ISMB BioVis '22 Talk on Data Transformations for Effective Visualization of Single-Cell Embeddings

This repository contains the code to reproduce plots presented in our BioVis talk+poster at ISMB '22. The talk itself is available on YouTube.

For a much more elaborate R implementation that includes our FAUST clustering method, please take a look at https://github.com/RGLab/FAUST.

For details about our clustering and visualization methods, please take a look at the related publication:

Greene et al., 2021, New interpretable machine-learning method for single-cell data reveals correlates of clinical response to cancer immunotherapy. Pattern.

Requirements

• Conda

Install

git clone [email protected]:flekschas-ozette/ismb-biovis-2022.git
cd ismb-biovis-2022
conda env create -f environment.yml
conda activate ozette-ismb-biovis-2022

Example Data

Download the example data from https://figshare.com/articles/dataset/ISMB_BioVis_2022_Data/20301639 and place the files under data/mair-2022. The data is from Mair et al., 2022, Extricating human tumour immune alterations from tissue inflammation, Nature.

Get Started

1. Start JupyterLab:

   jupyter-lab
   

2. Open one of the following notebooks:

• Explanation of our transformation embedding approach: annotation-embedding.ipynb

• Comparison of our transformation approach using different non-linear embedding methods: compare-annotation-embedding.ipynb

• Joint embedding of two samples showing how our transformation approach helps to reduce batch effects: joint-annotation-embedding.ipynb