](C1)N)N', 'C1CCN(CC1)N', 'C1C=CC[C@H]([C@@H]1N)N', 'C(CNCCN)CNCCN', 'CCNCCN', 'C=CCCN', 'C1CNC[C@H]1N', 'C1CC(CC(C1)N)N', 'C1CNCCC1N', 'N1CCNCC1', 'N1CCNC[C@H]1C', 'CC1CCC(CC1)CCN', 'CCC1CCC(CC1)CCN', 'N1CCNCC1c1ccccc1', 'N1CCCNCCCNCCC1', 'N1CCNCCNCCNCC1', 'N1CCCNCCNCCNCC1', 'N1CCCNCCNCCCNCC1', 'N1(C)CCCNCCN(C)CCCNCC1', 'N1CCCNCCCNCCCNCC1', 'N1CCNCCNCCOCC1', 'C1C[C@H](CC[C@@H]1N)N', 'C1CC(CC(C1)CN)CN', 'C1=CC(=CC(=C1)N)N', 'C1=CC(=CC=C1N)N', 'CC1=CC=CC(=C1N)N', 'CC1=C(C=C(C=C1)N)N', 'CC1=C(C=CC=C1N)N', 'CC1=CC(=C(C=C1)N)N', 'C1=CC2=C(C(=C1)N)C(=CC=C2)N', 'N1C(C)NC(C)NC1C', 'C1(=NC(=NC(=N1)N)N)N', 'C1CC(CCC1CN)CN', 'C(CCCCN)CCCN', 'C(COCCOCCN)N', 'C(COCCOCCOCCN)N', 'C(CCOCCCN)COCCCN', 'COCCCN', 'N(C)CCOCCOCCNC', 'C(CO)N', 'N(CCNCCO)CCO', 'CCCN', 'CCCCN', 'CCCCCN', 'CCCCCCN', 'CNCCCCCCNC', 'CCCNCCCN', 'C(CN)CNCCCN', 'C(CCCNCCCCCCN)CCN', 'C(CN)CNCCCNCCCN', 'CC(C)[NH]', 'CC(C)(C)[NH]', 'CCC(CC)N', 'CCC(C)(C)N', 'CC(C)CN', 'CCC(C)N', 'CC(C)C(C)N', 'CCCC(C)N', 'CC(C)CCN', 'CC(C)(C)CCN', 'CC(C)(C)CC(C)(C)N', 'CCC(C)CN', 'CCCCC(CC)CN', 'CC(CO)N', 'CC(C)C[C@@H](CO)N', 'CC(C)(C)C(CO)N', 'C(CO)(CO)(CO)N', 'CC(C)OCCCN', 'N(C(C)(C)C)CCNC(C)(C)C', 'C(CCN)CCN', 'C(CCCN)CCN', 'C(CCCCN)CCN', 'C(CCCCCN)CCN', 'C(CCCCCCCN)CCN', 'C(CCCCCCCCCN)CCN', 'CC(CC(C)(C)CCN)CN', 'C(CCCN(C)Cc1ccccc1)N', 'CN1CCN(CC1)CCCCN', 'CCN1CCC(CC1)N', 'CN1CCN(CC1)CCN', 'CN1CCOCC1CN', 'CCCN1CCC(CC1)N', 'C(CN(C)Cc1ccccc1)N', 'CN(C)CCOCCN', 'C1CCN(CC1)CCOCCN', 'C1CCN(C1)CCOCCN', 'CN(C1=CC=CC=C1)C(=O)CN', 'CN1C=CN=C1CN', 'CN1CCCC1CN', 'CN1CCCC1C(=O)N', 'CN1CCC(CC1)CN', 'CN1CCC(CC1)N', 'CC(C)N1CCN(CC1)CCCN', 'C1CN(CCN1)C(=O)CCN', 'C(C)(CN(C)C)(C)CN', 'CC(C)NCCN', 'C1[CH]C1N', 'C1CC(C1)N', 'C1CCC(C1)N', 'CCN1CCC[C@H]1CN', 'CCN1CCC[C@@H]1CN', 'C1CCC(CC1)N', 'C1CCC(CC1)CN', 'CC(C)(CN(C)C)CN', 'C1CC2CC1CC2N', 'C1CCCC(CCC1)N', 'C1C2CC3CC1CC(C2)(C3)N', 'C1C2CC3CC1CC(C2)(C3)CN', 'C1=CC=C(C=C1)CCCN', 'C1=CC=C(C=C1)CCCCN', 'CC(C)(C)C1=CC=C(C=C1)N', 'CC(C)(C)C1=CC=CC=C1N', 'CCC1=C(C(=CC=C1)CC)N', 'COC1=CC=CC=C1N', 'COC1=CC=CC=C1CN', 'CC1=CC=C(C=C1)C(C)N', 'CC(C)C1=CC=CC=C1N', 'CC(C)C1=CC=C(C=C1)N', 'CC(C)OC1=CC=CC(=C1)N', 'COC1=CC=C(C=C1)CCN', 'CCOC1=C(C=CC=C1)N', 'CCCCCCCCOC1=CC=C(C=C1)N', 'CCOC1=CC=CC=C1CN', 'CC(C)(C)C1=CC(=C(C=C1)C(C)(C)C)N', 'COC1=CC(=C(C=C1)OC)N', 'CC(C)(C)c1ccc(OC)c(c1)N', 'C(Cc1cc(OC)c(OC)cc1)N', 'C(c1cc(C)c(c(OC)c1)OC)N', 'C1=CC=C(C=C1)NCCN', 'C1=CC=C(C=C1)CNCCN', 'CNC1=CC=CC=C1N', 'C1=CC(=CC(=C1)N)CN', 'Cc1c(N)c(C)c(C)c(c1C)N', 'C1=CC=C(C(=C1)CN)N', 'CC1=CC(=C(C=C1C)N)N', 'c1(F)cc(ccc1)N', 'c1(F)ccc(cc1)N', 'C1=CC(=C(C=C1)F)CN', 'C(c1cc(F)ccc1)N', 'C1=CC(=C(C=C1)F)CCN', 'c1(cc(F)ccc1F)N', 'C1=C(C=C(C(=C1)F)F)N', 'C1=C(C=C(C=C1F)F)N', 'C(c1cc(F)ccc1F)N', 'C1=C(C=C(C(=C1F)F)F)N', 'c1(F)cc(F)c(c(F)c1)N', 'C(F)(F)(F)c1cc(ccc1)N', 'C(F)(F)(F)c1c(ccc(c1)F)N', 'CC1=CC=C(C=C1F)N', 'COC1=CC=C(C=C1F)N', 'c1(OC(F)F)ccc(cc1)N', 'C(F)(F)(F)Oc1ccc(cc1)N', 'CC1CC(CCC1N)CC2CCC(C(C2)C)N', 'N1C2NCCNC2NCC1', 'C1CC(C2=C1C=CC=C2)N', 'C1=CC2=C(C=C1)CC(C2)N', 'C1CC2=C(C1)C(=CC=C2)N', 'C1OC2=CC=C(C=C2O1)N', 'C1CCC2=C(C1)C=CC(=C2)N', 'C1C2=C(C=CC(=C2)N)C3=C1C=C(C=C3)N', 'c1(cc2cc3ccccc3cc2cc1)N', 'C1=CC=C2C(=C1)C3=C(C=CC=C3)C(=C2N)N', 'C1=CC=C(C=C1)C2=CC(=CC=C2)N', 'C1=CC=C(C=C1)C2=CC=C(C=C2)N', 'Cc1c(N)ccc(c1)-c1ccc(c(C)c1)N', 'C1=C(C=C(C(=C1)N)N)C2=CC=C(C(=C2)N)N', 'Cc1c(N)c(C)cc(c1)-c1cc(C)c(c(c1)C)N', 'C1=CC=C(C=C1)CC2=CC=CC=C2N', 'C1=CC(=CC=C1CC2=CC=C(C=C2)N)N', 'CNC1=CC=C(C=C1)CC2=CC=C(C=C2)N', 'C1=CC=C(C=C1)C(C2=CC=CC=C2)N', 'C(CC(c1ccccc1)c1ccccc1)N', 'C1=CC=C(C=C1)OC2=CC=CC=C2N', 'C1=CC=C(C=C1)OC2=CC=C(C=C2)N', 'c1(Oc2ccc(F)cc2)ccc(cc1)N', 'C1=CC(=CC=C1N)OC2=CC=C(C=C2)N', 'C1=CC=C(C=C1)COC2=CC(=CC=C2)N', 'C1=CC=C(C=C1)NC2=CC=C(C=C2)N', 'C1=CC=C(C=C1)NC2=CC=CC=C2N', 'C1=CC=C(C=C1)[C@H]([C@H](C2=CC=CC=C2)N)N', 'C1=CC=C(C=C1)C(C2=CC=CC=C2)(C3=CC=CC=C3)N', 'Nc1c(cc(cc1-c1ccccc1)-c1ccccc1)-c1ccccc1', 'C1=CC=C2C(=C1)C=CC(=C2C3=C(C=CC4=CC=CC=C43)N)N', 'N(c1ccccc1)c1ccc(cc1)-c1ccc(cc1)Nc1ccccc1', 'C1(c2c(cccc2)-c2ccccc21)(c1ccc(N)cc1)c1ccc(cc1)N', 'CC(NCCN(CCNC(C)C)CCNC(C)C)C']
frags_whitehead_amines = np.unique([Chem.CanonSmiles(x) for x in frags_whitehead_amines])
frags_whitehead_df = pd.DataFrame({
    'SMILES': frags_whitehead_amines, 
    'Name': [f"Whitehead_{i}" for i in range(len(frags_whitehead_amines))]
})

print("{} heads from Whitehead et al. 2014 library.".format(len(frags_whitehead_amines)))
mols2grid.display([Chem.MolFromSmiles(x) for x in frags_whitehead_amines],mol_col="mol")

252 heads from Whitehead et al. 2014 library.

The Descriptor Libraries Project from the Molecular Sciences Software Institute (MolSSI) runs quantum mechanical calculations on a subset of individual fragments. These are usefully sorted by molecule class. The primary amines are displayed below.

molssi_amines_primary_df = pd.read_excel("https://descriptor-libraries.molssi.org/primary-amines/content/primary-amines_library.xlsx", index_col=0)
molssi_amines_primary_df

	Canonical_SMILES	Compound_Name	Reaxys	Drug_Repurposing_Hub	Enamine	Notes
ID
1	COCCN	N1	1.0	NaN	NaN	NaN
2	NCC1CCCCC1	N2	1.0	NaN	NaN	NaN
3	NCCc1ccccc1Cl	N3	1.0	NaN	NaN	NaN
4	NCc1ccc(-c2ccccc2)cc1	N4	1.0	NaN	NaN	NaN
5	NC1CCOCC1	N5	1.0	NaN	NaN	NaN
...	...	...	...	...	...	...
4268	CN1CCC(N)CC1	N-TAK-901	NaN	1.0	NaN	NaN
4269	NC[C@@H]1C[C@H]1c1cccc2c1CCO2	N-tasimelteon	NaN	1.0	NaN	NaN
4270	CN1CCN(CCCN)CC1	N-TEN-010	NaN	1.0	NaN	NaN
4271	NC1CCC1	N-VU0360172	NaN	1.0	NaN	NaN
4272	CCC[C@H](N)c1ccccc1	N-WP1130	NaN	1.0	NaN	NaN

4272 rows × 6 columns

Secondary amines

Secondary amines typically offer less scope for versatile modification in syntheses, since they are already participating in a covalent bond with a carbon. MolSSI has a library of secondary amines that can be used for this purpose.

molssi_amines_secondary_df = pd.read_excel("https://descriptor-libraries.molssi.org/secondary-amines/content/secondary-amines_library.xlsx", index_col=0)
molssi_amines_secondary_df

	Canonical_SMILES	Compound_Name	Reaxys	Drug_Repurposing_Hub	Enamine	Notes
ID
1	C1COCCN1	SecN1	1.0	NaN	NaN	NaN
2	C1CCNCC1	SecN2	1.0	NaN	NaN	NaN
3	C1CCNC1	SecN3	1.0	NaN	NaN	NaN
4	CNC	SecN4	1.0	NaN	NaN	NaN
5	CNCCCN1c2ccccc2CCc2ccccc21	SecN5	1.0	NaN	NaN	NaN
...	...	...	...	...	...	...
3845	CN1CC2(CCNC2)C1	SecNEnamine5	NaN	NaN	1.0	NaN
3846	C1COCC2(C1)CNC2	SecNEnamine6	NaN	NaN	1.0	NaN
3847	C1CC(C2CNCCO2)C1	SecNEnamine7	NaN	NaN	1.0	NaN
3848	CC1(C)CNCCS1(=O)=O	SecNEnamine8	NaN	NaN	1.0	NaN
3849	CC1(C)CNCC2(CCOC2)O1	SecNEnamine9	NaN	NaN	1.0	NaN

3849 rows × 6 columns

Consolidating amine fragments

Taken together, these represent a hefty cross-section of amine fragments in commercial use. There is overlap with manufacturers’ databases, such as those of Enamine and WuXi. With the right access permissions, those too can be mined as necessary.

Having gathered this set, it’s useful to inspect these to get an idea of the staggering variety of drug-like amine fragments available (primarily because of their use in small-molecule drug design over the years).

smiles_list = []
name_list = []

molssi_amines_primary_df.rename(columns={'Canonical_SMILES': 'SMILES', 'Compound_Name': 'Name'}, inplace=True)
molssi_amines_secondary_df.rename(columns={'Canonical_SMILES': 'SMILES', 'Compound_Name': 'Name'}, inplace=True)

for new_df in [frags_whitehead_df, molssi_amines_primary_df, molssi_amines_secondary_df, chemspace_sdf, fch_sdf]:
    new_smiles_list = [Chem.CanonSmiles(x) for x in new_df['SMILES'].values]
    new_names_list = [f"{new_df['Name'].values[i]}" for i in range(new_df.shape[0])]
    for i in range(new_df.shape[0]):
        if new_smiles_list[i] in smiles_list:
            ndx = np.where(np.array(smiles_list) == new_smiles_list[i])[0][0]
            name_list[ndx] = name_list[ndx] + "|" + new_names_list[i]
        else:
            smiles_list.append(new_smiles_list[i])
            name_list.append(new_names_list[i])

frags_amines_df = pd.DataFrame({
    'SMILES': smiles_list,
    'Name': name_list
})

frags_amines_df

	SMILES	Name
0	C1CNC2NCCNC2N1	Whitehead_0
1	C1CNCCCNCCCNC1	Whitehead_1
2	C1CNCCCNCCNCCCNC1	Whitehead_2
3	C1CNCCN1	Whitehead_3
4	C1CNCCNC1	Whitehead_4|CSSS00000210015
...	...	...
26417	Fc1ccc(Nc2nnc(C3CC3)[nH]2)cc1	FCH_2137
26418	Cc1ccsc1CNC1CC[C@H]2CN(C)C[C@@H]12	FCH_2138
26419	NC1CC12CC(NC(=O)c1cccc3c1CCN3)C2	FCH_2151
26420	Cc1cc(=O)[nH]c(NC2CCCc3c(C)cccc32)n1	FCH_2152
26421	C[C@H](NC1CCCC(F)(F)CC1)C(=O)N(C)C	FCH_2153

26422 rows × 2 columns

amines_all = set(frags_amines_df["SMILES"])
print(f"Number of amine fragments: {len(amines_all)}")

display_df = pd.DataFrame(list(amines_all), columns=["SMILES"])
mols2grid.display(display_df)

Number of amine fragments: 26422