Predicting Interactions of m6A-Regulated RNA-Binding Proteins

Author: Sofia Martello
Master's Thesis Project | TU Munich Campus Straubing | June 30, 2023

Overview

This repository contains the code and data for my master's thesis project on predicting interactions of m6A-regulated RNA-binding proteins, with the aim of understanding the impact of the methylation m6A on the binding preferences of RNA-binding proteins.

Input Data

1. RNA-seq for HEK293 cells (Sun et al. 2018)

   • Dataset: (Sun et al. 2018) --> GSE122425_all.counts.293_vs_293NK.edgeR_all

2. RNA-seq for HepG2 (Wold, ENCODE)

   • Dataset: (Wold, ENCODE)

3. ENCODE eCLIP for protein-RNA interactions (Wold, ENCODE)

   • Dataset: (Wold, ENCODE)

4. miCLIP for m6A modifications on HEK293 cells (Linder et al. 2015)

   • Dataset: (Linder et al. 2015), GSE63753_hek293.abcam.CIMS.m6A.9536.bed (processed)

Dataset Preparation

For dataset preparation, consider three settings:

1. Baseline:

   • Does not contain m6A information.

2. Setting A:

   • Contains only sequences with at least one m6A site.

3. Setting B:

   • Ratio 1:1 between sequences containing and not containing m6A sites.

Setup

1. Duplicate the environment:

   • Use denbi-conda_env.yml and denbi-jupyterlab_env.yml.

2. Create a package 'src':

   • Utilize encoding.py, filtering.py, and model.py.
   • Install with pip install -e .

3. Organize folders as suggested:

   |-- data
   |-- docs
   |-- results
   |-- scripts
   |-- src
   |-- tests
   

Workflow

Run the Jupyter Notebooks in the following order:

1. Preprocessing + map_gene_ids.R
2. Encoding
3. Plots
4. Model
5. Baseline_vs_Setting_A
6. Baseline_vs_Setting_B
7. Baseline_vs_Setting_A_aug
8. Baseline_vs_Setting_B_aug

For the Baseline_vs_Setting notebooks, input either negative-1 (unbound sequences) or negative-2 (sequences bound to other RBPs) as class 0.