Smiles inchi round trip validation

SMILES–InChI Round-Trip Validation

Summary

Validate the chemical correctness and consistency of local SMILES↔InChI conversions by performing bidirectional conversions on test molecules and comparing structural equivalence. This ensures that a ComputeConverter subclass using RDKit produces chemically valid and reversible transformations without relying on external web services.

When to use

You have implemented a new RDKit-based ComputeConverter for SMILES↔InChI conversions and need to verify that the conversion methods preserve molecular structure integrity across round-trip transformations (SMILES → InChI → SMILES or vice versa) before registering it in the MSMetaEnhancer pipeline.

When NOT to use

• You are validating web service converters (e.g., CTS, CIR, PubChem) — use their API response schemas and rate-limit handling instead.
• Input molecules contain features unsupported by RDKit (e.g., exotic metals, unusual bonding) — RDKit may fail to parse or generate InChI for these.
• Your goal is to benchmark speed or throughput rather than correctness — round-trip validation is a correctness check, not a performance benchmark.

Inputs

• Reference SMILES strings (e.g., 'CC(=O)O', 'c1ccccc1')
• InChI strings generated by RDKit from test SMILES
• MSMetaEnhancer ComputeConverter subclass source code

Outputs

• pytest test report showing pass/fail for each round-trip conversion
• Validated converter implementation with confirmed chemical equivalence
• Assertion results for output dictionary structure and content correctness

How to apply

Create a pytest test file that instantiates the converter and tests conversion accuracy by: (1) selecting a set of reference SMILES inputs with known chemical structures (e.g., simple organic molecules, isomers); (2) converting SMILES to InChI using RDKit's Chem.inchi.MolToInchi; (3) converting the InChI back to canonical SMILES using RDKit's Chem.MolFromInchi and Chem.MolToSmiles with isomeric=False; (4) comparing the canonical SMILES output against the input canonical form to detect structural loss or artifact generation; (5) verifying output dictionary structure contains correct target attribute keys; (6) running pytest to confirm the converter passes all chemical correctness assertions and does not break existing tests.

Related tools

• RDKit (Performs local SMILES parsing (Chem.MolFromSmiles), InChI generation (Chem.inchi.MolToInchi), and canonical SMILES export for round-trip validation)
• pytest (Executes test suite to validate converter instantiation, conversion accuracy, output structure, and chemical correctness of round-trip transformations)
• MSMetaEnhancer (Framework providing ComputeConverter base class and registration mechanism for the validated SMILES–InChI converter) — https://github.com/RECETOX/MSMetaEnhancer

Examples

pytest tests/test_rdkit_converter.py -v

Evaluation signals

• Converter instantiation succeeds without import or initialization errors.
• Round-trip SMILES → InChI → canonical SMILES produces identical or chemically equivalent canonical SMILES output (e.g., Tautomers and stereoisomers handled consistently).
• Output dictionary keys match the target attribute names specified in the conversions list (e.g., 'inchi', 'canonical_smiles').
• All pytest assertions pass for reference SMILES test set covering simple, aromatic, and chiral molecules.
• pytest run confirms no regression in existing MSMetaEnhancer tests after converter registration.

Limitations

• RDKit may fail or produce invalid InChI for molecules with exotic metal centers, radical states, or non-standard bonding not representable in standard chemical structure formats.
• Round-trip conversion is sensitive to SMILES canonicalization settings; non-canonical input SMILES may not round-trip to identical strings, requiring comparison via canonical form or InChIKey.
• InChI generation from SMILES may lose stereo or charge information depending on RDKit configuration; test cases should document expected information loss.
• The validation approach does not cover dynamic properties (e.g., tautomeric stability, reactivity) that web services might infer from cheminformatics models.

Evidence

• [other] Implement conversion methods using RDKit to perform local molecular structure transformations (e.g., parse SMILES strings with RDKit Chem.MolFromSmiles and generate InChI with Chem.inchi.MolToInchi).: "Implement conversion methods using RDKit to perform local molecular structure transformations (e.g., parse SMILES strings with RDKit Chem.MolFromSmiles and generate InChI with Chem.inchi.MolToInchi)."
• [other] Create a pytest test file that validates converter instantiation and tests conversion accuracy against reference SMILES inputs, verifying output structure and chemical correctness.: "Create a pytest test file that validates converter instantiation and tests conversion accuracy against reference SMILES inputs, verifying output structure and chemical correctness."
• [other] MSMetaEnhancer uses RDKit as a reference implementation for implementing ComputeConverter subclasses that perform local chemical structure conversions.: "MSMetaEnhancer uses RDKit as a reference implementation for implementing ComputeConverter subclasses that perform local chemical structure conversions."
• [other] Run pytest to confirm existing tests still pass and new converter tests execute successfully.: "Run pytest to confirm existing tests still pass and new converter tests execute successfully."
• [readme] All functionality is tested with the pytest framework.: "All functionality is tested with the pytest framework."