STOUT V2.0: SMILES to IUPAC name conversion using transformer models
Author Information
Author(s): Rajan Kohulan, Zielesny Achim, Steinbeck Christoph
Primary Institution: Friedrich Schiller University Jena
Hypothesis
Can a transformer-based model accurately translate SMILES notation into IUPAC names?
Conclusion
STOUT V2 demonstrates high accuracy in generating IUPAC names from SMILES strings, even for complex chemical structures.
Supporting Evidence
- STOUT V2 was trained on a dataset of nearly 1 billion SMILES strings.
- The model achieved high accuracy in generating IUPAC names.
- Character-level tokenisation improved performance compared to word-level tokenisation.
- Retraining is necessary when IUPAC nomenclature rules change.
Takeaway
This study shows that a computer program can help scientists name chemical compounds more easily by translating a special code (SMILES) into proper names (IUPAC).
Methodology
The model was trained on a dataset of nearly 1 billion SMILES strings and their corresponding IUPAC names using transformer models.
Potential Biases
Potential biases may arise from the datasets used, which may not represent all chemical structures equally.
Limitations
The model may struggle with very complex chemical structures and requires retraining when nomenclature rules change.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website