The RDKit database cartridge

What is this?

This document is a tutorial and reference guide for the RDKit PostgreSQL cartridge.

If you find mistakes, or have suggestions for improvements, please either fix them yourselves in the source document (the .md file) or send them to the mailing list: rdkit-discuss@lists.sourceforge.net (you will need to subscribe first)

Tutorial

Introduction

Creating databases

Configuration

The timing information below was collected on a commodity desktop PC (Dell Studio XPS with a 2.9GHz i7 CPU and 8GB of RAM) running Ubuntu 12.04 and using PostgreSQL v9.1.4. The database was installed with default parameters.

To improve performance while loading the database and building the index, I changed a couple of postgres configuration settings in postgresql.conf :

synchronous_commit = off      # immediate fsync at commit
full_page_writes = off            # recover from partial page writes

And to improve search performance, I allowed postgresql to use more memory than the extremely conservative default settings:

shared_buffers = 2048MB           # min 128kB
                  # (change requires restart)
work_mem = 128MB              # min 64kB

Creating a database from a file

In this example I show how to load a database from the SMILES file of commercially available compounds that is downloadable from emolecules.com at URL http://www.emolecules.com/doc/plus/download-database.php

If you choose to repeat this exact example yourself, please note that it takes several hours to load the 6 million row database and generate all fingerprints.

First create the database and install the cartridge:

~/RDKit_trunk/Data/emolecules > createdb emolecules
~/RDKit_trunk/Data/emolecules > psql -c 'create extension rdkit' emolecules

Now create and populate a table holding the raw data:

~/RDKit_trunk/Data/emolecules > psql -c 'create table raw_data (id SERIAL, smiles text, emol_id integer, parent_id integer)' emolecules
NOTICE:  CREATE TABLE will create implicit sequence "raw_data_id_seq" for serial column "raw_data.id"
CREATE TABLE
~/RDKit_trunk/Data/emolecules > zcat emolecules-2013-02-01.smi.gz | sed '1d; s/\\/\\\\/g' | psql -c "copy raw_data (smiles,emol_id,parent_id) from stdin with delimiter ' '" emolecules

Create the molecule table, but only for SMILES that the RDKit accepts:

~/RDKit_trunk/Data/emolecules > psql emolecules
psql (9.1.4)
Type "help" for help.
emolecules=# select * into mols from (select id,mol_from_smiles(smiles::cstring) m from raw_data) tmp where m is not null;
WARNING:  could not create molecule from SMILES 'CN(C)C(=[N+](C)C)Cl.F[P-](F)(F)(F)(F)F'
... a lot of warnings deleted ...
SELECT 6008732
emolecules=# create index molidx on mols using gist(m);
CREATE INDEX

The last step is only required if you plan to do substructure searches.

Loading ChEMBL

Start by downloading and installing the postgresql dump from the ChEMBL website ftp://ftp.ebi.ac.uk/pub/databases/chembl/ChEMBLdb/latest

Connect to the database, install the cartridge, and create the schema that we’ll use:

chembl_14=# create extension if not exists rdkit;
chembl_14=# create schema rdk;

Create the molecules and build the substructure search index:

chembl_14=# select * into rdk.mols from (select molregno,mol_from_ctab(molfile::cstring) m  from compound_structures) tmp where m is not null;
SELECT 1210823
chembl_14=# create index molidx on rdk.mols using gist(m);
CREATE INDEX
chembl_14=# alter table rdk.mols add primary key (molregno);
NOTICE:  ALTER TABLE / ADD PRIMARY KEY will create implicit index "mols_pkey" for table "mols"
ALTER TABLE

Create some fingerprints and build the similarity search index:

chembl_14=# select molregno,torsionbv_fp(m) as torsionbv,morganbv_fp(m) as mfp2,featmorganbv_fp(m) as ffp2 into rdk.fps from rdk.mols;
SELECT 1210823
chembl_14=# create index fps_ttbv_idx on rdk.fps using gist(torsionbv);
CREATE INDEX
chembl_14=# create index fps_mfp2_idx on rdk.fps using gist(mfp2);
CREATE INDEX
chembl_14=# create index fps_ffp2_idx on rdk.fps using gist(ffp2);
CREATE INDEX
chembl_14=# alter table rdk.fps add primary key (molregno);
NOTICE:  ALTER TABLE / ADD PRIMARY KEY will create implicit index "fps_pkey" for table "fps"
ALTER TABLE

Here is a group of the commands used here (and below) in one block so that you can just paste it in at the psql prompt:

create extension if not exists rdkit;
create schema rdk;
select * into rdk.mols from (select molregno,mol_from_ctab(molfile::cstring) m  from compound_structures) tmp where m is not null;
create index molidx on rdk.mols using gist(m);
alter table rdk.mols add primary key (molregno);
select molregno,torsionbv_fp(m) as torsionbv,morganbv_fp(m) as mfp2,featmorganbv_fp(m) as ffp2 into rdk.fps from rdk.mols;
create index fps_ttbv_idx on rdk.fps using gist(torsionbv);
create index fps_mfp2_idx on rdk.fps using gist(mfp2);
create index fps_ffp2_idx on rdk.fps using gist(ffp2);
alter table rdk.fps add primary key (molregno);
create or replace function get_mfp2_neighbors(smiles text)
returns table(molregno integer, m mol, similarity double precision) as

$$ select molregno,m,tanimoto_sml(morganbv_fp(mol_from_smiles($1::cstring)),mfp2) as similarity from rdk.fps join rdk.mols using (molregno) where morganbv_fp(mol_from_smiles($1::cstring))%mfp2 order by morganbv_fp(mol_from_smiles($1::cstring))<%>mfp2; $$ language sql stable ;

Substructure searches

Example query molecules taken from the eMolecules home page:

chembl_14=# select count(*) from rdk.mols where m@>'c1cccc2c1nncc2' ;
 count
-------
   281
(1 row)

Time: 184.043 ms
chembl_14=# select count(*) from rdk.mols where m@>'c1ccnc2c1nccn2' ;
 count
-------
   671
(1 row)

Time: 449.998 ms
chembl_14=# select count(*) from rdk.mols where m@>'c1cncc2n1ccn2' ;
 count
-------
   930
(1 row)

Time: 568.378 ms
chembl_14=# select count(*) from rdk.mols where m@>'Nc1ncnc(N)n1' ;
 count
-------
  4478
(1 row)

Time: 721.758 ms
chembl_14=# select count(*) from rdk.mols where m@>'c1scnn1' ;
 count
-------
 10908
(1 row)

Time: 701.036 ms
chembl_14=# select count(*) from rdk.mols where m@>'c1cccc2c1ncs2' ;
 count
-------
 12823
(1 row)

Time: 1585.473 ms
chembl_14=# select count(*) from rdk.mols where m@>'c1cccc2c1CNCCN2' ;
 count
-------
  1155
(1 row)

Time: 4567.222 ms

Notice that the last two queries are starting to take a while to execute and count all the results.

Given we’re searching through 1.2 million compounds these search times aren’t incredibly slow, but it would be nice to have them quicker.

One easy way to speed things up, particularly for queries that return a large number of results, is to only retrieve a limited number of results:

chembl_14=# select * from rdk.mols where m@>'c1cccc2c1CNCCN2' limit 100;
 molregno |                                                                                      m
----------+------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
  1292129 | Cc1ccc2c(c1)C(=O)N(N(C)C)CC(=O)N2
  1013311 | CCCCC(=O)N1CC(=O)Nc2ccc(F)cc2C1c1ccccc1
  1294754 | COc1cc2c(cc1OCc1ccccc1)NC(=O)[C@@H]1CCCN1C2=O
  1012025 | O=C(c1cc2ccccc2oc1=O)N1CC(=O)Nc2ccc(Br)cc2C1c1ccc(F)cc1
   995226 | CC1Cc2ccccc2N1C(=O)CN1c2ccccc2C(=O)N(C)CC1=O
  1291875 | COC(=O)C1=NN2c3ccccc3CN([C@@H](C)c3ccccc3)C(=O)[C@@H]2[C@H]1c1ccccc1
  ...
  1116370 | COc1ccc(CC(=O)N2CC(=O)Nc3ccc(Br)cc3C2c2ccc(F)cc2)cc1OC
  1114872 | O=C1[C@@H]2[C@H](C(=O)N1Cc1ccccc1)[C@@H]1C(=O)Nc3ccccc3C(=O)N1[C@@H]2c1ccccc1
Time: 375.747 ms

SMARTS-based queries

Oxadiazole or thiadiazole:

chembl_14=# select * from rdk.mols where m@>'c1[o,s]ncn1'::qmol limit 500;
 molregno |                                                                      m
----------+----------------------------------------------------------------------------------------------------------------------------------------------
   534296 | Clc1ccccc1CNc1noc(-c2sccc2Br)n1
     1178 | CCCCc1oc2ccccc2c1Cc1cccc(/C(C)=C/Cn2oc(=O)[nH]c2=O)c1
   566382 | COC(=O)CCc1nc(C2CC(c3ccc(O)c(F)c3)=NO2)no1
   499261 | CS/C=C(/C)n1c(=O)onc1C(=O)c1ccc(Br)cc1
   450499 | CS(=O)(=O)c1ccc(Nc2ncnc(N3CCC(c4nc(-c5cccc(C(F)(F)F)c5)no4)CC3)c2[N+](=O)[O-])cc1
   600176 | Cc1nc(-c2c(Cl)cc(Cl)cc2-c2cnc([C@@H](C)NC(=O)N(C)O)c(F)c2)no1
     1213 | CC/C(=C\Cn1oc(=O)[nH]c1=O)c1cccc(OCc2nc(-c3ccc(C(F)(F)F)cc3)oc2C)c1
   659277 | Cn1c(N)c(CCCN)c[n+]1CC1=C(C(=O)O)N2C(=O)[C@@H](NC(=O)/C(=N\OC(C)(C)C(=O)O)c3nsc(N)n3)[C@H]2SC1
     1316 | CCCCCCCC/C(=C\Cn1oc(=O)[nH]c1=O)c1cccc(OCc2nc(-c3ccc(C(F)(F)F)cc3)oc2C)c1
   ...
     1206 | C/C(Cn1oc(=O)[nH]c1=O)=C(/C)c1cccc(OCc2nc(-c3ccc(C(F)(F)F)cc3)oc2C)c1
     1496 | Cc1oc(-c2ccccc2)nc1COc1cccc(C#CC(C)n2oc(=O)[nH]c2=O)c1
Time: 3365.309 ms

This is slower than the pure SMILES query, this is generally true of SMARTS-based queries.

Using Stereochemistry

Note that by default stereochemistry is not taken into account when doing substructure queries:

chembl_14=# select * from rdk.mols where m@>'NC(=O)[C@@H]1CCCN1C=O' limit 10;
 molregno |                                                                                        m
----------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
  1295889 | COc1ccc(C[C@@H](C(=O)NCC(N)=O)N(C)C(=O)[C@@H]2CCCN2C(=O)[C@H](CC(C)C)NC(=O)C(C)NC(=O)OCc2ccccc2)cc1
  1293815 | CN1C(=O)C23CC4=CC=CC(O)C4N2C(=O)C1(CO)SS3
  1293919 | CNC(=O)CNC(=O)C(NC(=O)CNC(=O)C1CCCN1C(=O)C(C)NC(=O)C(NC(=O)OC(C)(C)C)C(C)C)C(C)C
  1011887 | COC(=O)C(C)NC(=O)C1CCCN1C(=O)CNC(=O)OCc1ccccc1
  1293021 | CCC(C)C1NC(=O)C(NC(=O)C(CC(C)C)N(C)C(=O)[C@@H]2CC(O)CN2C(=O)[C@H](C)O)C(C)OC(=O)[C@H](Cc2ccc(OC)cc2)N(C)C(=O)[C@@H]2CCCN2C(=O)[C@H](CC(C)C)NC(=O)C(C)C(=O)[C@H](C(C)C)OC(=O)CC1O
  1287353 | CCC(C)C1NC(=O)C(NC(=O)C(CC(C)C)N(C)C(=O)C2CCCN2C(=O)C(C)O)C(C)OC(=O)C(Cc2ccc(OC)cc2)N(C)C(=O)C2CCCN2C(=O)C(CC(C)C)NC(=O)[C@H](C)C(=O)[C@H](C(C)C)OC(=O)CC1O
  1293647 | CCC(C)[C@@H]1NC(=O)[C@@H]2CCCN2C(=O)C(CC(O)CCl)OC(=O)CCNC(=O)[C@H](C)N(C)C(=O)[C@H](C(C)C)N(C)C1=O
  1290320 | C=CCOC(=O)[C@@H]1C[C@@H](OC(C)(C)C)CN1C(=O)[C@@H]1[C@H]2OC(C)(C)O[C@H]2CN1C(=O)OCC1c2ccccc2-c2ccccc21
  1281392 | COC1=CC2C(=O)N(C)[C@@H](C)C(=O)N3NCCC[C@@H]3C(=O)N3[C@@H](C[C@@]4(O)c5ccc(Cl)cc5N[C@@H]34)C(=O)N[C@H](C(C)C)C(=O)N3NCCC[C@@H]3C(=O)N2N=C1
  1014237 | CC(C)COC(=O)N1CC(O)CC1C(=O)Nc1ccc2c(c1)OCO2
(10 rows)

Time: 9.447 ms

This can be changed using the rdkit.do_chiral_sss configuration variable:

chembl_14=# set rdkit.do_chiral_sss=true;
SET
Time: 0.241 ms
chembl_14=# select * from rdk.mols where m@>'NC(=O)[C@@H]1CCCN1C=O' limit 10;
 molregno |                                                                                                                                                                                                                                                                                 m
----------+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
  1295889 | COc1ccc(C[C@@H](C(=O)NCC(N)=O)N(C)C(=O)[C@@H]2CCCN2C(=O)[C@H](CC(C)C)NC(=O)C(C)NC(=O)OCc2ccccc2)cc1
  1293021 | CCC(C)C1NC(=O)C(NC(=O)C(CC(C)C)N(C)C(=O)[C@@H]2CC(O)CN2C(=O)[C@H](C)O)C(C)OC(=O)[C@H](Cc2ccc(OC)cc2)N(C)C(=O)[C@@H]2CCCN2C(=O)[C@H](CC(C)C)NC(=O)C(C)C(=O)[C@H](C(C)C)OC(=O)CC1O
  1293647 | CCC(C)[C@@H]1NC(=O)[C@@H]2CCCN2C(=O)C(CC(O)CCl)OC(=O)CCNC(=O)[C@H](C)N(C)C(=O)[C@H](C(C)C)N(C)C1=O
  1290320 | C=CCOC(=O)[C@@H]1C[C@@H](OC(C)(C)C)CN1C(=O)[C@@H]1[C@H]2OC(C)(C)O[C@H]2CN1C(=O)OCC1c2ccccc2-c2ccccc21
  1281392 | COC1=CC2C(=O)N(C)[C@@H](C)C(=O)N3NCCC[C@@H]3C(=O)N3[C@@H](C[C@@]4(O)c5ccc(Cl)cc5N[C@@H]34)C(=O)N[C@H](C(C)C)C(=O)N3NCCC[C@@H]3C(=O)N2N=C1
  1007418 | C/C=C\C=C\C(=O)N1CC2(CC(c3cccc(NC(=O)/C=C\C=C/C)c3)=NO2)C[C@H]1C(N)=O
   785530 | C/C=C/C(=O)N1CC2(CC(c3cccc(NC(=O)CC)c3)=NO2)C[C@H]1C(N)=O
  1292152 | CCCCCCCC(=O)N[C@H](C(=O)N[C@H](C(=O)N(C)[C@H](C(=O)N1CCC[C@H]1C(=O)N(C)[C@H](C)C(=O)NCc1ccc(OC)cc1OC)C(C)C)C(C)C)C(C)C
  1281390 | CC(C)[C@@H]1NC(=O)[C@@H]2C[C@@]3(O)c4ccc(Cl)cc4N[C@H]3N2C(=O)[C@H]2CCCNN2C(=O)[C@@H](C)N(C)C(=O)[C@H]2CCCNN2C(=O)[C@@H]2CCCNN2C1=O
  1057962 | CC[C@H](C)[C@@H]1NC(=O)[C@H](CCCNC(=N)N)NC(=O)[C@H](CC(=O)O)NC(=O)[C@H](CCSC)NC(=O)[C@H](CCCCN)NC(=O)[C@H](CCCNC(=N)N)NC(=O)CNC(=O)[C@H](Cc2ccccc2)NC(=O)[C@@H](NC(=O)CNC(=O)[C@H](CO)NC(=O)CNC(=O)[C@H](CCC(N)=O)NC(=O)[C@@H](NC(=O)[C@H](CCSC)NC(=O)[C@H](CCCCN)NC(=O)[C@@H]2CCCN2C(=O)[C@@H](N)CO)C(C)C)CSSC[C@@H](C(=O)N[C@@H](CCCCN)C(=O)N[C@H](C(=O)N[C@@H](CC(C)C)C(=O)N[C@@H](CCCNC(=N)N)C(=O)N[C@@H](CCCNC(=N)N)C(=O)N[C@@H](Cc2cnc[nH]2)C(=O)O)C(C)C)NC(=O)CNC(=O)[C@H](CC(C)C)NC(=O)CNC(=O)[C@H](CO)NC(=O)[C@H](CO)NC(=O)[C@H](CO)NC(=O)[C@H](CO)NC1=O
(10 rows)

Time: 35.383 ms

Tuning queries

It is frequently useful to be able to exert a bit more control over substructure queries without having to construct complex SMARTS queries. The cartridge function mol_adjust_query_properties() can be used to do just this. Here is an example of the default behavior, using aquery for 2,6 di-substituted pyridines:

chembl_21=# select molregno,m from rdk.mols where m@>mol_adjust_query_properties('*c1cccc(NC(=O)*)n1') limit 10;
 molregno |                                         m                                          
----------+------------------------------------------------------------------------------------
   562310 | COc1ccc(C2C(C(=O)Nc3cccc(C)n3)c3ccccc3C(=O)N2C2CCCCC2)cc1
  1607792 | O=C(Nc1cccc(-c2ccccc2)n1)N(CCC(c1ccccc1)c1ccccc1)CCN1CCOCC1
  1587116 | O=C(Nc1cccc(F)n1)c1cc(F)cc(Oc2cncnc2)c1
  1587131 | O=C(Nc1cccc(Cl)n1)c1cc(Cl)cc(Oc2cncnc2)c1
  1592611 | Cc1cccc(NC(=O)[C@@H](C[C@H](O)[C@@H](N)CN2CC(=O)N(c3ccccc3Cl)CC2(C)C)C(C)C)n1
  1033789 | Cc1cccc(NC(=O)C(C)n2cc([N+](=O)[O-])cn2)n1
  1571533 | CCCOc1ccc(C(=O)Nc2cccc(CC)n2)cc1
  1592338 | CN1CCC(C(=O)c2cccc(NC(=O)c3c(F)cc(F)cc3F)n2)CC1
  1053301 | Cc1cccc(NC(=O)C(Cc2ccccc2)NS(=O)(=O)c2cccc3nsnc23)n1
  1323252 | O=C(Nc1cccc(NC(=O)C(=O)C2CCC3=C(Sc4ccccc4N3)C2=O)n1)C(=O)C1CCC2=C(Sc3ccccc3N2)C1=O
(10 rows)

By default mol_adjust_query_properties() makes the following changes to the molecule:

  • Converts dummy atoms into “any” queries
  • Adds a degree query to every ring atom so that its substitution must match what was provided
  • Aromaticity perception is done (if it hasn’t been done already)

We can control the behavior by providing an additional JSON argument. Here’s an example where we disable the additional degree queries:

chembl_21=# select molregno,m from rdk.mols where m@>mol_adjust_query_properties('*c1cccc(NC(=O)*)n1',
chembl_21(# '{"adjustDegree":false}') limit 10;
 molregno |                                     m                                      
----------+----------------------------------------------------------------------------
   526211 | COc1ccc2nccc(N3CCCN(CCNCc4ccc5c(n4)NC(=O)CS5)CC3)c2n1.Cl
   531033 | COc1ccc2nccc(N3CCC(CCNCc4ccc5c(n4)NC(=O)CS5)CC3)c2c1
   527205 | COc1ccc2nccc(N3CCC(O)(CCNCc4ccc5c(n4)NC(=O)CS5)CC3)c2n1.Cl
   528941 | COc1cc(F)c2ncc(F)c(CCN3C[C@@H](O)[C@@H](CNCc4ccc5c(n4)NC(=O)CS5)C3)c2c1.Cl
   539575 | COc1cc(F)c2ncc(Cl)c([C@@H](O)CN3CCC(NCc4ccc5c(n4)NC(=O)CO5)CC3)c2c1.Cl
   542344 | COc1ccc2ncc(C#N)c(CCN3C4CCC3CC(NCc3ccc5c(n3)NC(=O)CS5)C4)c2c1
   530834 | COc1ccc2ncc(F)c(CCN3CC[C@H](NC(=O)c4ccc5c(n4)NC(=O)CS5)[C@H](O)C3)c2n1
   613181 | O=C(Nc1ccc(-c2ccncc2)c(-c2ccccn2)n1)C1CC1
   527469 | COc1ccc2nccc(N3CCC(CCNCc4ccc5c(n4)NC(=O)CO5)CC3)c2c1
   527419 | COc1ccc2ncc(C#N)c(CCN3CCC(NCc4ccc5c(n4)NC(=O)CS5)CC3)c2n1
(10 rows)

or where we don’t add the additional degree queries to ring atoms or dummies (they are only added to chain atoms):

chembl_21=# select molregno,m from rdk.mols where m@>mol_adjust_query_properties('*c1cccc(NC(=O)*)n1',
chembl_21(# '{"adjustDegree":true,"adjustDegreeFlags":"IGNORERINGS|IGNOREDUMMIES"}') limit 10;
 molregno |                                     m                                      
----------+----------------------------------------------------------------------------
   526211 | COc1ccc2nccc(N3CCCN(CCNCc4ccc5c(n4)NC(=O)CS5)CC3)c2n1.Cl
   531033 | COc1ccc2nccc(N3CCC(CCNCc4ccc5c(n4)NC(=O)CS5)CC3)c2c1
   527205 | COc1ccc2nccc(N3CCC(O)(CCNCc4ccc5c(n4)NC(=O)CS5)CC3)c2n1.Cl
   528941 | COc1cc(F)c2ncc(F)c(CCN3C[C@@H](O)[C@@H](CNCc4ccc5c(n4)NC(=O)CS5)C3)c2c1.Cl
   539575 | COc1cc(F)c2ncc(Cl)c([C@@H](O)CN3CCC(NCc4ccc5c(n4)NC(=O)CO5)CC3)c2c1.Cl
   542344 | COc1ccc2ncc(C#N)c(CCN3C4CCC3CC(NCc3ccc5c(n3)NC(=O)CS5)C4)c2c1
   530834 | COc1ccc2ncc(F)c(CCN3CC[C@H](NC(=O)c4ccc5c(n4)NC(=O)CS5)[C@H](O)C3)c2n1
   613181 | O=C(Nc1ccc(-c2ccncc2)c(-c2ccccn2)n1)C1CC1
   527469 | COc1ccc2nccc(N3CCC(CCNCc4ccc5c(n4)NC(=O)CO5)CC3)c2c1
   527419 | COc1ccc2ncc(C#N)c(CCN3CCC(NCc4ccc5c(n4)NC(=O)CS5)CC3)c2n1
(10 rows)

The options available are:

  • adjustDegree (default: true) : adds a query to match the input atomic degree
  • adjustDegreeFlags (default: ADJUST_IGNOREDUMMIES | ADJUST_IGNORECHAINS) controls where the degree is adjusted
  • adjustRingCount (default: false) : adds a query to match the input ring count
  • adjustRingCountFlags (default: ADJUST_IGNOREDUMMIES | ADJUST_IGNORECHAINS) controls where the ring count is adjusted
  • makeDummiesQueries (default: true) : convert dummy atoms in the input structure into any-atom queries
  • aromatizeIfPossible (default: true) : run the aromaticity perception algorithm on the input structure (note: this is largely redundant since molecules built from smiles always have aromaticity perceived)
  • makeBondsGeneric (default: false) : convert bonds into any-bond queries
  • makeBondsGenericFlags (default: false) : controls which bonds are made generic
  • makeAtomsGeneric (default: false) : convert atoms into any-atom queries
  • makeAtomsGenericFlags (default: false) : controls which atoms are made generic

The various Flags arguments mentioned above, which control where particular options are applied, are constructed by combining operations from the list below with the | character.

  • IGNORENONE : apply the operation to all atoms
  • IGNORERINGS : do not apply the operation to ring atoms
  • IGNORECHAINS : do not apply the operation to chain atoms
  • IGNOREDUMMIES : do not apply the operation to dummy atoms
  • IGNORENONDUMMIES : do not apply the operation to non-dummy atoms
  • IGNOREALL : do not apply the operation to any atoms

Similarity searches

Basic similarity searching:

chembl_14=# select count(*) from rdk.fps where mfp2%morganbv_fp('Cc1ccc2nc(-c3ccc(NC(C4N(C(c5cccs5)=O)CCC4)=O)cc3)sc2c1');
 count
-------
    66
(1 row)

Time: 826.886 ms

Usually we’d like to find a sorted listed of neighbors along with the accompanying SMILES. This SQL function makes that pattern easy:

chembl_14=# create or replace function get_mfp2_neighbors(smiles text)
    returns table(molregno integer, m mol, similarity double precision) as
  $$
  select molregno,m,tanimoto_sml(morganbv_fp(mol_from_smiles($1::cstring)),mfp2) as similarity
  from rdk.fps join rdk.mols using (molregno)
  where morganbv_fp(mol_from_smiles($1::cstring))%mfp2
  order by morganbv_fp(mol_from_smiles($1::cstring))<%>mfp2;
  $$ language sql stable ;
CREATE FUNCTION
Time: 0.856 ms
chembl_14=#
chembl_14=# select * from get_mfp2_neighbors('Cc1ccc2nc(-c3ccc(NC(C4N(C(c5cccs5)=O)CCC4)=O)cc3)sc2c1') limit 10;
 molregno |                              m                              |    similarity
----------+-------------------------------------------------------------+-------------------
   472512 | Cc1ccc2nc(-c3ccc(NC(=O)C4CCN(C(=O)c5cccs5)CC4)cc3)sc2c1     | 0.772727272727273
   471317 | Cc1ccc2nc(-c3ccc(NC(=O)C4CCCN(S(=O)(=O)c5cccs5)C4)cc3)sc2c1 | 0.657534246575342
   471461 | Cc1ccc2nc(-c3ccc(NC(=O)C4CCN(S(=O)(=O)c5cccs5)CC4)cc3)sc2c1 | 0.647887323943662
   471319 | Cc1ccc2nc(-c3ccc(NC(=O)C4CCN(S(=O)(=O)c5cccs5)C4)cc3)sc2c1  | 0.638888888888889
  1032469 | O=C(Nc1nc2ccc(Cl)cc2s1)[C@@H]1CCCN1C(=O)c1cccs1             | 0.623188405797101
   751668 | COc1ccc2nc(NC(=O)[C@@H]3CCCN3C(=O)c3cccs3)sc2c1             | 0.619718309859155
   471318 | Cc1ccc2nc(-c3ccc(NC(=O)C4CN(S(=O)(=O)c5cccs5)C4)cc3)sc2c1   | 0.611111111111111
   740754 | Cc1ccc(NC(=O)C2CCCN2C(=O)c2cccs2)cc1C                       | 0.606060606060606
   732905 | O=C(Nc1ccc(S(=O)(=O)N2CCCC2)cc1)C1CCCN1C(=O)c1cccs1         | 0.602941176470588
  1087495 | Cc1ccc(NC(=O)C2CCCN2C(=O)c2cccs2)c(C)c1                     | 0.597014925373134
(10 rows)

Time: 5453.200 ms
chembl_14=# select * from get_mfp2_neighbors('Cc1ccc2nc(N(C)CC(=O)O)sc2c1') limit 10;
 molregno |                           m                           |    similarity
----------+-------------------------------------------------------+-------------------
   412312 | Cc1ccc2nc(N(C)CCN(C)c3nc4ccc(C)cc4s3)sc2c1            | 0.692307692307692
   470082 | CN(CC(=O)O)c1nc2cc([N+](=O)[O-])ccc2s1                | 0.583333333333333
  1040255 | CC(=O)N(CCCN(C)C)c1nc2ccc(C)cc2s1                     | 0.571428571428571
   773946 | Cl.CC(=O)N(CCCN(C)C)c1nc2ccc(C)cc2s1                  | 0.549019607843137
  1044892 | Cc1ccc2nc(N(CCN(C)C)C(=O)c3cc(Cl)sc3Cl)sc2c1          | 0.518518518518518
  1040496 | Cc1ccc2nc(N(CCCN(C)C)C(=O)CCc3ccccc3)sc2c1            | 0.517857142857143
  1049393 | Cc1ccc2nc(N(CCCN(C)C)C(=O)CS(=O)(=O)c3ccccc3)sc2c1    | 0.517857142857143
   441378 | Cc1ccc2nc(NC(=O)CCC(=O)O)sc2c1                        | 0.510204081632653
  1042958 | Cc1ccc2nc(N(CCN(C)C)C(=O)c3ccc4ccccc4c3)sc2c1         | 0.509090909090909
  1047691 | Cc1ccc(S(=O)(=O)CC(=O)N(CCCN(C)C)c2nc3ccc(C)cc3s2)cc1 | 0.509090909090909
(10 rows)

Time: 1797.656 ms

Adjusting the similarity cutoff

By default, the minimum similarity returned with a similarity search is 0.5. This can be adjusted with the rdkit.tanimoto_threshold (and rdkit.dice_threshold) configuration variables:

chembl_14=# select count(*) from get_mfp2_neighbors('Cc1ccc2nc(N(C)CC(=O)O)sc2c1');
 count
-------
    18
(1 row)

Time: 1199.751 ms
chembl_14=# set rdkit.tanimoto_threshold=0.7;
SET
Time: 0.191 ms
chembl_14=# select count(*) from get_mfp2_neighbors('Cc1ccc2nc(N(C)CC(=O)O)sc2c1');
 count
-------
     0
(1 row)

Time: 826.058 ms
chembl_14=# set rdkit.tanimoto_threshold=0.6;
SET
Time: 0.220 ms
chembl_14=# select count(*) from get_mfp2_neighbors('Cc1ccc2nc(N(C)CC(=O)O)sc2c1');
 count
-------
     1
(1 row)

Time: 1092.303 ms
chembl_14=# set rdkit.tanimoto_threshold=0.5
chembl_14-# ;
SET
Time: 0.257 ms
chembl_14=# select count(*) from get_mfp2_neighbors('Cc1ccc2nc(N(C)CC(=O)O)sc2c1');
 count
-------
    18
(1 row)

Time: 1081.721 ms

Using the MCS code

The most straightforward use of the MCS code is to find the maximum common substructure of a group of molecules:

chembl_20=# select fmcs(m::text) from rdk.mols join compound_records using (molregno) where doc_id=3;                                                                                           fmcs
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
 [#6]1(-[#7](-[#6](-[#6]2:[#6]:[#6]:[#6](:[#6]:[#6]:2)-[#7]-[#6](-[#6]2:[#6](-[#6]3:[#6]:[#6]:[#6]:[#6]:[#6]:3):[#6]:[#6]:[#6]:[#6]:2)=[#8])=[#8])-[#6]-[#6]-[#6]):[#6]:[#16]:[#6]:[#6]:1
(1 row)

chembl_20=# select fmcs(m::text) from rdk.mols join compound_records using (molregno) where doc_id=4;
                                  fmcs
------------------------------------------------------------------------
 [#6](-[#6]-,:[#6]-,:[#6]-,:[#6]-,:[#6])-[#7]-[#6]-[#6](-,:[#6])-,:[#6]
(1 row)

The same thing can be done with a SMILES column:

chembl_20=# select fmcs(canonical_smiles) from compound_structures join compound_records using (molregno) where doc_id=4;
                                  fmcs
------------------------------------------------------------------------
 [#6](-[#7]-[#6]-[#6]-,:[#6]-,:[#6]-,:[#6]-,:[#6])-[#6](-,:[#6])-,:[#6]
(1 row)

It’s also possible to adjust some of the parameters to the FMCS algorithm, though this is somewhat more painful as of this writing (the 2015_03_1 release). Here are a couple of examples:

chembl_20=# select fmcs_smiles(str,'{"Threshold":0.8}') from
chembl_20-#   (select string_agg(m::text,' ') as str from rdk.mols
chembl_20(#   join compound_records using (molregno) where doc_id=4) as str ;
                                                                           fmcs_smiles
------------------------------------------------------------------------------------------------------------------------------------------------------------------
 [#6]-[#6]-[#8]-[#6](-[#6](=[#8])-[#7]-[#6](-[#6])-[#6](-,:[#6])-,:[#6])-[#6](-[#8])-[#6](-[#8])-[#6](-[#8]-[#6]-[#6])-[#6]-[#7]-[#6](-[#6])-[#6](-,:[#6])-,:[#6]
(1 row)

chembl_20=# select fmcs_smiles(str,'{"AtomCompare":"Any"}') from
chembl_20-# (select string_agg(m::text,' ') as str from rdk.mols
chembl_20(# join compound_records using (molregno) where doc_id=4) as str ;
                                                                              fmcs_smiles
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
 [#6]-,:[#6,#7]-[#8,#6]-[#6,#7](-[#6,#8]-[#7,#6]-,:[#6,#7]-,:[#6,#7]-,:[#7,#6]-,:[#6])-[#6,#7]-[#6]-[#6](-[#8,#6]-[#6])-[#6,#7]-[#7,#6]-[#6]-,:[#6,#8]-,:[#7,#6]-,:[#6]
(1 row)

Note The combination of "AtomCompare":"Any" and a value of "Threshold" that is less than 1.0 does a quite generic search and can results in very long search times. Using "Timeout" with this combination is recommended:

chembl_20=# select fmcs_smiles(str,'{"AtomCompare":"Any","CompleteRingsOnly":true,"Threshold":0.8,"Timeout":60}') from
chembl_20-#  (select string_agg(m::text,' ') as str from rdk.mols
chembl_20(#   join compound_records using (molregno) where doc_id=3) as str ;
WARNING:  findMCS timed out, result is not maximal
                                                                                          fmcs_smiles
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
 [#8]=[#6](-[#7]-[#6]1:[#6]:[#6]:[#6](:[#6]:[#6]:1)-[#6](=[#8])-[#7]1-[#6]-[#6]-[#6]-[#6,#7]-[#6]2:[#6]-1:[#6]:[#6]:[#16]:2)-[#6]1:[#6]:[#6]:[#6]:[#6]:[#6]:1-[#6]1:[#6]:[#6]:[#6]:[#6]:[#6]:1
(1 row)

Available parameters and their default values are:

  • MaximizeBonds (true)
  • Threshold (1.0)
  • Timeout (-1, no timeout)
  • MatchValences (false)
  • MatchChiralTag (false) Applies to atoms
  • RingMatchesRingOnly (false)
  • CompleteRingsOnly (false)
  • MatchStereo (false) Applies to bonds
  • AtomCompare (“Elements”) can be “Elements”, “Isotopes”, or “Any”
  • BondCompare (“Order”) can be “Order”, “OrderExact”, or “Any”

Reference Guide

New Types

  • mol : an rdkit molecule. Can be created from a SMILES via direct type conversion, for example: ‘c1ccccc1’::mol creates a molecule from the SMILES ‘c1ccccc1’
  • qmol : an rdkit molecule containing query features (i.e. constructed from SMARTS). Can be created from a SMARTS via direct type conversion, for example: ‘c1cccc[c,n]1’::qmol creates a query molecule from the SMARTS ‘c1cccc[c,n]1’
  • sfp : a sparse count vector fingerprint (SparseIntVect in C++ and Python)
  • bfp : a bit vector fingerprint (ExplicitBitVect in C++ and Python)

Parameters

  • rdkit.tanimoto_threshold : threshold value for the Tanimoto similarity operator. Searches done using Tanimoto similarity will only return results with a similarity of at least this value.
  • rdkit.dice_threshold : threshold value for the Dice similiarty operator. Searches done using Dice similarity will only return results with a similarity of at least this value.
  • rdkit.do_chiral_sss : toggles whether or not stereochemistry is used in substructure matching. (available from 2013_03 release).
  • rdkit.sss_fp_size : the size (in bits) of the fingerprint used for substructure screening.
  • rdkit.morgan_fp_size : the size (in bits) of morgan fingerprints
  • rdkit.featmorgan_fp_size : the size (in bits) of featmorgan fingerprints
  • rdkit.layered_fp_size : the size (in bits) of layered fingerprints
  • rdkit.rdkit_fp_size : the size (in bits) of RDKit fingerprints
  • rdkit.torsion_fp_size : the size (in bits) of topological torsion bit vector fingerprints
  • rdkit.atompair_fp_size : the size (in bits) of atom pair bit vector fingerprints
  • rdkit.avalon_fp_size : the size (in bits) of avalon fingerprints

Operators

Molecule comparison

  • < : returns whether or not the left mol is less than the right mol
  • > : returns whether or not the left mol is greater than the right mol
  • = : returns whether or not the left mol is equal to the right mol
  • <= : returns whether or not the left mol is less than or equal to the right mol
  • >= : returns whether or not the left mol is greater than or equal to the right mol

Note Two molecules are compared by making the following comparisons in order. Later comparisons are only made if the preceding values are equal:

# Number of atoms # Number of bonds # Molecular weight # Number of rings

If all of the above are the same and the second molecule is a substructure of the first, the molecules are declared equal, Otherwise (should not happen) the first molecule is arbitrarily defined to be less than the second.

There are additional operators defined in the cartridge, but these are used for internal purposes.

Functions

Other

  • rdkit_version() : returns a string with the cartridge version number.

There are additional functions defined in the cartridge, but these are used for internal purposes.

Using the Cartridge from Python

The recommended adapter for connecting to postgresql is pyscopg2 (https://pypi.python.org/pypi/psycopg2).

Here’s an example of connecting to our local copy of ChEMBL and doing a basic substructure search:

>>> import psycopg2
>>> conn = psycopg2.connect(database='chembl_16')
>>> curs = conn.cursor()
>>> curs.execute('select * from rdk.mols where m@>%s',('c1cccc2c1nncc2',))
>>> curs.fetchone()
(9830, 'CC(C)Sc1ccc(CC2CCN(C3CCN(C(=O)c4cnnc5ccccc54)CC3)CC2)cc1')

That returns a SMILES for each molecule. If you plan to do more work with the molecules after retrieving them, it is much more efficient to ask postgresql to give you the molecules in pickled form:

>>> curs.execute('select molregno,mol_send(m) from rdk.mols where m@>%s',('c1cccc2c1nncc2',))
>>> row = curs.fetchone()
>>> row
(9830, <read-only buffer for 0x...>)

These pickles can then be converted into molecules:

>>> from rdkit import Chem
>>> m = Chem.Mol(str(row[1]))
>>> Chem.MolToSmiles(m,True)
'CC(C)Sc1ccc(CC2CCN(C3CCN(C(=O)c4cnnc5ccccc54)CC3)CC2)cc1'

License

This document is copyright (C) 2013-2016 by Greg Landrum

This work is licensed under the Creative Commons Attribution-ShareAlike 4.0 License. To view a copy of this license, visit http://creativecommons.org/licenses/by-sa/4.0/ or send a letter to Creative Commons, 543 Howard Street, 5th Floor, San Francisco, California, 94105, USA.

The intent of this license is similar to that of the RDKit itself. In simple words: “Do whatever you want with it, but please give us some credit.”