Penguinn

PENGUINN (Precise Exploration of Nuclear G-quadruplexes Using Interpretable Neural Networks) is a machine learning method based on Convolutional Neural Networks, that learns the characteristics of G4 sequences and predicts probability of forming G4s for given sequences.

Have a look at our paper PENGUINN: Precise Exploration of Nuclear G-Quadruplexes Using Interpretable Neural Networks for more information about our work.

Installation

Using Git:

git clone https://gitlab.com/RBP_Bioinformatics/penguinn.git

or

git clone git@gitlab.com:RBP_Bioinformatics/penguinn.git

Prerequisities

Penguinn is implemented in python using Keras and Tensorflow backend.

Required:

• python, recommended version 3.7
  • Keras 2.3.1
  • tensorflow 2.3.0
  • Biopython
  • numpy

Installing

Firstly, create a virtual environment.

python -m venv venv

Then, activate it and install the necessary libraries.

source venv/bin/activate
pip install -r requirements.txt

Running prediction

Follow the instructions:

cd path/to/Penguinn/directory
#add rights to execute
chmod +x penguinn.py
#if you've followed the installation steps above and aren't actively sourcing from the previously created virtualenv
source venv/bin/activate
#run the prediction
./penguinn.py --input <input_fasta_file> --output <output_file> --model <path_to_model.h5>

Default model is set to model trained on human dataset with positive:negative samples ratio 1:1, as described in our paper.

Web application

https://ml-bioinfo-ceitec.github.io/penguinn/

Contact information

CEITEC MU, RBP Bioinformatics - Panagiotis Alexiou, https://www.ceitec.eu/rbp-bioinformatics-panagiotis-alexiou/rg281