Skip to content

Latest commit

 

History

History
46 lines (35 loc) · 2.46 KB

drp.md

File metadata and controls

46 lines (35 loc) · 2.46 KB
Raw

Drug-Response Prediction

Drug-response prediction aims at predicting if a drug is responsive on cells, tissues, or patients, based on the gene expression profile.

Features

  • Supported models: TGDRP. More models will be supported in the future.
  • Supported dataset: GDSC, TCGA.
  • Supproted evaluation: ROC_AUC for classification and RMSE, MAE, $r^2$ index and pearson coefficient.

Data Preparation

Install GDSC and TCGA here (password is 7a6b, GDSC/ and TCGA/) and put them under datasets/drp/. Install supplementary data here (password is 7a6b, drp/) and put them under assets/drp/. Install the STRING database (homo sapiens) here (9606.protein.info.v11.0.txt, 9606.protein.links.v11.0.txt and 9606.protein.sequences.v11.0.fasta) and put it under assets/kg/STRING/.

Training and Evaluation

You can run the Bash scripts under dair_biomed/scripts/mtr/:

open_biomed/scripts/drp/
├── train.sh				# train TGDRP on GDSC dataset
└── transfer.sh             # train TGDRP on GDSC dataset and zero-shot test on TCGA dataset

You can also modify the scripts or directly use the following command:

python tasks/mol_task/drp.py \
--device DEVICE \                   # gpu device id
--task TASK \                       # task type, select from [classification, regression]
--mode MODE \                       # training mode, select from [train, test, zero_shot_transfer]
--config_path CONFIG_PATH \         # configuration file, see configs/dti/ for more details
--dataset DATASET \                 # dataset name
--dataset_path DATASET_PATH \       # path to the dataset
--init_checkpoint INIT_CHECKPOINT \ # checkpoint path used for efficient validation
--param_key PARAM_KEY \             # key of the checkpoint dict that contains model parameters
--num_workers NUM_WORKERS \         # number of workers when loading data
--epochs EPOCHS \                   # number of training epochs
--patience PATIENCE \               # number of tolerant epochs for early-stopping
--weight_decay WEIGHT_DECAY \       # weight decay, default is 1e-4
--lr LR \                           # learning rate, default is 0
--batch_size BATCH_SIZE \           # batch size, default is 128
--transfer_dataset_path TRANSFER_DATASET_PATH \ # path to the zero-shot transfer dataset
--seed SEED           							# random seed