sr-smiles 0.1.2

A Python library to convert reaction SMILES into superimposed reaction (sr) SMILES.

Overview | Installation | Usage | Contributing | Citation | References

sr-SMILES: Superimposed Reaction SMILES

sr-SMILES is a Python library for transforming reaction SMILES into a more compact and change-aware representation called superimposed reaction (sr) SMILES. This representation explicitly encodes changes in chemical reactions, making it suitable for machine learning and data-driven applications.

---

Overview

The sr‑SMILES is inspired by the Condensed Graph of Reaction (CGR) representation, a concept originating from graph‑based cheminformatics [1].

In a classical CGR, the reactant and product graphs of a chemical reactions are superimposed and represented as a single unified graph. Atoms common to both sides are merged, and bonds are annotated with their changes (e.g., “single → double”, “added”, or “removed”).

sr‑SMILES brings this concept to the string domain, making it suitable for language modeling applications. Instead of representing reactions as separate reactants and products, sr‑SMILES combines them into a compact, local‑change‑aware representation that explicitly encodes how atoms and bonds transform. It is applicable to any organic reaction of the form {reactant(s)}>>{product(s)}.

While atom mappings are required to perform the transformation, the library provides workarounds for unmapped or partially mapped reactions by integrating atom‑mapping tools such as RXNMapper [2].

Let's take a look at an example:

👉 Notice how the sr‑SMILES is more compact and explicitly encodes where atoms and bonds change during the reaction.

---

Installation

pip install sr-smiles

---

Usage

Your sr-SMILES Toolkit: Seamless Reaction Conversions

The simplest use case involves mapped and balanced reactions. But don’t worry, the library also handles unmapped or unbalanced cases.

There are several ways to use sr‑SMILES depending on your workflow:

1. Core functions (simple, flexible)
   • rxn_to_sr()
   • sr_to_rxn()
2. Wrapper classes (convenient for bulk data)
   • RxnToSr()
   • SrToRxn()
3. CLI (file-based workflows)
   • rxn2sr
   • sr2rxn

In the following sections, we’ll walk through basic examples for each option.

📘 For more detailed setups and parameter guidance, see the example notebook.

Let’s start with the imports:

from sr_smiles import SrToRxn, RxnToSr, sr_to_rxn, rxn_to_sr

1. Core functions (rxn_to_sr() and sr_to_rxn())

These are the best place to start when exploring sr‑SMILES. They provide a simple, direct way to understand how the library transforms reactions between RXN and sr-SMILES.

rxn_smiles = "[F-:6].[Br:1][C@:2]([H:5])([CH3:3])[NH2:4]>>[Br-:1].[CH3:3][C@:2]([H:5])([F:6])[NH2:4]"

sr_smiles = rxn_to_sr(
    rxn_smiles
)
rxn_back_with_mapping = sr_to_rxn(
    sr_smiles,
    add_atom_mapping=True  # optionally, to show atom mapping in the output smiles, defaults to False
)

print(f"RXN SMILES (original input):\n\t{rxn_smiles}\n")
print(f"sr-SMILES (without mapping):\n\t{sr_smiles}\n")
print(f"RXN SMILES (without mapping numbers):\n\t{rxn_back_without_mapping}\n")

RXN SMILES (original input):
	[F-:6].[Br:1][C@:2]([H:5])([CH3:3])[NH2:4]>>[Br-:1].[CH3:3][C@:2]([H:5])([F:6])[NH2:4]

sr-SMILES (without mapping):
	{[F-]|[F]}{~|-}{[C@]|[C@@]}({-|~}{[Br]|[Br-]})([H])([CH3])[NH2]

RXN SMILES (without mapping numbers):
	O=C(C#C[H])[H]>>[O+]#[C-].C(#C[H])[H]

2. Wrapper classes (RxnToSr() and SrToRxn())

These offer a convenient, efficient interface for practical use, as they are ideal for processing large datasets or handling more complex cases like unmapped and unbalanced reactions.

RXN to sr-SMILES:

import pandas as pd

rxn_list = [
    "[N:1]#[C:2][C@@:3]1([H:6])[C:4]([H:7])([H:8])[O:5]1>>[N:1]#[C:2][C@@:3]([C:4][H:7])([O:5][H:8])[H:6]",
    "[O:1]([C@@:2]([C:3](=[O:4])[H:9])([C:5]#[C:6][H:10])[H:8])[H:7]>>[O:1]([C@@:2]([C:3][O:4][H:9])([C:5]#[C:6][H:10])[H:8])[H:7]",
]

# using the RxnToSr transform on a list of reactions
transform_to_sr = RxnToSr()
sr_results = transform_to_sr(rxn_list)

# using the RxnToSr transform on a pd.DataFrame
df_data = pd.DataFrame({"reactions": rxn_list})
transform_to_sr = RxnToSr(
    rxn_col="reactions"    # <- in this case we need to specify the column name!
)
df_data["sr_smiles"] = transform_to_sr(rxn_list)

assert sr_results == df_data["sr_smiles"].tolist()
print(sr-SMILES:\n\t" + "\n\t".join(sr_results))

sr-SMILES:
	[N]#[C][C@@]1([H])[C]2([H]){-|~}[H]{~|-}[O]1{-|~}2
	[O]([C@@]([C]1{=|-}[O]{~|-}[H]{-|~}1)([C]#[C][H])[H])[H]

And sr-SMILES back to RXN:

transform_to_rxn = SrToRxn(add_atom_mapping=True)
rxns = transform_to_rxn(sr_results)

print("RXNs:\n\t" + "\n\t".join(rxns))

RXNs:
	[N:1]#[C:2][C@@:3]1([H:4])[C:5]([H:6])([H:7])[O:8]1>>[N:1]#[C:2][C@@:3]([H:4])([C:5][H:6])[O:8][H:7]
	[O:1]([C@@:2]([C:3](=[O:4])[H:5])([C:6]#[C:7][H:8])[H:9])[H:10]>>[O:1]([C@@:2]([C:3][O:4][H:5])([C:6]#[C:7][H:8])[H:9])[H:10]

What if your reactions aren’t atom‑mapped, and/or some are unbalanced? No problem, simply set balance_rxn=True and/or enable the integrated mapper with use_rxnmapper=True.

# Example list of reactions
rxn_list = [
    "CCO>>CC=O",                # unmapped
    "N>>NC",                    # unmapped and unbalanced
    "[NH3:1]>>[NH2:1][CH3:2]",  # unbalanced
]

# using the wrapper with a pandas DataFrame
df_data = pd.DataFrame({"reaction": rxn_list})
transform_to_sr_df = RxnToSr(
    rxn_col="reaction",
    use_rxnmapper=True,
    balance_rxn=True,
)
df_data["sr_smiles"] = transform_to_sr_df(df_data)
print("\nDataFrame with sr-SMILES:\n", df_data)

DataFrame with sr-SMILES:
                   reaction                        sr_smiles
0                CCO>>CC=O  [CH3]{[CH2]|[CH]}{-|=}{[OH]|[O]}
1                    N>>NC           {[NH3]|[NH2]}{~|-}[CH3]
2  [NH3:1]>>[NH2:1][CH3:2]           {[NH3]|[NH2]}{~|-}[CH3]

3. Command Line Interface (rxn2sr and sr2rxn)

If your prefer working with a CLI tool , that be it:

╭────────── 🚀 sr‑SMILES Converter v0.0.1 ────-─────╮
│                                                   │
│   👋 Welcome to sr‑SMILES                         │
│   Transforming Reaction SMILES ➡️ sr‑SMILES       │
│                                                   │
│   Input column:   'rxn_smiles'                    │
│   Output column:  'sr_smiles'                     │
│   Input file:     path/to/input.csv               │
│   Output file:    path/to/output.csv              │
│                                                   │
╰───────────────────────────────────────────────────╯

The forward transformation:

rxn2sr path/to/input.csv \              # required (input CSV)
    -o path/to/output.csv \             # optional output CSV
    --rxn-col rxn_smiles \              # name of the RXN SMILES column
    --sr-col sr_smiles \                # name of the new sr‑SMILES column
    --use-rxnmapper \                   # use RXNMapper if rxns are unmapped
    --keep-atom-mapping \               # preserve atom mapping
    --balance-rxn                       # enable reaction balancing

And the backward transformation:

sr2rxn output_sr.csv \                  # required (input CSV)
    -o path/to/output.csv \             # optional output CSV
    --sr-col sr_smiles \                # name of the sr‑SMILES column
    --rxn-col rxn_back                  # name of the new RXN SMILES column

---

Contributing

🤝 Contributions are welcome! Here's how to get started:

• Report bugs or edge cases by opening an issue.
• Submit fixes or features via a pull request.
• For local development, clone the repo and install with dev dependencies:

  git clone https://github.com/heid-lab/sr-smiles.git
  cd sr-smiles
  poetry install --with dev
  pre-commit install
  

---

Citation

[WIP]

---

References

[1] Heid, E.; Green, W. H. Machine Learning of Reaction Properties via Learned Representations of the Condensed Graph of Reaction. J. Chem. Inf. Model. 2022, 62 (9), 2101–2110. DOI: 10.1021/acs.jcim.1c00975

[2] Schwaller, P.; Hoover, B.; Reymond, J.‑L.; Strobelt, H.; Laino, T. Extraction of Organic Chemistry Grammar from Unsupervised Learning of Chemical Reactions. Sci. Adv. 2021, 7 (15), eabe4166. DOI: 10.1126/sciadv.abe4166