This model is fine-tuned on the Tox21 dataset and is designed to classify chemical compounds using SELFIES (Self-referencing Embedded Strings) as input representations. It employs the APE (Atom Pair Encoding) tokenizer for tokenizing the input, with the vocabulary stored in the same repository as the model under the file name tokenizer.json. The model is intended for sequence classification tasks and should be loaded with the AutoModelForSequenceClassification class.

Model Details

Model Description

This is a 🤗 transformers model fine-tuned on the Tox21 dataset for classifying chemical compounds. It uses the SELFIES molecular representation format as input and tokenizes these inputs using the APE Tokenizer. The vocabulary for the APE tokenizer is stored in the file tokenizer.json, located in the same repository as the model.

Developed by: Miguelangel Leon
Funded by: This work was supported by national funds through FCT (Fundação para a Ciência e a Tecnologia), under the project - UIDB/04152/2020 (DOI:10.54499/UIDB/04152/2020) - Centro de Investigação em Gestão de Informação (MagIC)/NOVA IMS).
Model type: Sequence Classification
Language(s) (NLP): Not applicable (SELFIES molecular representation)
License: MIT
Finetuned from model [optional]: mikemayuare/SELFYAPE

Model Sources

Paper : Leon, M., Perezhohin, Y., Peres, F. et al. Comparing SMILES and SELFIES tokenization for enhanced chemical language modeling. Sci Rep 14, 25016 (2024). https://doi.org/10.1038/s41598-024-76440-8

Uses

Direct Use

This model can be used directly for classification tasks on chemical compounds. It is specifically designed for toxicity prediction tasks, and the inputs must be formatted as SELFIES.

Downstream Use

This model can be fine-tuned further for other chemical classification tasks if the downstream dataset also uses SELFIES representations.

Out-of-Scope Use

This model is not suited for tasks that do not involve molecular data or tasks that require natural language processing (NLP).

Bias, Risks, and Limitations

As this model has been trained on the Tox21 dataset, it may not generalize well to unseen chemical compounds that differ significantly from the training data. Moreover, since this model is designed for molecular data, it is not appropriate for use in non-chemical classification tasks.

Recommendations

Users should be aware of potential biases stemming from the training data (Tox21). Careful evaluation on the target chemical compounds is recommended to ensure the model's suitability for a given application.

How to Get Started with the Model

To use the model for classification, it must be loaded with the AutoModelForSequenceClassification class from 🤗 transformers. The APE tokenizer is required to process the input data, which should be formatted as SELFIES.

You can load the APE tokenizer and the model with the following steps:

python

# Install the APETokenizer from the repository
# !git clone https://github.com/mikemayuare/apetokenizer
# Load the tokenizer
from src.apetokenizer.ape_tokenizer import APETokenizer

tokenizer = APETokenizer()
tokenizer.load_vocabulary("tokenizer.json")

# Load the model
from transformers import AutoModelForSequenceClassification

model = AutoModelForSequenceClassification.from_pretrained("mikemayuare/SELFY-APE-tox21")