vqsynergy's Introduction

# VQSynergy In this study, we introduce VQSynergy, an innovative machine learning framework that leverages the Vector Quantization (VQ) bottlenecking mechanism, combined with gated residuals and a custom attention mechanism, to improve the data efficiency,accuracy and generalizability of drug synergy prediction models.

Requirements

refer to Model/requirements.txt

python==3.8
deepchem==2.7.1
numpy==1.24.4
pandas==2.0.3
rdkit==2023.3.3
scikit_learn==1.3.0
scipy==1.8.1
torch==1.10.1+cu111
torch_geometric==2.0.0

Usage

VQSynergy provides four main scripts for training models on the ALMANAC and O'Neil datasets. Each script is tailored to either quantized or unquantized training:

ALMANAC Dataset:
- almanac_quantized.sh: Trains on the ALMANAC dataset with quantization enabled.
- almanac_unquantized.sh: Trains on the ALMANAC dataset without quantization.
O'Neil Dataset:
- oneil_quantized.sh: Trains on the O'Neil dataset with quantization enabled.
- oneil_unquantized.sh: Trains on the O'Neil dataset without quantization.

To run a script, use the following command:

bash script_name.sh [GPU] [NOISE]

Replace script_name.sh with one of the four scripts mentioned above.

[GPU]: (Optional) Specify the GPU ID to use for training. If not provided, the default GPU (0) will be used.
[NOISE]: (Optional) Specify the noise level to apply. If not provided, the default noise level (0) will be used.

Recommend Projects