Native system or Virtual Machine with at least around 30 GB free.
The systems is required to provide the following commands:
mkdircpcurl
On NixOS, all dependencies are already included in the file shell.nix, no manual installation of dependencies is required.
Nvidia CUDA allows for faster execution of the Notebook at hand by utilizing the GPU instead of the CPU. From a select few tests, it seems to increase the iterations per second about tenfold when using a AMD Ryzen 9 5900HX CPU and a NVIDIA GeForce RTX 3070 Laptop (Mobile) GPU.
To run the Jupyter Notebook with CUDA support enabled, some manual adjusting needs to be done.
The file shell.nix has to be edited, specifically # in the line containing # cudaPackages.cudatoolkit has to be removed.
If you want the exact python pip package versions, remove all # from the file requirements.txt.
Furthermore, it seems to be necessary to include the following lines in /etc/nixos/configuration.nix.
If you have any advice on how to solve this via a shell.nix-File, please let me know! The documentation on how to use CUDA on NixOS seems to be outdated or at least didn't work out of the box for me.
systemd.services.nvidia-control-devices = {
wantedBy = [ "multi-user.target" ];
serviceConfig.ExecStart = "${pkgs.linuxPackages.nvidia_x11.bin}/bin/nvidia-smi";
};For Debian / Ubuntu, the following two commands should suffice:
sudo apt install gcc curl python3.11 python3-pip
python3.11 -m pip install --user virtualenvOn other systems, dependencies may need to be installed manually if the previous sections are not applicable. Please refer to the previous install commands for Debian / Ubuntu and the shell.nix file for NixOS to derive the required setup for your system.
On the first run, starting up might take a while, because the python dependencies required to run the Notebook will be installed.
To start JupyterLab on NixOS, the following command needs to be executed:
nix-shellTo start JupyterLab on Debian / Ubuntu and potentially other systems, the following command needs to be executed:
bash ./run.shOnce JupyterLab has opened, open the notebook ZDNA-prediction.local.ipynb in JupyterLab.
The big blue link at the very top with the title "Jump to Run Section" can be used to jump to the "Run"-Section, if you feel like it takes too long to scroll down.
The "Run"-Section contains further information on how to use the notebook.
This repository contains code and data for the article "Z-Flipon Variants reveal the many roles of Z-DNA and Z-RNA in health and disease"
The full genome predictions for human and mouse genomes can be downloaded here
To predict Z-DNA flipons on new data please use this colab notebook
The finetuned DNABERT weights can be downloaded from google drive:
1_HG_chipseq.ipynb - Generate data splits for HG data with Chipseq labels. Train the models. Generate full genome predictions.
1_HG_kousine.ipynb - Generate data splits for HG data with Kouzine labels. Train the models. Generate full genome predictions.
1_MM_curax.ipynb - Generate data splits for MM data with Curax labels. Train the models.
1_MM_kousine.ipynb - Generate data splits for MM data with Kouzine labels. Train the models.
2_Generate_stats_hg_chipseq.ipynb - Calculate most frequently attended k-mers for HG data with Chipseq labels.
2_Generate_stats_hg_kouzine.ipynb - Calculate most frequently attended k-mers for HG data with Kouzine labels.
2_Generate_stats_mm_curax.ipynb - Generate full genome predictions for MM data with Curax labels. Calculate most frequently attended k-mers.
2_Generate_stats_mm_kouzine.ipynb - Generate full genome predictions for MM data with Kouzine labels. Calculate most frequently attended k-mers.
README.md - This file
ZDNA-prediction.ipynb - Standalone notebook for prediction of Z-DNA. Intended to be run in colab enviroment via: https://colab.research.google.com/github/mitiau/Z-DNABERT/blob/main/ZDNA-prediction.ipynb