Drug Response Prediction with NTK

File Description

1. data.npy: Processed Drug Response data from PDX database with 1634 examples, taken from [1]
2. process.py: Divides data.npy into train/test split in a ratio of roughly 2:1, taken from [1]
3. NTK.py: Neural Tangent Kernel, taken from [2]
4. script.py: Calculates Mean Squared Error using SVM with both linear kernel and NTK

Execution

1. Execute ths script: python3 script.py

Results

1. MSE with linear kernel SVM reported in [1]: 0.824 +- 0.034
2. MSE with linear kernel SVM I got: 0.807 +- 0.134
3. MSE with linear kernel regression I got: 0.817 +- 0.120
4. MSE with NTK kernel SVM I got: 0.755 +- 0.128
5. MSE with NTK kernel regression I got: 0.788 +- 0.109
6. MSE with Baysian Regression I got: 0.804 +- 0.122
7. MSE with Random Forest Regression I got: 0.813 +- 0.141

References

1. DrugOrchestra
2. NTK

Drug Task Prediction with GNN

Refer to GNN folder. Datasets explored: Drugbank and Repurposing Hub Earlier AUROC score: Drugbank - 0.752 and Repurposing Hub - 0.703 New AUROC score: Combined - 0.825