SimonKitSangChu / EsmTherm

EsmTherm

Installation

conda create -n esmtherm python=3.10 --file=environment.yml
conda activate esmtherm
pip install -e .

# To download the model checkpoints
gdown https://drive.google.com/drive/u/1/folders/1z3_IbeD8oORmLndqCCiqYRCxmbKNbWXb -O output_dir/checkpoint-best --folder

Dataset Preparation

Download the supplementary materials from Tsuboyama et al. (2023), extract K50_dG_Dataset1_Dataset2.csv and place it under data directory.

# analyze and filter dataset
python prebuild_dataset.py

# create training dataset
python build_dataset.py \
    --dataset_dir datasets/dataset \
    --csv datasets/analysis/filtered_data.csv \
    --split_csv datasets/wildtype_split.csv

Training and Evaluation

# training
python train.py \
    --dataset_dir datasets/dataset \
    --output_dir output_dir \
    --model_name facebook/esm2_t12_35M_UR50D

# evaluation
python evaluate.py \
    --model_name_or_path output_dir/checkpoint-best \
    --input_csv _your_input_csv_ \
    --output_csv __your_output_csv_

More instructions can be found with --help flag.