RNA-KG analysis with homogeneous methods

This repository contains the code for analyzing the RNA Knowledge Graph (RNA-KG) as detailed in the paper and available on GitHub.

The results from this analysis were submitted in the article "RNA Knowledge Graph analysis through homogeneous embedding methods" to Bioinformatics Advances.

Objectives

The main goal of this project is to explore RNA-KG by applying various homogenous embedding methods and prediction models to carry out the following tasks:

• Node/edge type prediction
• Generic edge prediction
• Specific edge prediction

To perform these tasks, we used two embedding methods:

• Node2Vec Skip-Gram
• LINE

The embeddings generated from these methods were used to train the following prediction models:

• Decision Trees
• Random Forest

This project extensively utilizes the GRAPE library, which can be found in the paper and on GitHub.

Repo structure

• RNA-KG_notebooks/: Contains all the Jupyter notebooks used.
  • Default_RNA-KG/: Notebooks utilizing the full RNA-KG for visualization and prediction tasks.
    • RNA-KG.ipynb: Loads and visualizes the RNA-KG graph using GRAPE functions.
    • RNA-KG_import_fixed.ipynb: Imports the RNA-KG graph from .nt format and creates nodes.csv and edges.csv.
    • RNA-KG_visualization_fixed.ipynb: Generates selective plots of specific embedding types.
    • RNA-KG_visualization_selective.ipynb: Predicts node/edge types using a sample of the RNA-KG graph with 2D embeddings.
    • RNA-KG_node_type_prediction_full_dim_fixed_unbiased.ipynb generates the node type prediction results using the full RNA-KG graph with full-dimensionality embeddings, ensuring unbiased results.
    • edge_prediction.ipynb: Conducts edge prediction using GRAPE, biased by full graph embeddings.
    • edge_prediction_unbiased.ipynb: Performs unbiased edge prediction using holdout-generated embeddings.
    • edge_prediction_bias_comp.ipynb: Compares edge prediction performance between full graph and holdout embeddings.
  • Views/: Notebooks focused on edge prediction using RNA-KG views.
    • ViewX/edge_pred_final_res.ipynb: Implements edge prediction for specific type pairs in ViewX.
    • view_stats_generator.ipynb: Generates statistics for views used in the paper.
• helper_lib/: Custom library for organizing code used across notebooks, primarily leveraging GRAPE.
  • cache.py: Implements caching functions to replace unreliable GRAPE caching.
  • graph.py: Provides utilities for loading and exploring graphs.
  • predict.py: Contains core prediction functions, including a custom edge prediction pipeline.
  • visualize.py: Offers visualization functions for GRAPE-generated embeddings.

Setup

1. Clone the repository and go into the folder.

   git clone https://github.com/AnacletoLAB/RNA-KG_homogeneous_emb_analysis && cd RNA-KG_homogeneous_emb_analysis

2. (Optional but recommended) Create a new virtual environment to install the python packages and activate it.

   python3 -m venv ./.venv && source ./.venv/bin/activate

3. Install dependencies

   pip install -r requirements.txt

4. Start Jupyter

   jupyter-lab

5. Download the RNA-KG.nt file from zenodo: https://zenodo.org/records/10418431

6. Run the notebook RNA-KG_import_fixed.ipynb to generate the nodes.csv and edges.csv files from RNA-KG.nt