mylzwq / pporank Goto Github PK

Deep reinforcement learning for personalized drug recommendation

Python>=3.7.10; torch(1.18.1); yaml(0.1.7); numpy(>=1.18.1); pandas(1.0.0); scipy(1.4.1); seaborn(0.8.1); tensorboard (1.15.5); matplotlib(3.1.1); joblib(0.14.1); json5(0.8.5);jupyter(1.0.0); tqdm (4.42.0) For drugs processing: deepchem (2.5.0); h5py(2.10.0); hdf5(1.10.4)

The implementation of PPORank is in "main.py": (cpu version) : The following is an exmaple for GDSC dataset when training with 16 actors, projection dimension of 100 without normalize y more details can be found in arguments.py

python main.py   --num_processes 16  --nlayers_deep 2 --Data GDSC_ALL --analysis FULL  --algo ppo --f 100  --normalize_y 

model logs dir : ./logs

model saved dir: ./Saved

model prediction saved dir: ./results

The clean data could be found in

Data Sharing

Download and preprocess the GDSC dataset:

python ./preprocess/load_dataset.py load_GDSC.txt

Split the data for training and testing, create folds for cross-validation and Pretrain the MF layers' weight

python prepare.py config.yaml

Runing the experiments on ppo (with config file "./configs/configS_base.yaml"):

python results.py > results_ppo.txt

PPO experiment with TCGA cohort

 python load_TCGA.py
 
 python results_TCGA.py ./TCGA/TCGA_BRCA.npz