JosieHong / DrugCell_Exp

A visible neural network model for drug response prediction

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

I529 - Experiments on DrugCell

This is the final project of I-529. We did the following experiments based on DrugCell:

  • Using unhashed fingerprints of drugs; Comparing the Baseline1 and Exp1-1, using hashed or unhashed fingerprint will not effect significantly.
  • Using Graph Convolution Networks (GCN/GAT) to embed drugs; To traina GCN/GAT in batchwise: We build up the model parallelly shown in following figure, referring this issue. Comparing the Baseline2 and Exp2-2, our model performs better in MSE, but not good in PC. More results are coming soon.
  • More metrics: mean mean squared error (MSE), pearson correalation (PC). Different features of these two metrics are here.

The results are:

Model Note PC MSE Scripts
Baseline0 Pretrained model* 0.822805 0.014052
Baseline1 Train on drugcell_all.txt 0.828568 0.013232 &
Exp1-1 Train on drugcell_all.txt & using unhashed FP 0.813499 0.013995 &
Exp1-2 Train on drugcell_all_cut.txt & GCN &
Exp1-3 Train on drugcell_all_cut.txt & GAT &
Baseline2 Train on drugcell_train.txt 0.315630 0.282851 &
Exp2-2 Train on drugcell_train.txt & GCN -0.036170 0.040641 &
Exp2-3 Train on drugcell_train.txt & GAT -0.023885 0.040629 &

The pretrained model can be downloaded here.


The whole dataset can be download here.

$ cat drugcell_all.txt | wc -l
$ cat drugcell_all_cut.txt | wc -l
$ cat drugcell_train.txt | wc -l
$ cat drugcell_test.txt | wc -l


Please set up the environment as described in ./ Then install rdkit for loading drug graph and tqdm for showing the process bar by following command:

conda activate pytorch3drugcell
conda install -c rdkit rdkit
conda install -c conda-forge tqdm

All the experiments' scripts are in ./sample/. Please run them as following example:

conda activate pytorch3drugcell
cd sample

# test the pretrained model

# train and test our own model

# More experiments' scripts can be found in the table. 


  title={Predicting drug response and synergy using a deep learning model of human cancer cells},
  author={Kuenzi, Brent M and Park, Jisoo and Fong, Samson H and Sanchez, Kyle S and Lee, John and Kreisberg, Jason F and Ma, Jianzhu and Ideker, Trey},
  journal={Cancer cell},


A visible neural network model for drug response prediction

License:MIT License


Language:Python 100.0%