---
license: apache-2.0
metrics:
- accuracy
- bleu
pipeline_tag: text2text-generation
tags:
- chemistry
- biology
- medical
- smiles
- iupac
- text-generation-inference
widget:
- text: ethanol
  example_title: CCO
---
# IUPAC2SMILES-canonical-small

IUPAC2SMILES-canonical-small was designed to accurately translate IUPAC chemical names to SMILES. 

## Model Details

### Model Description

IUPAC2SMILES-canonical-small is based on the MT5 model with optimizations in implementing different tokenizers for the encoder and decoder. 
- **Developed by:** Knowladgator Engineering
- **Model type:** Encoder-Decoder with attention mechanism
- **Language(s) (NLP):** SMILES, IUPAC (English)
- **License:** Apache License 2.0

### Model Sources
- **Paper:** coming soon
- **Demo:** [ChemicalConverters](https://huggingface.co/spaces/knowledgator/ChemicalConverters)

## Quickstart
Firstly, install the library:
```commandline
pip install chemical-converters
```
### IUPAC to SMILES
#### To perform simple translation, follow the example:
```python
from chemicalconverters import NamesConverter

converter = NamesConverter(model_name="knowledgator/IUPAC2SMILES-canonical-small")
print(converter.iupac_to_smiles('ethanol'))
print(converter.iupac_to_smiles(['ethanol', 'ethanol', 'ethanol']))
```
```text
['CCO']
['CCO', 'CCO', 'CCO']
```
#### Processing in batches:
```python
from chemicalconverters import NamesConverter

converter = NamesConverter(model_name="knowledgator/IUPAC2SMILES-canonical-small")
print(converter.iupac_to_smiles(["buta-1,3-diene" for _ in range(10)], num_beams=1, 
                                process_in_batch=True, batch_size=1000))
```
```text
['<SYST>C=CC=C', '<SYST>C=CC=C'...]
```
Our models also predict IUPAC styles from the table:

| Style Token | Description                                                                                        |
|-------------|----------------------------------------------------------------------------------------------------|
| `<BASE>`    | The most known name of the substance, sometimes is the mixture of traditional and systematic style |
| `<SYST>`    | The totally systematic style without trivial names                                                 |
| `<TRAD>`    | The style is based on trivial names of the parts of substances                                     |

## Bias, Risks, and Limitations

This model has limited accuracy in processing large molecules and currently, doesn't support isomeric and isotopic SMILES.

### Training Procedure

<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->

The model was trained on 100M examples of SMILES-IUPAC pairs with lr=0.0003, batch_size=1024 for 2 epochs.

## Evaluation

| Model                               | Accuracy | BLEU-4 score | Size(MB) |
|-------------------------------------|---------|------------------|----------|
| IUPAC2SMILES-canonical-small        |88.9%    |0.966             |23        |
| IUPAC2SMILES-canonical-base         |93.7%    |0.974             |180       |
| STOUT V2.0\*                        |68.47%   |0.92              |128       |
*According to the original paper https://jcheminf.biomedcentral.com/articles/10.1186/s13321-021-00512-4

## Citation
Coming soon.

## Model Card Authors

[Mykhailo Shtopko](https://huggingface.co/BioMike)

## Model Card Contact

info@knowledgator.com