Skip to content
/ mmpdb Public
forked from rdkit/mmpdb

A package to identify matched molecular pairs and use them to predict property changes.

License

Notifications You must be signed in to change notification settings

ricrogz/mmpdb

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

mmpdb 3.1 - matched molecular pair database generation and analysis

Synopsis

A package to identify matched molecular pairs and use them to predict property changes and generate new molecular structures.


Installation

mmpdb 3.1 must be installed before use. (Earlier versions of mmpdb could be run in-place, in the top-level directory.) This will also ensure that the SciPy, peewee, and click packages are installed.

To install from PyPI using pip, which comes with Python:

On macOS and other Unix-like systems:

python -m pip install mmpdb

On Windows:

py -m pip install mmpdb

If you are using a virtual environment (eg, with venv or conda then the installer will place a copy of the relevant files into the virtual environment's Python package directory, and place the mmpdb command-line driver in your path.

If you are not using a virtual environment (and you should be using a virtual environment) then the installer will place the relevant files into Python's system directory. If you do not have write permissions to that directory, or only want to install it for your personal use, then add the --user flag at the end of the install command.

To install from source directory, go to the top-level directory then do:

On macOS and other Unix-like systems:

python -m pip install .

On Windows:

py -m pip install .

If you plan to modify the source code and want the installation to use the local directory rather than make a copy into the package directory, then you need an "editable" installation,

On macOS and other Unix-like systems:

python -m pip install -e .

On Windows:

py -m pip install -e .

(Assuming the dependencies are installed then it is possible to use mmpdb without installation by going to the top-level directory and using python -m mmpdblib or, for Windows, py -m mmpdlib. This is not recommended.)

Requirements

The package has been tested on Python 3.9 and 3.10. It should work under newer versions of Python.

You will need a copy of the RDKit cheminformatics toolkit, available from http://rdkit.org/ , which in turn requires NumPy. You will also need SciPy, peewee, and click. The latter three are listed as dependencies in setup.cfg and should be installed automatically.

Optional components you may find useful are:

  • The matched molecular pairs may instead by be stored in a Postgres database. These were tested using the psycopg2 adapter. See mmpdb help-postgres for more information.

  • The "--memory" option in the index command requires the psutil module to get memory use information.

NOTE: mmpdb 2 used a JSON-Lines format for the fragment files, and suggested an optional package with faster JSON parsing. mmpdb 3 no longer uses this format.


How to run the program and get help

The package includes a command-line program named "mmpdb". This support many subcommands. For examples:

  • "mmpdb fragment" -- fragment a SMILES file

  • "mmpdb index" -- find matched molecular pairs in a fragment file

Use the "--help" option to get more information about any of the commands. For example, "mmpdb fragment --help" will print the command-line arguments, describe how they are used, and show examples of use.

The subcommands starting with "help-" print additional information about a given topic. Much of the text of this README come from the output of

 % mmpdb help-analysis 
 % mmpdb help-distributed 

If you wish to experiment with a simple test set, use tests/test_data.smi, with molecular weight and melting point properties in tests/test_data.csv.


Publication

An open-access publication describing this package has been published in the Journal of Chemical Information and Modeling:

A. Dalke, J. Hert, C. Kramer. mmpdb: An Open-Source Matched Molecular Pair Platform for Large Multiproperty Data Sets. J. Chem. Inf. Model., 2018, 58 (5), pp 902–910. https://pubs.acs.org/doi/10.1021/acs.jcim.8b00173

For more about the methods to scale mmpdb to larger datasets and generate new molecules, see:

M. Awale, J. Hert, L. Guasch, S. Riniker, C. Kramer. The Playbooks of Medicinal Chemistry Design Moves. J. Chem. Inf. Model., 2021, 61 (2), pp 729-742. https://pubs.acs.org/doi/abs/10.1021/acs.jcim.0c01143


Background

The overall process is:

  1. Fragment structures in a SMILES file, to produce fragments.

  2. Index the fragments to produces matched molecular pairs. (you might include property information at this point)

  3. Load property information.

  4. Find transforms for a given structure; and/or

  5. Predict a property for a structure given the known property for another structure; and/or

  6. Apply 1-cut rules to generate new structures from a given structure.

Some terminology:

A fragmentation cuts 1, 2, or 3 non-ring bonds to convert a structure into a "constant" part and a "variable" part. The substructure in the variable part is a single fragment, and often considered the R-groups, while the constant part contains one fragment for each cut, and it often considered as containing the core.

The matched molecular pair indexing process finds all pairs which have the same constant part, in order to define a transformation from one variable part to another variable part. A "rule" stores information about a transformation, including a list of all the pairs for that rule.

The "rule environment" extends the transformation to include information about the local environment of the attachment points on the constant part. The environment fingerprint is based on the RDKit circular fingerprints for the attachment points, expressed as a canonical SMARTS pattern, and alternatively, as a "pseudo"-SMILES string, which is a bit less precise but easier to understand and visualize.

The fingerprint SMARTS pattern describes the Morgan circular fingerprint invariants around the attachment points. Here's a 2-cut example split across three lines:

[#0;X1;H0;+0;!R:1]-[#6;X4;H1;+0;R](-[#6;X4;H2;+0;R])-[#6;X4;H2;+0;R].
[#0;X1;H0;+0;!R:2]-[#7;X3;H0;+0;R](-[#6;X4;H2;+0;R])-[#6;X4;H2;+0;R].
[#0;X1;H0;+0;!R:3]-[#6;X3;H0;+0;R](:[#6;X3;H1;+0;R]):[#6;X3;H1;+0;R]

The SMARTS modifiers, like "H0" to require no hydrogens, are needed to match the Morgan invariants but are quite the eye-full. The psuedosmiles alternative is:

[*:1]-[CH](-[CH2](~*))-[CH2](~*).
[*:2]-[N](-[CH2](~*))-[CH2](~*).
[*:3]-[c](:[cH](~*)):[cH](~*)

This can be processed by RDKit, if sanitization is disabled, and turned into an image.

CAUTION! The "(~*)" terms are used to represent the SMARTS connectivity terms "X", but they do not necessarily all represent distinct atoms!

There is one rule environment for each available radius. Larger radii correspond to more specific environments. The "rule environment statistics" table stores information about the distribution of property changes for all of the pairs which contain the given rule and environment, with one table for each property.

1) Fragment structures

Use "smifrag" to see how a given SMILES is fragmented. Use "fragment" to fragment all of the compounds in a SMILES file.

"mmpdb smifrag" is a diagnostic tool to help understand how a given SMILES will be fragmented and to experiment with the different fragmentation options. For example:

% mmpdb smifrag 'c1ccccc1OC'
                   |-------------  variable  -------------|       |---------------------  constant  --------------------
#cuts | enum.label | #heavies | symm.class | smiles       | order | #heavies | symm.class | smiles           | with-H   
------+------------+----------+------------+--------------+-------+----------+------------+------------------+----------
  1   |     N      |    2     |      1     | [*]OC        |    0  |    6     |      1     | [*]c1ccccc1      | c1ccccc1 
  1   |     N      |    6     |      1     | [*]c1ccccc1  |    0  |    2     |      1     | [*]OC            | CO       
  2   |     N      |    1     |     11     | [*]O[*]      |   01  |    7     |     12     | [*]C.[*]c1ccccc1 | -        
  1   |     N      |    1     |      1     | [*]C         |    0  |    7     |      1     | [*]Oc1ccccc1     | Oc1ccccc1
  1   |     N      |    7     |      1     | [*]Oc1ccccc1 |    0  |    1     |      1     | [*]C             | C        

Use "mmpdb fragment" to fragment a SMILES file and produce a fragment file for the MMP analysis. Start with the test data file named "test_data.smi" containing the following structures:

Oc1ccccc1 phenol  
Oc1ccccc1O catechol  
Oc1ccccc1N 2-aminophenol  
Oc1ccccc1Cl 2-chlorophenol  
Nc1ccccc1N o-phenylenediamine  
Nc1cc(O)ccc1N amidol  
Oc1cc(O)ccc1O hydroxyquinol  
Nc1ccccc1 phenylamine  
C1CCCC1N cyclopentanol  

then run the following command generate a fragment database.

% mmpdb fragment test_data.smi -o test_data.fragdb

Fragmentation can take a while. You can save time by asking the code to reuse fragmentations from a previous run. If you do that then the fragment command will reuse the old fragmentation parameters. (You cannot override them with command-line options.). Here is an example:

% mmpdb fragment data_file.smi -o new_data_file.fragdb \
       --cache old_data_file.fragdb

The "--cache" option will greatly improve the fragment performance when there are only a few changes from the previous run.

The fragmentation algorithm is configured to ignore structures which are too big or have too many rotatable bonds. There are also options which change where to make cuts and the number of cuts to make. Use the "--help" option on each command for details.

Use "mmpdb help-smiles-format" for details about to parse different variants of the SMILES file format.

2) Index the MMPA fragments to create a database

The "mmpa index" command indexes the output fragments from "mmpa fragment" by their variable fragments, that is, it finds fragmentations with the same R-groups and puts them together. Here's an example:

% mmpdb index test_data.fragdb -o test_data.mmpdb

The output from this is a SQLite database.

If you have activity/property data and you do not want the database to include structures where there is no data, then you can specify the properties file as well:

% mmpdb index test_data.fragdb -o test_data.mmpdb --properties test_data.csv

Use "mmpdb help-property-format" for more details about the property file format.

For more help use "mmpdb index --help".

3) Add properties to a database

Use "mmpdb loadprops" to add or modify activity/property data in the database. Here's an example property file named 'test_data.csv' with molecular weight and melting point properties:

ID      MW      MP  
phenol  94.1    41  
catechol        110.1   105  
2-aminophenol   109.1   174  
2-chlorophenol  128.6   8  
o-phenylenediamine      108.1   102  
amidol  124.1   *  
hydroxyquinol   126.1   140  
phenylamine     93.1    -6  
cyclopentanol   86.1    -19  

The following loads the property data to the MMPDB database file created in the previous section:

% mmpdb loadprops -p test_data.csv test_data.mmpdb
Using dataset: MMPs from 'test_data.fragdb'
Reading properties from 'tests/test_data.csv'
Read 2 properties for 9 compounds from 'tests/test_data.csv'
Imported 9 'MW' records (9 new, 0 updated).
Imported 8 'MP' records (8 new, 0 updated).
Number of rule statistics added: 533 updated: 0 deleted: 0
Loaded all properties and re-computed all rule statistics.

Use "mmpdb help-property-format" for more details about the property file format.

For more help use "mmpdb loadprops --help". Use "mmpdb list" to see what properties are already loaded.

4) Identify possible transforms

Use "mmpdb transform" to transform an input structure using the rules in a database. For each transformation, it can estimate the effect on any properties. The following looks at possible ways to transform 2-pyridone using the test dataset created in the previous section, and predict the effect on the "MW" property (the output is reformatted for clarity):

% mmpdb transform --smiles 'c1cccnc1O' test_data.mmpdb --property MW
ID     SMILES    MW_from_smiles    MW_to_smiles    MW_radius
1    Clc1ccccn1     [*:1]O          [*:1]Cl           1
2     Nc1ccccn1     [*:1]O          [*:1]N            1
3     c1ccncc1      [*:1]O          [*:1][H]          1

      MW_smarts                        MW_pseudosmiles    MW_rule_environment_id 
[#0;X1;H0;+0;!R:1]-[#6;X3;H0;+0;R]    [*:1]-[#6](~*)(~*)    299
[#0;X1;H0;+0;!R:1]-[#6;X3;H0;+0;R]    [*:1]-[#6](~*)(~*)    276
[#0;X1;H0;+0;!R:1]-[#6;X3;H0;+0;R]    [*:1]-[#6](~*)(~*)    268

MW_count    MW_avg    MW_std    MW_kurtosis    MW_skewness
    1        18.5
    3        -1         0            0
    4       -16         0            0

MW_min  MW_q1  MW_median  MW_q3  MW_max  MW_paired_t    MW_p_value
 18.5    18.5   18.5      18.5    18.5
 -1      -1     -1        -1      -1      1e+08    
-16     -16    -16       -16     -16      1e+08 

This says that "c1cccnc1O" can be transformed to "Clc1ccccn1" using the transformation [*:1]O>>[*:1]Cl (that is, replace the oxygen with a chlorine). The best transformation match has a radius of 1, which includes the aromatic carbon at the attachment point but not the aromatic nitrogen which is one atom away.

There is only one pair for this transformation, and it predicts a shift in molecular weight of 18.5. This makes sense as the [OH] is replaced with a [Cl].

On the other hand, there are three pairs which transform it to pyridine. The standard deviation of course is 0 because it's a simple molecular weight calculation. The 1e+08.0 is the mmpdb way of writing "positive infinity".

Melting point is more complicated. The following shows that in the transformation of 2-pyridone to pyridine there are still 3 matched pairs and in this case the average shift is -93C with a standard deviation of 76.727C:

% mmpdb transform --smiles 'c1cccnc1O' test_data.mmpdb --property MP
ID   SMILES    MP_from_smiles   MP_to_smiles   MP_radius   
1   Clc1ccccn1    [*:1]O           [*:1]Cl        1
2    Nc1ccccn1    [*:1]O           [*:1]N         1
3    c1ccncc1     [*:1]O           [*:1][H]       1

MP_smarts                            MP_pseudosmiles     MP_rule_environment_id
[#0;X1;H0;+0;!R:1]-[#6;X3;H0;+0;R]   [*:1]-[#6](~*)(~*)   299
[#0;X1;H0;+0;!R:1]-[#6;X3;H0;+0;R]   [*:1]-[#6](~*)(~*)   276
[#0;X1;H0;+0;!R:1]-[#6;X3;H0;+0;R]   [*:1]-[#6](~*)(~*)   268

MP_count   MP_avg   MP_std   MP_kurtosis   MP_skewness   
   1       -97            
   3       -16.667   75.235     -1.5        -0.33764   
   3       -93       76.727     -1.5        -0.32397   

MP_min   MP_q1   MP_median   MP_q3   MP_max   MP_paired_t   MP_p_value
 -97      -97       -97       -97     -97      
 -72      -65.75    -47        40      69        0.3837      0.73815
-180     -151       -64       -42.25  -35       -2.0994      0.17062

You might try enabling the "--explain" option to see why the algorithm selected a given tranformation.

For more help use "mmpdb transform --help".

5) Use MMP to make a prediction

Use "mmpdb predict" to predict the property change in a transformation from a given reference structure to a given query structure. Use this when you want to limit the transform results when you know the starting and ending structures. The following predicts the effect on molecular weight in transforming 2-pyridone to pyridone:

% mmpdb predict --smiles 'c1cccnc1' --reference 'c1cccnc1O' \
          test_data.mmpdb --property MP
predicted delta: -93 +/- 76.7268

This is the same MP_value and MP_std from the previous section using 'transform'.

The reference value may also be included in the calulation, to give a predicted value.

% mmpdb predict --smiles 'c1cccnc1' --reference 'c1cccnc1O' \
          test_data.mmpdb --property MP --value -41.6
predicted delta: -93 predicted value: -134.6 +/- 76.7268

I'll redo the calculation with the molecular weight property, and have mmpdb do the trival calculation of adding the known weight to the predicted delta:

% mmpdb predict --smiles 'c1cccnc1' --reference 'c1cccnc1O' \
          test_data.mmpdb --property MW --value 95.1
predicted delta: -16 predicted value: 79.1 +/- 0

You might try enabling the "--explain" option to see why the algorithm selected a given transformation, or use "--save-details" to save the list of possible rules to the file pred_detail_rules.txt and to save the list of rule pairs to pred_detail_pairs.txt.

6) Use MMP to generate new structures

The rules in a MMP database give a sort of "playbook" about the transformations which might be explored in medicinal chemistry. These rule can be applied to a given structure to generate new related structures, following a method related to the transform command but ignoring any property information. Here's an example using the default radius of 0, which means the environment fingerprint is ignored. (The columns have been re-formatted for the documentation.)

% mmpdb generate --smiles 'c1ccccc1C(O)C' test_data.mmpdb
start             constant  from_smiles  to_smiles          r  pseudosmiles  final
CC(O)c1ccccc1  *C(C)c1ccccc1  [*:1]O    [*:1][H]            0  [*:1](~*)  CCc1ccccc1
CC(O)c1ccccc1  *C(C)c1ccccc1  [*:1]O    [*:1]N              0  [*:1](~*)  CC(N)c1ccccc1
CC(O)c1ccccc1  *C(C)c1ccccc1  [*:1]O    [*:1]Cl             0  [*:1](~*)  CC(Cl)c1ccccc1
CC(O)c1ccccc1  *C(C)O     [*:1]c1ccccc1 [*:1]c1ccccc1O      0  [*:1](~*)  CC(O)c1ccccc1O
CC(O)c1ccccc1  *C(C)O     [*:1]c1ccccc1 [*:1]c1ccccc1N      0  [*:1](~*)  CC(O)c1ccccc1N
CC(O)c1ccccc1  *C(C)O     [*:1]c1ccccc1 [*:1]c1cc(O)ccc1N   0  [*:1](~*)  CC(O)c1cc(O)ccc1N
CC(O)c1ccccc1  *C(C)O     [*:1]c1ccccc1 [*:1]c1ccc(O)cc1N   0  [*:1](~*)  CC(O)c1ccc(O)cc1N
CC(O)c1ccccc1  *C(C)O     [*:1]c1ccccc1 [*:1]C1CCCC1        0  [*:1](~*)  CC(O)C1CCCC1
CC(O)c1ccccc1  *C(C)O     [*:1]c1ccccc1 [*:1]c1ccccc1Cl     0  [*:1](~*)  CC(O)c1ccccc1Cl
CC(O)c1ccccc1  *C(C)O     [*:1]c1ccccc1 [*:1]c1ccc(N)c(N)c1 0  [*:1](~*)  CC(O)c1ccc(N)c(N)c1
CC(O)c1ccccc1  *C(C)O     [*:1]c1ccccc1 [*:1]c1cc(O)ccc1O   0  [*:1](~*)  CC(O)c1cc(O)ccc1O
CC(O)c1ccccc1  *C(C)O     [*:1]c1ccccc1 [*:1]c1ccc(O)c(O)c1 0  [*:1](~*)  CC(O)c1ccc(O)c(O)c1
CC(O)c1ccccc1  *C(C)O     [*:1]c1ccccc1 [*:1]c1ccc(O)cc1O   0  [*:1](~*)  CC(O)c1ccc(O)cc1O

#pairs  pair_from_id  pair_from_smiles  pair_to_id  pair_to_smiles
4       2-aminophenol  Nc1ccccc1O     phenylamine        Nc1ccccc1
3       phenol         Oc1ccccc1      phenylamine        Nc1ccccc1
1       catechol       Oc1ccccc1O     2-chlorophenol     Oc1ccccc1Cl
2       phenylamine    Nc1ccccc1      2-aminophenol      Nc1ccccc1O
2       phenylamine    Nc1ccccc1      o-phenylenediamine Nc1ccccc1N
1       phenylamine    Nc1ccccc1      amidol             Nc1ccc(O)cc1N
1       phenylamine    Nc1ccccc1      amidol             Nc1ccc(O)cc1N
1       phenylamine    Nc1ccccc1      cyclopentanol      NC1CCCC1
1       phenol         Oc1ccccc1      2-chlorophenol     Oc1ccccc1Cl
1       phenol         Oc1ccccc1      amidol             Nc1ccc(O)cc1N
1       phenol         Oc1ccccc1      hydroxyquinol      Oc1ccc(O)c(O)c1
1       phenol         Oc1ccccc1      hydroxyquinol      Oc1ccc(O)c(O)c1
1       phenol         Oc1ccccc1      hydroxyquinol      Oc1ccc(O)c(O)c1

The second half the output shows the number of known pairs for the given rule environment (use --min-pairs N to require at least N pairs), and gives a representative pair from the dataset.

In the above example, all of the fragmentations in the specified --smiles are used. Alternatively, you may specify --smiles and one of --constant or --query to use that specific fragmentation, or use --constant and --query (without --smiles) to specify the exact pair.

There is also an option to generate --subqueries. This generates all of the unique 1-cut fragmentations of the query, and uses them as additional queries. I'll use the --constant to specify the phynol group, leaving the aminomethanol available as the query. I'll use --subqueries to include fragments of the query. I'll limit the output --columns to the start and final SMILES structures, and the number of pairs. I'll use --explain to display debug information, and finally, I'll use --no-header to make the output a bit less complicated:

% mmpdb generate --smiles 'c1ccccc1C(O)N' --constant '*c1ccccc1' test_data.mmpdb \
     --subqueries --columns start,final,#pairs --explain --no-header
Number of subqueries: 4
Subqueries are: ['*CN', '*CO', '*N', '*O']
Using constant SMILES *c1ccccc1 with radius 0.
Environment SMARTS: [#0;X1;H0;+0;!R:1] pseudoSMILES: [*:1](~*)
Number of matching environment rules: 42
Query SMILES [*:1]C(N)O is not a rule_smiles in the database.
Query SMILES [*:1]CN is not a rule_smiles in the database.
Query SMILES [*:1]CO is not a rule_smiles in the database.
Nc1ccccc1     Oc1ccccc1       3
Nc1ccccc1     c1ccccc1        2
Nc1ccccc1     Clc1ccccc1      1
Number of rules for [*:1]N: 3
Oc1ccccc1     c1ccccc1        4
Oc1ccccc1     Nc1ccccc1       3
Oc1ccccc1     Clc1ccccc1      1
Number of rules for [*:1]O: 3

Distributed computing

These commands enable MMP generation on a distributed compute cluster, rather than a single machine.

NOTE: This method does not support properties, and you must use the SQLite- based "mmpdb" files, not Postgres databases. The Postgres wiki mentions pgloader as a possible tool to have Postgres load a SQLite database.

These examples assume you work in a queueing environment with a shared file system, and a queueing system which lets you submit a command and a list of filenames, to enqueue the command once for each filename.

This documentation will use the command 'qsub' as a wrapper around GNU Parallel:

alias qsub="parallel --no-notice -j 1 --max-procs 4"

This alias suppresses the request to cite GNU parallel in scientific papers, and has it process one filename at a time, with at most 4 processes in parallel.

I'll pass the filenames to process via stdin, like this example:

% ls /etc/passwd ~/.bashrc | qsub wc
       2       5      88 /Users/dalke/.bashrc
     120     322    7630 /etc/passwd

This output shows that wc received only a single filename because with two filenames it also shows a 'total' line.

% wc /etc/passwd ~/.bashrc
     120     322    7630 /etc/passwd
       2       5      88 /Users/dalke/.bashrc
     122     327    7718 total

Distributed fragmentation generation

NOTE: This method can also be used to process larger data sets on a single machine because the mmpdb merge step uses less memory than the mmpdb index.

The fragment command supports multi-processing with the -j flag, which scales to about 4 or 8 processors. For larger data sets you can break the SMILES dataset into multiple files, fragment each file indepenently, then merge the results.

These steps are:

  • smi_split - split the SMILES file into smaller files
  • fragment - fragment the each smaller SMILES file into its own fragb file.
  • fragdb_merge - merge the smaller fragdb files together.

Use smi_split to create N smaller SMILES files

I'll start with a SMILES file containing a header and 20267 SMILES lines:

% head -3 ChEMBL_CYP3A4_hERG.smi
SMILES  CMPD_CHEMBLID
[2H]C([2H])([2H])Oc1cc(ncc1C#N)C(O)CN2CCN(C[C@H](O)c3ccc4C(=O)OCc4c3C)CC2       CHEMBL3612928
[2H]C([2H])(N[C@H]1C[S+]([O-])C[C@@H](Cc2cc(F)c(N)c(O[C@H](COC)C(F)(F)F)c2)[C@@H]1O)c3cccc(c3)C(C)(C)C  CHEMBL2425617
% wc -l ChEMBL_CYP3A4_hERG.smi
   20268 ChEMBL_CYP3A4_hERG.smi

By default the "smi_split" command splits a SMILES file into 10 files. (Use -n or --num-files to change the number of files, or use --num-records to have N records per file.)

% mmpdb smi_split ChEMBL_CYP3A4_hERG.smi
Created 10 SMILES files containing 20268 SMILES records.

That "20268 SMILES record" shows that all 20268 lines were used to generate SMILES records, which is a mistake as it includes the header line. I'll re-do the command with --has-header to have it skip the header:

% mmpdb smi_split ChEMBL_CYP3A4_hERG.smi --has-header
Created 10 SMILES files containing 20267 SMILES records.

By default this generates files which look like:

% ls -l ChEMBL_CYP3A4_hERG.*.smi
-rw-r--r--  1 dalke  admin  141307 Feb 10 15:10 ChEMBL_CYP3A4_hERG.0000.smi
-rw-r--r--  1 dalke  admin  152002 Feb 10 15:10 ChEMBL_CYP3A4_hERG.0001.smi
-rw-r--r--  1 dalke  admin  127397 Feb 10 15:10 ChEMBL_CYP3A4_hERG.0002.smi
-rw-r--r--  1 dalke  admin  137930 Feb 10 15:10 ChEMBL_CYP3A4_hERG.0003.smi
-rw-r--r--  1 dalke  admin  130585 Feb 10 15:10 ChEMBL_CYP3A4_hERG.0004.smi
-rw-r--r--  1 dalke  admin  150072 Feb 10 15:10 ChEMBL_CYP3A4_hERG.0005.smi
-rw-r--r--  1 dalke  admin  139620 Feb 10 15:10 ChEMBL_CYP3A4_hERG.0006.smi
-rw-r--r--  1 dalke  admin  133347 Feb 10 15:10 ChEMBL_CYP3A4_hERG.0007.smi
-rw-r--r--  1 dalke  admin  131310 Feb 10 15:10 ChEMBL_CYP3A4_hERG.0008.smi
-rw-r--r--  1 dalke  admin  129344 Feb 10 15:10 ChEMBL_CYP3A4_hERG.0009.smi

The output filenames are determined by the --template option, which defaults to {prefix}.{i:04}.smi, where i is the output file index. See smi_split --help for details.

Fragment the SMILES files

These files can be fragmented in parallel:

% ls ChEMBL_CYP3A4_hERG.*.smi | qsub mmpdb fragment -j 1

I used the -j 1 flag to have mmpdb fragment use only a single thread, otherwise each of the four fragment commands will use 4 threads even though my laptop only has 4 cores. You should adjust the value to match the resources available on your compute node.

The parallel command doesn't forward output until the program is done, so it takes a while to see messages like:

Using 'ChEMBL_CYP3A4_hERG.0002.fragdb' as the default --output file.
Fragmented record 249/2026 (12.3%)[15:04:16] Conflicting single bond
directions around double bond at index 5.
[15:04:16]   BondStereo set to STEREONONE and single bond directions set to NONE.

If no -o/--output is specified, the fragment command uses a named based on the input name, for example, if the input file is ChEMBL_CYP3A4_hERG.0002.smi then the default output file is ChEMBL_CYP3A4_hERG.0002.mmpdb.

Merge the fragment files

NOTE: This step is only needed if you want to use the merged file as a --cache for new fragmentation. The fragdb_constants and fragdb_partition commands can work directly on the un-merged fragdb files.

About 28 minutes later I have 10 fragdb files:

% ls -l ChEMBL_CYP3A4_hERG.*.fragdb
-rw-r--r--  1 dalke  admin  17862656 Feb 10 15:17 ChEMBL_CYP3A4_hERG.0000.fragdb
-rw-r--r--  1 dalke  admin  38285312 Feb 10 15:27 ChEMBL_CYP3A4_hERG.0001.fragdb
-rw-r--r--  1 dalke  admin  15024128 Feb 10 15:16 ChEMBL_CYP3A4_hERG.0002.fragdb
-rw-r--r--  1 dalke  admin  15929344 Feb 10 15:16 ChEMBL_CYP3A4_hERG.0003.fragdb
-rw-r--r--  1 dalke  admin  18063360 Feb 10 15:23 ChEMBL_CYP3A4_hERG.0004.fragdb
-rw-r--r--  1 dalke  admin  20586496 Feb 10 15:24 ChEMBL_CYP3A4_hERG.0005.fragdb
-rw-r--r--  1 dalke  admin  24911872 Feb 10 15:26 ChEMBL_CYP3A4_hERG.0006.fragdb
-rw-r--r--  1 dalke  admin  16875520 Feb 10 15:28 ChEMBL_CYP3A4_hERG.0007.fragdb
-rw-r--r--  1 dalke  admin  12451840 Feb 10 15:28 ChEMBL_CYP3A4_hERG.0008.fragdb
-rw-r--r--  1 dalke  admin  11010048 Feb 10 15:29 ChEMBL_CYP3A4_hERG.0009.fragdb

I'll merge these with the fragdb_merge command:

% mmpdb fragdb_merge ChEMBL_CYP3A4_hERG.*.fragdb -o ChEMBL_CYP3A4_hERG.fragdb
Merge complete. #files: 10 #records: 18759 #error records: 1501

This took about 4 seconds.

Use the merged fragment file as cache

The merged file can be used a a cache file for future fragmentations, such as:

% ls ChEMBL_CYP3A4_hERG.*.smi | \
    qsub mmpdb fragment --cache ChEMBL_CYP3A4_hERG.fragdb -j 1

This re-build using cache takes about 20 seconds.

Distributed indexing

The mmpdb index command is single-threaded. It's possible to parallelize indexing by partitioning the fragments with the same constant SMILES into their own fragdb data sets, indexing those files, then merging the results back into a full MMP database.

Note: the merge command can only be used to merge MMP databases with distinct constants. It cannot be used to merge arbitrary MMP databases.

Note: the MMP database only stores aggregate information about pair properties, and the aggregate values cannot be meaningfully merged, so the merge command will ignore any properties in the database.

Partitioning on all constants

The mmpdb fragdb_partition command splits one or more fragment databases into N smaller files. All of the fragmentations with the same constant are in the same file.

NOTE: the fragdb files from the fragment command have a slightly different structure than the ones from the partition command. The fragment fragdb files only contain the input records that were fragmented. Each partition fragdb file contains all of the input records from the input fragment file(s). This is needed to handle 1-cut hydrogen matched molecular pairs.

If you specify multiple fragdb files then by default the results are put into files matching the template "partition.{i:04d}.fragdb", as in the following:

% mmpdb fragdb_partition ChEMBL_CYP3A4_hERG.*.fragdb
Analyzed 'ChEMBL_CYP3A4_hERG.0000.fragdb': #constants: 48895 #fragmentations: 109087
Analyzed 'ChEMBL_CYP3A4_hERG.0001.fragdb': #constants: 70915 #fragmentations: 212777
Analyzed 'ChEMBL_CYP3A4_hERG.0002.fragdb': #constants: 52370 #fragmentations: 100594
Analyzed 'ChEMBL_CYP3A4_hERG.0003.fragdb': #constants: 49021 #fragmentations: 103350
Analyzed 'ChEMBL_CYP3A4_hERG.0004.fragdb': #constants: 52318 #fragmentations: 112930
Analyzed 'ChEMBL_CYP3A4_hERG.0005.fragdb': #constants: 55977 #fragmentations: 123463
Analyzed 'ChEMBL_CYP3A4_hERG.0006.fragdb': #constants: 64083 #fragmentations: 164259
Analyzed 'ChEMBL_CYP3A4_hERG.0007.fragdb': #constants: 51605 #fragmentations: 114113
Analyzed 'ChEMBL_CYP3A4_hERG.0008.fragdb': #constants: 44149 #fragmentations: 80613
Analyzed 'ChEMBL_CYP3A4_hERG.0009.fragdb': #constants: 35889 #fragmentations: 69029
Analyzed 10 databases. Found #constants: 467865 #fragmentations: 1190215
Exporting 1 constants to 'partition.0000.fragdb' (#1/10, weight: 334589647)
Exporting 1 constants to 'partition.0001.fragdb' (#2/10, weight: 270409141)
Exporting 1 constants to 'partition.0002.fragdb' (#3/10, weight: 225664391)
Exporting 1 constants to 'partition.0003.fragdb' (#4/10, weight: 117895691)
Exporting 77977 constants to 'partition.0004.fragdb' (#5/10, weight: 52836587)
Exporting 77978 constants to 'partition.0005.fragdb' (#6/10, weight: 52836587)
Exporting 77975 constants to 'partition.0006.fragdb' (#7/10, weight: 52836586)
Exporting 77976 constants to 'partition.0007.fragdb' (#8/10, weight: 52836586)
Exporting 77977 constants to 'partition.0008.fragdb' (#9/10, weight: 52836586)
Exporting 77978 constants to 'partition.0009.fragdb' (#10/10, weight: 52836586)

The command's --template option lets you specify how to generate the output filenames.

Why are there so few constants in first files and so many in the other? And what are the "weight"s?

I'll use the fragdb_constants command to show the distinct constants in each file and the number of occurrences.

% mmpdb fragdb_constants partition.0000.fragdb
constant        N
*C      25869

That's a lot of methyls (25,869 to be precise).

The indexing command does N*(N-1)/2 indexing comparisions, plus a 1-cut hydrogen match, so the cost estimate for the methyls is 25869*(25869-1)/2+1 = 334589647, which is the weight value listed above.

I'll next list the three most common and least constants in ChEMBL_CYP3A4_hERG.0004.fragdb:

% mmpdb fragdb_constants partition.0004.fragdb --limit 3
constant        N
*C.*C.*OC       7076
*C.*Cl  4388
*C.*C.*CC       3261
% mmpdb fragdb_constants partition.0004.fragdb | tail -3
*n1nnnc1SCC(=O)Nc1nc(-c2ccc(Cl)cc2)cs1  1
*n1nnnc1SCc1nc(N)nc(N2CCOCC2)n1 1
*n1s/c(=N/C)nc1-c1ccccc1        1

The values of N are much smaller, so the corresponding weight is significantly smaller.

By default the partition command tries to split the constants evenly (by weight) across -n / --num-files files, defaulting to 10, which combined with the quadratic weighting is why the first few files have only a single, very common, constant, and why all of the "1" counts are used to fill space in the remaining files

You can alternatively use --max-weight to set an upper bound for the weights in each file. In this example I'll use the merged fragdb file from the previous step:

% mmpdb fragdb_partition ChEMBL_CYP3A4_hERG.fragdb --max-weight 50000000
Analyzed 'ChEMBL_CYP3A4_hERG.fragdb': #constants: 467865 #fragmentations: 1190215
Exporting 1 constants to 'ChEMBL_CYP3A4_hERG-partition.0000.fragdb' (#1/11, weight: 334589647)
Exporting 1 constants to 'ChEMBL_CYP3A4_hERG-partition.0001.fragdb' (#2/11, weight: 270409141)
Exporting 1 constants to 'ChEMBL_CYP3A4_hERG-partition.0002.fragdb' (#3/11, weight: 225664391)
Exporting 1 constants to 'ChEMBL_CYP3A4_hERG-partition.0003.fragdb' (#4/11, weight: 117895691)
Exporting 10 constants to 'ChEMBL_CYP3A4_hERG-partition.0004.fragdb' (#5/11, weight: 49918518)
Exporting 11 constants to 'ChEMBL_CYP3A4_hERG-partition.0005.fragdb' (#6/11, weight: 49916276)
Exporting 13 constants to 'ChEMBL_CYP3A4_hERG-partition.0006.fragdb' (#7/11, weight: 49899719)
Exporting 7 constants to 'ChEMBL_CYP3A4_hERG-partition.0007.fragdb' (#8/11, weight: 49896681)
Exporting 43 constants to 'ChEMBL_CYP3A4_hERG-partition.0008.fragdb' (#9/11, weight: 49893145)
Exporting 9 constants to 'ChEMBL_CYP3A4_hERG-partition.0009.fragdb' (#10/11, weight: 49879752)
Exporting 467768 constants to 'ChEMBL_CYP3A4_hERG-partition.0010.fragdb' (#11/11, weight: 17615427)

If you specify a single fragdb filename then the default output template is "{prefix}-partition.{i:04}.fragdb" where "{prefix}" is the part of the fragdb filename before its extension. The idea is to help organize those files together.

Odds are, you don't want to index the most common fragments. The next two sections help limits which constants are used.

Selecting constants

As you saw, the mmpdb fragdb_constants command can be used to list the constants. It can also be used to list a subset of the constants.

The count for each constant quickly decreases to something a bit more manageable.

% mmpdb fragdb_constants ChEMBL_CYP3A4_hERG.*.fragdb --limit 20
constant        N
*C      25869
*C.*C   23256
*C.*C.*C        21245
*C.*C.*O        15356
*C.*O   8125
*C.*C.*OC       7076
*C.*OC  6878
*F      6201
*C.*F   6198
*C.*c1ccccc1    5124
*C.*O.*O        5117
*c1ccccc1       5073
*OC     4944
*Cl     4436
*C.*Cl  4388
*O      4300
*F.*F   4281
*C.*F.*F        3935
*C.*C.*F        3656
*F.*F.*F        3496

I'll select those constants which occur only 2,000 matches or fewer, and limit the output to the first 5.

% mmpdb fragdb_constants ChEMBL_CYP3A4_hERG.*.fragdb --max-count 2000 --limit 5
constant        N
*C.*CC.*O       1954
*C.*C(F)(F)F    1915
*C.*C.*OC(C)=O  1895
*C(F)(F)F       1892
*Cl.*Cl 1738

or count the number of constants which only occur once (the 1-cut constants might match with a hydrogen substitution while the others will never match). I'll use --no-header so the number of lines of output matches the number of constants:

% mmpdb fragdb_constants ChEMBL_CYP3A4_hERG.fragdb --max-count 1 --no-header | wc -l
  370524

These frequent constants are for small fragments. I'll limit the selection to constants where each part of the constant has at least 5 heavy atoms:

% mmpdb fragdb_constants ChEMBL_CYP3A4_hERG.*.fragdb --min-heavies-per-const-frag 5 --limit 4
constant        N
*c1ccccc1       5073
*c1ccccc1.*c1ccccc1     1116
*Cc1ccccc1      1050
*c1ccc(F)cc1    921

I'll also require N be between 10 and 1000.

% mmpdb fragdb_constants ChEMBL_CYP3A4_hERG.*.fragdb --min-heavies-per-const-frag 5 \
   --min-count 10 --max-count 1000 --no-header | wc -l
1940

That's a much more tractable size for this example.

As you saw earlier, the mmpdb fragdb_partition command by default partitions on all constants. Alternatively, use the --constants flag to pass in a list of constants to use. This can be a file name, or - to accept constants from stdin, as in the following three lines:

% mmpdb fragdb_constants ChEMBL_CYP3A4_hERG.*.fragdb --min-heavies-per-const-frag 5 \
     --min-count 10 --max-count 1000 | \
     mmpdb fragdb_partition ChEMBL_CYP3A4_hERG.*.fragdb --constants -
Exporting 1 constants to 'ChEMBL_CYP3A4_hERG.0000.fragdb' (weight: 423661)
Exporting 1 constants to 'ChEMBL_CYP3A4_hERG.0001.fragdb' (weight: 382376)
Exporting 109 constants to 'ChEMBL_CYP3A4_hERG.0002.fragdb' (weight: 382044)
Exporting 261 constants to 'ChEMBL_CYP3A4_hERG.0003.fragdb' (weight: 382013)
Exporting 261 constants to 'ChEMBL_CYP3A4_hERG.0004.fragdb' (weight: 382013)
Exporting 260 constants to 'ChEMBL_CYP3A4_hERG.0005.fragdb' (weight: 382010)
Exporting 261 constants to 'ChEMBL_CYP3A4_hERG.0006.fragdb' (weight: 382010)
Exporting 262 constants to 'ChEMBL_CYP3A4_hERG.0007.fragdb' (weight: 382010)
Exporting 262 constants to 'ChEMBL_CYP3A4_hERG.0008.fragdb' (weight: 382009)
Exporting 262 constants to 'ChEMBL_CYP3A4_hERG.0009.fragdb' (weight: 382003)

Note: the --constants parser expects the first line to be a header, which is why I don't use --no-header in the fragdb_constants command. Alternatively, also use --no-header in the fragdb_partition command if the input does not have a header.

Partitioning in parallel

Partioning large data sets may take significant time because the export process is single-threaded.

The fragdb_partition command can be configured to export only subset of the partitions using a simple round-robin scheme. If you specify --task-id n and --num-tasks N then the given fragdb_partition will only export partitions i such that i % N == n.

The expected approach is to create a single constants files which will be shared by multiple partition commands.

% mmpdb fragdb_constants ChEMBL_CYP3A4_hERG.*.fragdb --min-heavies-per-const-frag 5 \
     --min-count 10 --max-count 1000 -o constants.dat

The following splits the job across two partition commands, with task ids 0 and 1, respectively:

% mmpdb fragdb_partition ChEMBL_CYP3A4_hERG.*.fragdb --constants constants.dat --task-id 0 --num-tasks 2
Exporting 1 constants to 'partition.0000.fragdb' (#1/10, weight: 423661)
Exporting 109 constants to 'partition.0002.fragdb' (#3/10, weight: 382044)
Exporting 261 constants to 'partition.0004.fragdb' (#5/10, weight: 382013)
Exporting 261 constants to 'partition.0006.fragdb' (#7/10, weight: 382010)
Exporting 262 constants to 'partition.0008.fragdb' (#9/10, weight: 382009)
% mmpdb fragdb_partition ChEMBL_CYP3A4_hERG.*.fragdb --constants constants.dat --task-id 1 --num-tasks 2
Exporting 1 constants to 'partition.0001.fragdb' (#2/10, weight: 382376)
Exporting 261 constants to 'partition.0003.fragdb' (#4/10, weight: 382013)
Exporting 260 constants to 'partition.0005.fragdb' (#6/10, weight: 382010)
Exporting 262 constants to 'partition.0007.fragdb' (#8/10, weight: 382010)
Exporting 262 constants to 'partition.0009.fragdb' (#10/10, weight: 382003)

Use the --dry-run option to get an idea of how many files will be created:

% mmpdb fragdb_partition ChEMBL_CYP3A4_hERG.*.fragdb --constants constants.dat --dry-run
i       #constants      weight  filename
0       10      423661  'partition.0000.fragdb'
1       10      382376  'partition.0001.fragdb'
2       10      382044  'partition.0002.fragdb'
3       10      382013  'partition.0003.fragdb'
4       10      382013  'partition.0004.fragdb'
5       10      382010  'partition.0005.fragdb'
6       10      382010  'partition.0006.fragdb'
7       10      382010  'partition.0007.fragdb'
8       10      382009  'partition.0008.fragdb'
9       10      382003  'partition.0009.fragdb'

Indexing in parallel

The partitioned fragdb files can be indexed in parallel:

% ls partition.*.fragdb | qsub mmpdb index
WARNING: No --output filename specified. Saving to 'partition.0000.mmpdb'.
WARNING: No --output filename specified. Saving to 'partition.0001.mmpdb'.
WARNING: No --output filename specified. Saving to 'partition.0002.mmpdb'.
WARNING: No --output filename specified. Saving to 'partition.0003.mmpdb'.
WARNING: No --output filename specified. Saving to 'partition.0004.mmpdb'.
WARNING: No --output filename specified. Saving to 'partition.0005.mmpdb'.
WARNING: No --output filename specified. Saving to 'partition.0006.mmpdb'.
WARNING: No --output filename specified. Saving to 'partition.0007.mmpdb'.
WARNING: No --output filename specified. Saving to 'partition.0008.mmpdb'.
WARNING: No --output filename specified. Saving to 'partition.0009.mmpdb'.

(If you don't like these warning messages, use the --quiet flag.)

Merging partitioned mmpdb files

The last step is to merge the partitioned mmpdb files with the merge option, which only works if no two mmpdb files share the same constant:

% mmpdb merge partition.*.mmpdb -o ChEMBL_CYP3A4_hERG_distributed.mmpdb
[Stage 1/7] Merging compound records ...
[Stage 1/7] Merged 4428 compound records in 0.046 seconds.
[Stage 2/7] Merging rule_smiles tables ...
[Stage 2/7] Merged 3159 rule_smiles records in 0.030 seconds.
[Stage 3/7] Merging rule tables ...
[Stage 3/7] Merged 21282 rule records in 0.072 seconds.
[Stage 4/7] Merging environment_fingerprint records ...
[Stage 4/7] Merged 1753 environment_fingerprint records in 0.035 seconds.
[Stage 5/7] Merging rule environment records ...
[Stage 5/7] Merged 143661 rule environment records in 0.47 seconds.
[Stage 6/7] Merging constant_smiles and pair records ...
[Stage 6/7] Merged 893 constant SMILES and 203856 pair records in 0.26 seconds
[Stage 7/7] Indexed and analyzed the merged records in 0.33 seconds.
Merged 10 files in 1.3 seconds.

Let's take a look:

% mmpdb list ChEMBL_CYP3A4_hERG_distributed.mmpdb
                Name                 #cmpds #rules #pairs #envs  #stats  |-------- Title --------| Properties
ChEMBL_CYP3A4_hERG_distributed.mmpdb   4428  21282 203856 143661      0  Merged MMPs from 10 files <none>

Finally, I'll cross-check this with a normal mmpdb index. I need to create the same subset

% mmpdb fragdb_partition ChEMBL_CYP3A4_hERG.fragdb --constants constants.dat \
      -n 1 --template ChEMBL_CYP3A4_hERG_subset.fragdb
Exporting 1940 constants to 'ChEMBL_CYP3A4_hERG_subset.fragdb' (#1/1, weight: 3862149)

Then index the subset:

% mmpdb index ChEMBL_CYP3A4_hERG_subset.fragdb
WARNING: No --output filename specified. Saving to 'ChEMBL_CYP3A4_hERG_subset.mmpdb'.

And finally, compare the two:

% mmpdb list ChEMBL_CYP3A4_hERG_subset.mmpdb ChEMBL_CYP3A4_hERG_distributed.mmpdb
                Name                 #cmpds #rules #pairs #envs  #stats  |----------------- Title ------------------| Properties
     ChEMBL_CYP3A4_hERG_subset.mmpdb   4428  21282 203856 143661      0  MMPs from 'ChEMBL_CYP3A4_hERG_subset.fragdb' <none>
ChEMBL_CYP3A4_hERG_distributed.mmpdb   4428  21282 203856 143661      0  Merged MMPs from 10 files                    <none>

They are the same, except for the title.


History and Acknowledgements

The project started as a fork of the matched molecular pair program 'mmpa' written by Jameed Hussain, then at GlaxoSmithKline Research & Development Ltd.. Many thanks to them for contributing the code to the RDKit project under a free software license.

Since then it has gone through two rewrites before the 1.0 release. Major changes to the first version included:

  • performance improvements,

  • support for property prediction

  • environmental fingerprints

That version supported both MySQL and SQLite, and used the third-party "peewee.py" and "playhouse" code to help with for database portability. Many thanks to Charlies Leifer for that software.

The second version dropped MySQL support but added APSW support, which was already available in the peewee/playhouse modules. The major goals in version 2 were:

  • better support for chiral structures

  • canonical variable fragments, so the transforms are canonical on both the left-hand and right-hand sides. (Previously only the entire transform was canonical.)

The project then forked into three branches:

  1. The public GitHub branch, with a few improvements by Christian Kramer

  2. Andrew Dalke's crowd-funded branch which:

  • replaced the Morgan fingerprint-based hashed environment fingerprint with its canonical SMARTS equivalent, and a "pseudo-SMILES" which might be used in depictions
  • added Postgres support
  • added export methods to tab-separated and database dump formats
  1. Mahendra Awale's improvements for:
  • large-database mmpdb generation by partitioning on fragment constants
  • playbook generation

Roche funded Andrew Dalke to merge these three branches, resulting in mmpdb 3.0.


Copyright

The mmpdb package is copyright 2015-2023 by F. Hoffmann-La Roche Ltd and Andrew Dalke Scientific AB, and distributed under the 3-clause BSD license. See LICENSE for details.


License information

The software derives from software which is copyright 2012-2013 by GlaxoSmithKline Research & Development Ltd., and distributed under the 3-clause BSD license. To the best of our knowledge, mmpdb does not contain any of the mmpa original source code. We thank the authors for releasing this package and include their license in the credits. See LICENSE for details.

The file fileio.py originates from chemfp and is therefore copyright by Andrew Dalke Scientific AB under the MIT license. See LICENSE for details. Modifications to this file are covered under the mmpdb license.

About

A package to identify matched molecular pairs and use them to predict property changes.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%