PanRBPNet: A RBP-binding-informed RNA Foundation Model

PanRBPNet - or parnet - is a multi-task extension of our previous RBPNet model for prediction RNA-protein binding at nucleotide resolution.

Warning

This package is under heavy development and while it's API is somewhat stable, it can change at any time and without warning. If you are using this package in your own research, it's highly recommended to either 1) fork the respository or 2) pin the version / commit of the package you are using.

Installation

Using Pip

pip install parnet

or

pip install git+https://github.com/github.com/mhorlacher/parnet.git@SOME_BRANCH

to install from a specific branch (e.g. the latest development version of package version).

Using Conda

git clone https://github.com/github.com/mhorlacher/parnet
cd parnet
make env
conda activate parnet

Using Docker

First, pull the docker image.

docker pull 

Then, run commands as shown below.

docker run parnet --help 

To provide input files and capture output file, mount host files and/or directories using --mount (see the Docker docs).

Dataset

Parnet's training datasets are stored in Huggingface's dataset format (HFDS). The primary training dataset of this study, which is composed of 223 eCLIP tracks from the ENCODE project, can be obtained via:

Download compressed TFDS dataset

wget https://zenodo.org/records/14176118/files/encode.filtered.5.hfds.tar.gz

Unpack TFDS dataset

tar -xJf encode.hfds.tar.xz

Usage

Usage: parnet [OPTIONS] COMMAND [ARGS]...

Options:
  --help  Show this message and exit.

Commands:
  build-dataset
  predict
  train