azusakou/AbGraftBERT

AbGraftBERT

Introduction

Schematic illustration of AbGraftBERT

Dependencies

• pytorch==1.12.0
• fairseq==0.10.2
• numpy==1.23.3

Pretrain a model on sequence data

All preprocessed data is from ABGNN, with the code primarily based on this repository, thanks! For training, we can run:

bash pretrain-abgraftbert.sh

Finetune on sequence and structure co-design tasks

For experiment 1, raw data is from MEAN.

For experiment 2, raw data is from HSRN.

For experiment 3, raw data is from RefineGNN.

The finetuning scripts are following:

# for exp1
bash finetune-exp1.sh

# for exp2
bash finetune-exp2.sh

# for exp3
# have to additionally install pytorch_lightning, matplotlib, and igfold
bash finetune-exp3.sh
bash covid-optimize.sh

We can simply run the following code for inference:

python inference.py \
    --cdr_type ${CDR} \
    --cktpath ${model_ckpt_path}}/checkpoint_best.pt \
    --data_path ${dataset_path}