Awesome

Python_SmallWorld_API

An unofficial Python3 module to query a SmallWorld chemical space search server.

Disclaimer

This is Unofficial So please do not abuse it or use it when you cannot legally use the site!

SmallWorld is a search engine for chemical space developed by John Mayfield and Roger Sayle at NextMove Software. John Irwin and Brian Shoichet at UCSF (the creators and maintainers of the ZINC database), host a version of it at sw.docking.org along with another NextMove Software product, Arthor.

Overview

SmallWorld allows one to search for similar compounds to a give SMILES in one of many databases —a very complex feat.

A copy is hosted at sw.docking.org by John Irwin, Brian Shoichet and co. This is a free service, but it is not intended for heavy use. To change the endpoint, one can change the class attribute SmallWorld.base_url to a different URL. The folk at NextMove Software deploy instances of it for paying customers —and with full support.

The API points of the site are described in wiki.docking.org/index.php/How_to_use_SmallWorld_API.

This Python3 module allows one to search it.

For searches in Arthor, Zinc and EnamineStore see elsewhere.

Install

pip install -q smallworld-api

Usage

The following searches for Aspirin in Enamine's make-on-demand space, Enamine REAL, which does not contain it as the latter is filtered by Lipinski's rule of five (Aspirin is actually a terrible placeholder drug)

from rdkit import Chem
from rdkit.Chem import PandasTools
import pandas as pd  # for typehinting below

from smallworld_api import SmallWorld

print(SmallWorld.base_url)  # 'https://sw.docking.org'
aspirin = 'O=C(C)Oc1ccccc1C(=O)O'
sw = SmallWorld()
results : pd.DataFrame = sw.search(aspirin, dist=5, db=sw.REAL_dataset)

from IPython.display import display
display(results)

The first two import lines are optional as the code works without rdkit. If pandas gets imported before PandasTools and Chem imported not in main then display issues happen, which can be fixed with a from rdkit.Chem.Draw import IPythonConsole.

So it's up to you to remember to run:

PandasTools.AddMoleculeColumnToFrame(results, 'smiles', 'molecule', includeFingerprints=True)

The argument db for .search is a string and is the name of the database. These do seem to change, so they get updated during initialisation or with the call:

dbs: pd.DataFrame = SmallWorld.retrieve_databases()  #: pd.DataFrame (.db_choices gets updated too)

The dynamic properties .REAL_dataset and .ZINC_dataset simply return the best value from the presets, which may have become out of date (unless updated).

Query terms

The first argument passed to .search can be:

• a str (SMILES)
• a Chem.Mol (rdkit is an optional requirement though)
• a list-like (sequence) or a dict-like (mapping) of the above, where the index or key becomes the name in the output table.

See the class attribute dictionary SmallWorld.default_submission for what the defaults are set to, which ought to be:

{'dist': 8, 'tdn': 6, 'rdn': 6, 'rup': 2, 'ldn': 2, 'lup': 2, 'maj': 6, 'min': 6, 'sub': 6, 'sdist': 12, 'tup': 6, 'scores': 'Atom Alignment,ECFP4,Daylight'}

If one is sure that the correct dataset is being used and any raised NoMatchError is due to the SMILES, then once can add for the last case the argument tolerate_tolerate_NoMatchError=True, which makes them ignored bar for a warning.

Debug

The instantiation is set up so for debugging, namely it has two attributes of interest:

• sw.last_reply, a requests.Response instance
• sw.hit_list_id an integer representing the search (AKA. hlid in the server responses)

The errors raised are generally either requests.HTTPError or smallworld.NoMatchError. The former is raised by a requests.Response.raise_for_status call and means there is a status code that isn't 200, the latter is raised by one of the various checks in sw.get_results().

For the former errors, i.e. those by a serverside-declared HTML-formatting error (eg. status code 404), if one is in a Jupyter notebook one can do sw.show_reply_as_html(). Generally if you get status code 500, it is best to try again tomorrow as the server is having a hard time and is probably not okay on the web.

For the latter, the result in .last_reply should be a JSON string, therefore should give something like this:

reply_data: dict = sw.last_reply.json()

A common issue is the change in database names, therefore do do and pick a different one (ATM, the index of the dataframe is the name to use, but in 2021 it was the name)

from IPython.display import display

from smallworld_api import SmallWorld
db_table : pd.DataFrame = SmallWorld.retrieve_databases()
display( db_table )

There will be a "ground control to major Tom" warning in the first query. This weird reply means that the stream has finished, but not closed or something. Ignore it.

Also, as a shorthand, mol: Chem.Mol = SmallWorld.check_smiles(aspirin) can be called to check if the molecules is fine.

For extreme debugging, open Chrome and go to sw.docking.org and open the developer tools (F12). Then go to the Network tab and do a search, eg. with CC2=CC=C(CNC(=O)C1CCC1)C=N2, this will be populated by all the figure requests, but /search/submit will be the first one to look at if the issue is with the submit method in the trace, /search/view if it's with the get_results method. Then simply copy the url query off the request and use it as parameters or compare them etc. For example:

import urllib.parse
from smallworld_api import SmallWorld

url_query = '👾=👾&👾=👾'
expected = dict(urllib.parse.parse_qsl(url_query))

class Debug(SmallWorld):
    
    def submit_query(self, params):
        # override the method to check the parameters
        print('Going to use:', params)
        print('Missing:', set(expected).difference(params))
        return super().submit_query(expected)

d = Debug()
d.search('CCO', db=d.REAL_dataset)

If it's a field that change, raise an issue and I'll update the class or do a pull request :pray:.

Choices

The database choices can seen with the preset list SmallWorld.db_choices. But also this can be recached via the classmethod SmallWorld.retrieve_databases().

Two databases, REAL_Space_21Q3_2B(public) and REAL_DB_20Q2, are Enamine REAL databases (aka. Enamine will make the compound on request). Previously, the repository, enamine-real-search was good for this, but unfortunately Enamine changed their endpoints. So I wrote this to take its place! Despite the smaller number of entries, REAL_DB_20Q2 gives the most hits and is less likely to "Major Tom out".

Likewise, the attribute SmallWorld.sf_choices (type list) and the classmethod SmallWorld.retrieve_scorefun_options() do the same. The values are less and are: ['Atom Alignment', 'SMARTS Alignment', 'ECFP4', 'Daylight'], but these are activated by default and will be visible as columns in the resulting dataframe from a search call.

Here is the full list of databases:

import pandas as pd

choices: pd.DataFrame = SmallWorld.retrieve_databases()

display(choices)

Which will return (as of writing on the 9th Dec 2021):

Names

There is a Python module called smallworld, which implements the small world algorithm. This is not an API to the sw.docking.org site.

The blog of the sw.docking.org site mentions a pysmallworld. There is no mention of this in Google so I am guessing it is for a future feature? I however need to use this now as I need it as a publicly usable example workflow of Fragmenstein.

Also, there is a great and wacky boardgame called Small World, with a curious/agonising dynamic which forces you to not be a collector.