David Hägele, Yuxuan Tang, Daniel Weiskopf
This repository contains the code for the visual analytics system prototype that was proposed as a solution for the 2024 Bio+MedVis Challenge. The software is a Bokeh server application written in Python that provides a web front-end.
| Challenge Description | Visualization to Redesign |
|---|---|
Spatial transcriptomics technology can detect cell types at different locations on cellular tissue. Due to limited resolution, a mixture of cell types is detected per spot. However, the proportions of the individual cell types for an individual spot can be determined, resulting in a composition such as [80% type1, 10% type2, 4% type3, ...]. This data can be visualized using pie chart glyphs superimposed onto a histological image, so analysts can relate the cell type proportions to locations on the tissue and observe areas of similar patterns. Such a visualization has certain limitations and therefore the challenge asks for a redesign. |
 |
We propose a visual analytics system to explore the cell type compositions and relate them to the histological image of the tissue.
There are 3 views that support brushing and linking, i.e., selections made in one view are reflected in the other views.
- Histological image view - shows tissue and locations of spots (toggleable) which can be highlighted when a selection of spots is issued.
- Stacked bar chart of cell type mixtures - shows the cell type proportions of selected spots.
- Dimensionality reduction (similarity) of cell type mixtures - using PCA of the proportions in Aitchison gemotry shows similar mixtures being grouped into blobs.
- additional k-means clustering for color coding.
To set up the project you need an up to date Python 3 installation. Then you can use the bash scripts to set up and run the server application.
# set up a virtual python environment
./setup_venv.sh
# install the dependencies
./setup_dependencies.sh
# start the server
./start_server.sh