Installations

For GPU support (in Windows Native):

conda install -c conda-forge cudatoolkit=11.2 cudnn=8.1.0
# Anything above 2.10 is not supported on the GPU on Windows Native
pip install "tensorflow<2.11"
# Verify the installation
python -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"

Others pip installations (using the requirements.txt file):

pip install -r requirements.txt

The environment can also be setup using the environment.yml file:

conda env create -f environment.yml
conda activate gramseq

Data Preparation

Preparing the RNA-Seq data from L1000 project:

python src/prepare_l1000.py

Training

To integrate RNA-seq data with the BindingDB dataset for model training, the following code can be used:

python src/train.py --dataset bindingdb\
                    --protenc CNN\
                    --rnaseq\
                    --epochs 500\
                    --folds 5 > ./result_logs/output_bindingdb_rnaseq_cnn.log 2> error_bindingdb_rnaseq_cnn.log

Similarly, the model can be trained on the Davis and KIBA datasets by replacing the bindingdb with davis or kiba for the --dataset argument, respectively. To use RNN as the protein encoder, CNN can be replaced with RNN and passed into the --protenc argument.