-
Notifications
You must be signed in to change notification settings - Fork 0
Installation
In order to train a BPNet model, you will need to have certain packages installed on your machine. First, it is highly recommended that you use a GPU for model training and have the necessary NVIDIA drivers and CUDA already installed. You can check if your machine is properly set up to use GPUs by running the command nvidia-smi and making sure it returns information about your system GPU(s) instead of an error. Additionally, there are two ways to ensure you have the necessary packages to train BPNet models:
Download and install the latest version of Docker for your operating system. You can find the appropriate installer for your platform at Docker Installers. Once Docker is installed, run the below shown docker run command followed by the necessary parameters to open an environment with all the necessary packages installed. Then navigate to the chrombpnet directory by running the command cd bpnet to start running the tutorial.
Note: To access your system GPU's from within the docker container, you must have NVIDIA Container Toolkit installed on your host machine.
docker pull vivekramalingam/tf-atlas:gcp-modeling_v2.1.0-rc.1
docker run -it --rm --cpus=10 --memory=200g --gpus device=1 --mount src=/mnt/bpnet-models/,target=/mydata,type=bind vivekramalingam/tf-atlas:gcp-modeling_v2.1.0-rc.1
Instead of installing the BPNet repo by yourself, you can also try the Anvil or Docker options to train/use BPNet models.
The NHGRI's AnVIL (Genomic Data Science Analysis, Visualization, and Informatics Lab-space) platform allows researchers with no/minimal computation skills to run analysis tools with just a few mouse clicks. Using our BPNet workflow on the AnVIL platform you can train state-of-art deep learning BPNet models for ChIP-seq data without any programming experience. We highly recommand this option for biologists. It is also highly scalable as GPU provisioning on the cloud etc., is automatically handled by AnVIL. We trained our Atlas scale models on the entire ENCODE compendium using the AnVIL platform and those models are also available for use on the AnVIL platform.
Create a clean conda environment with python ==3.7
conda create --name bpnet python=3.7
conda activate bpnet
Install non-Python requirements via conda
conda install -c conda-forge -c bioconda samtools=1.19.2 bamtools=2.5.2 ucsc-bedgraphtobigwig=445
Install from source
pip install git+https://github.com/kundajelab/bpnet-refactor.git