TransFactor

Usage

1. Install

git clone https://github.com/marsico-lab/TransFactor.git

conda env create -f environment.yml

As creating an environment from a file fails on some devices, one can create an environment from scratch

conda create -n transfactor python=3.11.8 conda install pytorch==2.2.1 torchvision==0.17.1 torchaudio==2.2.1 -c pytorch pip install -r requirements.txt

2. Run hyperparameter optimization with Optuna

conda activate transfactor

python transfactor_optuna.py

3. Predict using saved models

conda activate transfactor

python predict.py --ckpt_path <path_to_model_checkpoint> --out_dir <out_dir>

4. Train model on new data with full hyperparameter optimization

The data is expected to be in a pickled pandas DataFrame with the sequence in a column named seq and the label in a column named label. conda activate transfactor python transfactor_optuna.py --data_config_path <path_to_dataset_pickle_file>

Reproduce results from manuscript

Please find the corresponding notebooks and scripts in /manuscript. To download the saved model checkpoints and alanine scan results, please download those here: https://1drv.ms/f/s!AhK2hrQjl9NtlIgf9ueMG5xV94JBUA?e=k1kHZF