This repository contains all code and data necessary to replicate the results presented in the publication Transformers enable accurate prediction of acute and chronic chemical toxicity in aquatic organisms.

To use the best trained models, refer to: TRIDENT

https://hub.docker.com/repository/docker/styrbjornk/streamlit-app-trident-serve-v4.1/general

To replicate the study, refer to the documentation under the development section.

For very extensive predictions (>100 MB) consider cloning this repo and follow the tutorials under tutorials (requires basic python understanding).

Clone this repository:

git clone https://github.com/StyrbjornKall/TRIDENT

Replicate entire study Contains all packages required to reproduce this study:

conda env create -f trident_development_environment.yml

data contains all preprocessed data used for training our nine fine-tuned EC50, EC10 and combined models. Also contains QSAR comparison data.

development contains all code needed to replicate the findings presented in the publication.

TRIDENT contains the nine fine-tuned Deep Neural Network modules for the models. For the fine-tuned transformer (RoBERTa) modules, refer to Huggingface model-hub.

tutorials contains very simple tutorial notebooks for running inference using the fine-tuned models. Written in order to minimize programmatic interference so that very basic python knowledge suffice.

Refer to each sections README for further descriptions.

When using any of our models, please cite us!

@article{
doi:10.1126/sciadv.adk6669,
author = {Mikael Gustavsson  and Styrbjörn Käll  and Patrik Svedberg  and Juan S. Inda-Diaz  and Sverker Molander  and Jessica Coria  and Thomas Backhaus  and Erik Kristiansson },
title = {Transformers enable accurate prediction of acute and chronic chemical toxicity in aquatic organisms},
journal = {Science Advances},
volume = {10},
number = {10},
pages = {eadk6669},
year = {2024},
doi = {10.1126/sciadv.adk6669},
URL = {https://www.science.org/doi/abs/10.1126/sciadv.adk6669}}