This repository contains a collection of notebooks for generating VR content using the Barabasi-Albert network and expression profiles data. Each notebook focuses on different aspects of the VR visualization process.
To quickly begin running a notebook in google colab, view the file in github and edit the github.com in the URL to githubtocolab.com, e.g. for https://github.com/menchelab/ISMB_ECCB_2023_Network_VR_Visualization/blob/main/1_Minimal_Example_Notebook.ipynb
edit the URL to
https://githubtocolab.com/menchelab/ISMB_ECCB_2023_Network_VR_Visualization/blob/main/1_Minimal_Example_Notebook.ipynb
and you will be redirected to an active colab notebook containg the same code.
Build the dolphin friendship network as a simple example. The results are exported for DataDiVR upload.
The information needed for creating the PPI is downloaded form the STRING database THe PPI is created as a networkx object, with appropriate filtering for score. A gene list is provided and the subnetwork of genes in the gene list and the neighbors are selected. Leaves in the subnetwork are filtered out. The results are exported for DataDiVR upload.
Use the umap package to build a weighted similarity network on a dataset Try out different layouts of that network. The results are exported for DataDiVR upload.
Make a correlation network across the features in a dataset. Filter that network with betweenness centrality to make it sparser. The results are exported for DataDiVR upload.
Especially for those already familiar with network science.
The basic_barabasi_albert.ipynb notebook demonstrates the basic upload functionalities. It generates a random BA Graph using the networkx module and converts the node and edge structure into a JSON file required for the VR platform. The notebook introduces two core functions, make_json() and plotly_preview(), which will be used in other notebooks as well. It showcases the minimal usage of make_json() to quickly generate the first VR content.
The celltype_network.ipynb notebook constructs a network based on expression profiles from The Human Protein Atlas. In this network, cell types are represented as nodes, and links are shared genes between cell types. Edge weights are calculated using the Jaccard similarity score, measuring the number of shared genes.
This notebook showcases three main steps:
-
Edge Filtering: Edges are ranked and filtered based on edge-betweenness-centrality, with a chosen cutoff criterion. The aim is to retain the minimal set of edges before the network breaks down.
-
Layouts: Different layouts are introduced for network visualization. Initially, random and spring layouts provide basic placement of the nodes. More advanced layouts such as 'global' and 'importance' from Cartographs can be employed. Additionally, the concept of utilizing the z-axis is introduced to display an external attribute, where the number of different tissues a cell type can be found in is used as the value for the z-axis.
-
Coloring and Annotations: This section demonstrates coloring nodes and edges and transferring node annotations into the JSON format. This allows the annotations to be visible and accessible within the VR environment.
The edge_filtering.ipynb notebook demonstrates three different edge filtering methods for the expression profiles dataset:
- Cutoff by Edge Weights Directly
- Edge Betweenness Centrality
- Disparity Filter (Backbone)
Each method is explored in detail within the notebook.
Feel free to explore the notebooks and use them to generate VR content for your projects!
The VR backend github repository
A handbook that covers the main features of the interactive panels in VR