This is a companion repository to our preprint https://www.biorxiv.org/content/10.1101/453449v2 (Kobak & Berens 2018, The art of using t-SNE for single-cell transcriptomics). All code is in Python Jupyter notebooks. We use this t-SNE implementation: https://github.com/KlugerLab/FIt-SNE.

See demo.ipynb for a step-by-step guide using a data set from Tasic et al., Nature 2018 (24,000 cells sequenced with Smart-seq2).

The other notebooks generate all figures that we have in the paper:

• toy-example.ipynb
• tasic-et-al.ipynb
• umi-datasets.ipynb
• million-cells.ipynb
• two-million-cells.ipynb
• umap-comparison.ipynb

The last three notebooks require one to run server-10xdata.py and server-cao.py. One needs more than 32 Gb of RAM to process these datasets conveniently, so these Python scripts were run separately on a powerful machine. They pickle all the results (t-SNE embeddings). Unfortunately, these pickles are too large to be shared on github. We might share them later elsewhere.

For any technical questions, please start an Issue.