zhaohanm/fusiongda

we propose a novel FusionGDA model, which utilises a pre-training phase with a fusion module to enrich the gene and disease semantic representations encoded by pre-trained language models.

Heterogeneous biomedical entity representation learning for gene-disease association prediction

Installation

# Download the latest Anaconda installer
wget https://repo.anaconda.com/archive/Anaconda3-latest-Linux-x86_64.sh

# Install Anaconda
bash Anaconda3-latest-Linux-x86_64.sh -b

# Clean up the installer to save space
rm Anaconda3-latest-Linux-x86_64.sh

# Set path to conda
ENV PATH /root/anaconda3/bin:$PATH

# Updating Anaconda packages
conda update --all

# Install the latest version of PyTorch and related libraries with CUDA support
# Note: Replace 'cudatoolkit=x.x' with the version compatible with your CUDA version
RUN conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch

# Install other Python packages
pip install pytdc
pip install wandb
pip install lightgbm
pip install -U adapter-transformers
pip install pytorch-metric-learning

Executing program

Make sure you are in the directory ~/dpa_pretrain/scripts You adjust the required parameters directly.

Pre-training phase

bash run_pretrain_gda_ml_adapter_infoNCE.sh

Fine-tuning phase

TDC Dataset

bash run_finetune_gda_lightgbm_infoNCE_tdc.sh

DisGeNET Dataset

bash run_finetune_gda_lightgbm_infoNCE.sh

Check your results in the wandb account.

Datasets

We store all required datasets in the Google Drive. Here