MultiRBP

MultiRBP is a deep learning model for predicting protein-RNA binding interactions.

MultiRBP

MultiRBP is a deep learning model for predicting protein-RNA binding interactions. This model was developed by Jonathan Karin and Hagai Michel as their bachelor degree final project, under the supervision of Dr.Yaron Orentein.
School of Electrical and Computer Engineering, Ben-Gurion University of the Negev,Beer-Sheva, Israel.

Table of Contents

• About the Project
  • Built With
• Getting Started
  • Bring Data
  • Prepare Data
  • Train and Test
  • In vivo
• Usage
• Contact

About The Project

Deep learning-based method to predict RNA-binding intensity ofhundreds of proteins to a given RNA sequence with better accuracy than extant methods.

Built With

• Python - 3.7 , Numpy and pandas libraries.
• Tensorflow - 2.0.0
• Keras - 2.3.1

Getting Started

Open folder and terminal for the project and write:

git clone https://github.com/OrensteinLab/MultiRBP
cd MultiRBP

Bring Data

Download the normalize RNACompete data in this Link from RNAcompete site . Download the secondary structure data from this Link.
Unzip the files to 'Data' folder.

Prepare Data

python in_vitro/generate_data.py

Train and Test

python in_vitro/train_and_test.py

The results will be saved as res.csv, the model would be saved in h5 format (model_41_9.h5).

In vivo

For achieving the best results in predicting in vivo binding, train the model with input vector size of 75 nucleotides.

python in_vivo/train_75.py

download the in vivo data:
Positive sequences Negative sequences unzip the files to 'eclip' folder.

mkdir in_vivo/eclip/
unzip eCLIP_bed_control_ext_fa.zip -d in_vivo/eclip/
unzip eCLIP_bed_ext_fa.zip -d in_vivo/eclip/

Run the model

python in_vivo/test_in_vivo.py

Contact

Dr. Yaron Orenstein - yaronore [at] bgu.ac.il
Jonathan Karin - karinjo [at ] post.bgu.ac.il