rdcanon - SMARTS and Reaction SMARTS Canonicalization

Overview

rdcanon is a package designed for canonicalizing SMARTS and Reaction SMARTS templates. It reorders SMARTS to optimize querying speed. This optimization is invariant of atom mapping.

Installation

pip install rdcanon

Installing from source

Prerequisites

• Ensure you have rdkit installed (version > 2023.9.2).

• The following packages will be installed:

    rdkit > 2023.09.1
  
    matplotlib
  
    lark
  
    numpy
  
    networkx
  
    scikit-learn (optional, for kde generation)
  
    ipykernel
  
    pandas
  
    openpyxl
  

1. Create or activate a virtual environment.
2. Clone the repository.
3. Install the package with the command:

pip install -e rdcanon

Usage

Sanitizing Individual SMARTS

To sanitize individual SMARTS:

from rdcanon import canon_smarts 

test_smarts = [
 "[$([NX3H,NX4H2+]),$([NX3](C)(C)(C))]1[CX4H]([CH2][CH2][CH2]1)[CX3](=[OX1])[OX2H,OX1-,N]",
 "[$([NX3H2,NX4H3+]),$([NX3H](C)(C))][CX4H]([*])[CX3](=[OX1])[OX2H,OX1-,N]",
 "[CX3](=O)[OX1H0-,OX2H1]",
 "[CX3](=[OX1])[OX2][CX3](=[OX1])",
 "[N&H2&+0:4]-[C&H1&+0:2](-[C&H2&+0:8])-[O&H1&+0:3]"
]

# The second parameter is optional and flags whether atom mapping should be returned (defaults to False)
for smarts in test_smarts:
 print(smarts, canon_smarts(smarts), canon_smarts(smarts, True))

Sanitizing Reaction SMARTS

For sanitizing reaction SMARTS:

from rdcanon import canon_reaction_smarts

Unit Testing

To run all unit tests:

python rdcanon_tests.py

note: currently, "TestRecursive.test_validate_recursive_against_database" is expected to fail at: "[#1]C&X3=[O&X1] [#1,#6][C&X3;H1,H2]=[O&X1]". This is due to a strange edge case with merging query Hs, explicit hydrogens, and explicit/implicit connections.

Current Limitations

No consolidation or expansion of atomic queries is performed automatically, but a mechanism is provided to allow the user to systematically replace canonicalized atomic queries with an input dictionary (e.g., {"[O;H1]": "[O;H1;+0]"} would replace the canonicalized variant of [O;H1] with the canonicalized variant of [O;H1;+0]).

Replacement dictionaries should be processed first using

canon_repl_dict = gen_canon_repl_dict(repl_dict)

before passing as an argument into canon_smarts.

Chirality or directionality beyond tetrahedral centers and cis/trans isomerism is not currently supported.

Manuscript Figures and Tests

All data can be found in the manuscripts/data directory.

To create the bar charts of Figure 1, use the notebook within the manuscript directory named "prim_frequencies.ipynb".

To run the subgraph isomorphism experiments of Figure 3, use the notebook within the manuscript directory named "generate_plots_substruct_match.ipynb".

To run the template application experiments of Figure 4, use the notebook within the manuscript directory named "gen_plots_run_reactants_20240104.ipynb".

To run the retrosynthetic analysis experiments of Figure 5, use the notebook within the manuscript directory named "gen_retrosim_plots_20240105.ipynb".

RDCanon Files

The main workflow consists of two files, main.py and token_parser.py. The main.py file calls token_parser.py to parse and score atomic queries.

The files askcos_prims.py, drugbank_prims_with_nots.py, np_prims.py, and pubchem_prims.py are 4 query primitive frequency dictonaries, which are used for embedding leaf nodes in query trees.

The rdcanon_tests.py file contains all of the test cases using the abseil interface.

Finally, utils.py contains some helper functions for testing and plotting.