|
--- |
|
library_name: transformers |
|
tags: |
|
- safe |
|
- datamol-io |
|
- molecule-design |
|
- smiles |
|
- generated_from_trainer |
|
datasets: |
|
- sagawa/ZINC-canonicalized |
|
model-index: |
|
- name: SAFE_100M |
|
results: [] |
|
--- |
|
|
|
# SAFE_100M |
|
|
|
This model was trained from scratch on the ZINC dataset converted to SAFE format for molecule generation tasks. |
|
It achieves the following results on the evaluation set: |
|
|
|
- Loss: 0.3887 |
|
|
|
## Model description |
|
|
|
SAFE_100M is a transformer-based model designed for molecular generation tasks. It was trained on the ZINC dataset (https://huggingface.co./datasets/sagawa/ZINC-canonicalized), which has been converted to the SAFE (SMILES Augmented For Encoding) format. This format is specifically tailored for improved molecular representation in machine learning tasks. |
|
|
|
The model is intended to generate valid and diverse molecular structures, which can be useful in various applications such as drug discovery, materials science, and chemical engineering. |
|
|
|
This model utilizes the SAFE framework, which was introduced in the following paper: |
|
|
|
```bibtex |
|
@article{noutahi2024gotta, |
|
title={Gotta be SAFE: a new framework for molecular design}, |
|
author={Noutahi, Emmanuel and Gabellini, Cristian and Craig, Michael and Lim, Jonathan SC and Tossou, Prudencio}, |
|
journal={Digital Discovery}, |
|
volume={3}, |
|
number={4}, |
|
pages={796--804}, |
|
year={2024}, |
|
publisher={Royal Society of Chemistry} |
|
} |
|
``` |
|
|
|
We acknowledge and thank the authors for their valuable contribution to the field of molecular design. |
|
|
|
## Intended uses & limitations |
|
|
|
This model is primarily intended for: |
|
|
|
- Generating molecular structures |
|
- Exploring chemical space for drug discovery |
|
- Assisting in the design of new materials |
|
|
|
Limitations: |
|
|
|
- The model's output should be validated by domain experts before practical application |
|
- Generated molecules may not always be synthetically feasible |
|
- The model's knowledge is limited to the chemical space represented in the ZINC dataset |
|
|
|
## Training and evaluation data |
|
|
|
The model was trained on the ZINC dataset (https://huggingface.co./datasets/sagawa/ZINC-canonicalized), which was converted to the SAFE format. The ZINC dataset is a large collection of commercially available chemical compounds for virtual screening. |
|
|
|
## Training procedure |
|
|
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
|
|
- learning_rate: 0.0001 |
|
- train_batch_size: 100 |
|
- eval_batch_size: 100 |
|
- seed: 42 |
|
- gradient_accumulation_steps: 2 |
|
- total_train_batch_size: 200 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: linear |
|
- lr_scheduler_warmup_steps: 10000 |
|
- training_steps: 250000 |
|
|
|
### Framework versions |
|
|
|
- Transformers 4.44.2 |
|
- Pytorch 2.4.0+cu121 |
|
- Datasets 2.20.0 |
|
- Tokenizers 0.19.1 |
|
|
