rxkit 0.1.2

Chemical reaction and molecule SMILES canonicalization toolkit

RxKit

Chemical reaction and molecule SMILES canonicalization toolkit powered by RDKit.

Acknowledgements / References

The reaction SMILES handling in this package (in particular the conversion between STANDARD, STANDARD_AGENT, STANDARD_WITH_TILDE and EXTENDED formats, the |f:...| fragment-info parser, and the multi-component compound merging logic) is inspired by and adapted from the following open-source projects:

• rxn4chemistry/rxn-chemutils — reaction SMILES utilities, including extended-reaction SMILES parsing and conversion.
• MolecularAI/reaction_utils — reaction preprocessing utilities from Molecular AI (AstraZeneca).
• rxn4chemistry/rxn-reaction-preprocessing — reaction preprocessing pipeline used by the IBM RXN project.

Thanks to the maintainers and contributors of these projects.

Features

• SMILES canonicalization — normalize molecule SMILES strings to canonical forms
• InChI-based canonicalization (inchify) — canonicalize via InChI round-trip for tautomer-insensitive matching
• Reaction processing (process_reaction) — canonicalize reaction SMILES, deduplicate agents, and extract product
• SMILES tokenizer — tokenize SMILES strings for sequence models
• Augmented SMILES — generate randomized SMILES for data augmentation

Installation

From PyPI (recommended)

pip install rxkit

From source

git clone https://github.com/CreamyLong/RxKit.git
cd RxKit
pip install -e .

Note: RDKit is a required dependency. If you encounter RDKit installation issues, install it via conda first:

conda install -c conda-forge rdkit
pip install rxkit

Quick Start

from rxkit.canonicalize import canonicalize_smiles, inchify, process_reaction, smi_tokenizer

# Canonicalize a single molecule
canonicalize_smiles("CCO")
# 'CCO'

# Tautomer-insensitive canonicalization via InChI
inchify("CC(=O)O")
# 'CC(=O)O'

# Process a reaction
process_reaction("C=C.CC(=O)O>>CC(=O)OC")
# 'C=C.CC(=O)O>>COC(C)=O'

# Tokenize for sequence models
smi_tokenizer("CC(=O)O")
# 'C C ( = O ) O'

API Reference

mol_utils

Function	Description
canonicalize_smiles(smiles)	Canonicalize a SMILES string, stripping atom map numbers. Returns empty string on failure.
inchify(smi, extended_tautomer_check=False)	Canonicalize via InChI round-trip. If the result contains disconnected fragments (.) — indicating broken ionic bonds — the original SMILES is returned. Set extended_tautomer_check=True for stricter tautomer handling.
smi_tokenizer(smi)	Tokenize a SMILES string into space-separated tokens.
augm_smile(smi)	Generate a randomized SMILES string for data augmentation.
getNumHeavyAtoms(smi)	Return the number of heavy (non-hydrogen) atoms.
get_longest_smiles(smis)	Given dot-separated SMILES, return the longest fragment (excluding PPh₃).
remove_isotope_information(rxn)	Strip isotope mass numbers from a reaction SMILES (e.g. [13CH3] → [CH3]).
is_valid_molecule(smi, allow_empty=False)	Return True if smi parses to a non-empty RDKit molecule. Accepts a list of SMILES (all must validate). allow_empty=True treats empty strings as valid.

rxn_utils

Function	Description
detect_rxn_type(rxn)	Auto-detect reaction SMILES format. Returns EXTENDED, STANDARD_agent, STANDARD_WITH_TILDE, STANDARD, or UNKNOWN.
process_reaction(rxn, with_agent=True, rxn_type=None)	Canonicalize a reaction SMILES. When with_agent=True, reagents that also appear in products are removed as agents. rxn_type is auto-detected if not provided. The output is always normalised to STANDARD format (A.B>>D).
is_valid_reaction(rxn, rxn_type=None, allow_empty_products=False, is_strict=False)	Return True if rxn is a structurally well-formed reaction SMILES — non-empty, recognized format, and every fragment is a valid molecule SMILES. Set allow_empty_products=True to allow empty product sides (e.g. partial templates). Set is_strict=True to additionally reject empty fragments (..), duplicate compounds within a side (CCO.CCO>>...), and the same molecule appearing on multiple sides (CCO>>CCO, CCO>Cc1ccccc1>...). Solvent-only templates (CCO>CCO>CCO) remain valid.
tilde_to_standard(rxn)	Replace every ~ in a tilde reaction SMILES with . (e.g. CC.O.[Na+]~[Cl-]>>CCO → CC.O.[Na+].[Cl-]>>CCO).
standard_to_tilde(rxn)	Naive string-level rewrite: replace . with ~ inside each side of a STANDARD reaction SMILES.
parse_tilde_reaction_smiles(rxn)	Parse a STANDARD_WITH_TILDE reaction SMILES into (reactants, agents, products) lists of compounds.
tilde_to_extended(rxn)	Convert a STANDARD_WITH_TILDE reaction SMILES to EXTENDED form.
parse_extended_reaction_smiles(rxn)	Parse an EXTENDED reaction SMILES into (reactants, agents, products) lists of compounds, merging fragments that share a group index back into a single dot-separated compound.
to_extended_reaction_smiles(reactants, agents, products)	Build an EXTENDED reaction SMILES from three lists of compounds. Compounds that already contain . are recorded in the trailing |f:...| fragment info.
extended_to_standard(rxn)	Drop the trailing |f:...| fragment info from an EXTENDED reaction SMILES.
standard_to_extended(rxn)	Build an EXTENDED reaction SMILES from a STANDARD reaction SMILES (each . becomes a separate fragment; no fragment groups are recorded because they cannot be inferred from the string).
extended_to_tilde(rxn)	Convert an EXTENDED reaction SMILES to STANDARD_WITH_TILDE form by joining fragments in each group with ~ (e.g. CC.O.[Na+].[Cl-]>>CCO |f:2.3| → CC.O.[Na+]~[Cl-]>>CCO).
split_smiles_and_fragment_info(rxn)	Split an (extended) reaction SMILES into the pure SMILES part and the fragment info suffix.
determine_fragment_groups(extended_info)	Parse a fragment info string like |f:0.2,5.6| into [[0, 2], [5, 6]].

Supported reaction SMILES types:

Type	Example
STANDARD	C=C.CC(=O)O>>CC(=O)OC
STANDARD_agent	C=C.CC(=O)O>O>CC(=O)OC
STANDARD_WITH_TILDE	CCO~C>>CCO.CC
EXTENDED	|f:0.1,2.3|C=C.CC(=O)O>>CC(=O)OC

The output of process_reaction is always normalised to STANDARD form (A.B>>D), with ~ and |f:...| markers stripped.

Extended / Tilde Reaction SMILES

When a compound is a multi-component system (e.g. an ionic solid like NaCl written as [Na+].[Cl-]), the EXTENDED format uses a trailing fragment info marker to record which .-separated fragments must be grouped into the same compound. In STANDARD_WITH_TILDE form the same grouping is expressed inline with ~.

from rxkit.canonicalize import extended_to_tilde, tilde_to_extended, parse_extended_reaction_smiles

extended = "CC.O.[Na+].[Cl-]>>CCO |f:2.3|"

# EXTENDED -> TILDE
extended_to_tilde(extended)
# 'CC.O.[Na+]~[Cl-]>>CCO'

# TILDE -> EXTENDED
tilde_to_extended("CC.O.[Na+]~[Cl-]>>CCO")
# 'CC.O.[Na+].[Cl-]>>CCO |f:2.3|'

# Parse to a structured representation
parse_extended_reaction_smiles(extended)
# (['CC', 'O', '[Na+].[Cl-]'], [], ['CCO'])

Fragment indices in the |f:...| block are global (i.e. they index into the dot-separated fragments of the whole reaction SMILES, going through reactants → reagents → products in order). Multi-fragment compounds are placed after single-fragment compounds in each parsed list, matching the convention used by rxn-chemutils.

Reaction Processing Details

process_reaction takes a reaction SMILES in reactants>reagents>products format:

• with_agent=True (default): Reagents are merged with reactants as precursors. Any molecule appearing in both precursors and products is treated as an agent and removed from both sides. This is useful for cleaning up reactions where catalysts/solvents are listed as reagents.
• with_agent=False: Reactants and products are canonicalized independently. Reagents are ignored.

Run Tests

pip install pytest
pytest tests/

License

MIT