Compound models and Pearson residuals for single-cell RNA-seq data without UMIs

This repository holds the code needed to reproduce the analyses and figures presented in Lause et al. (2024). The code for the earlier version of the preprint from August 2023 can be found under the v1 release.

Code

Some of the notebooks depend on each other.

All plots based on the Tasic 2018 dataset require 01_prepare_tasic to run first. Then,

• to reproduce Figures for the homogeneous within-cluster data (Figure 2, S1, S2, S4), run notebook 02
• to reproduce Figures for the full Tasic data (Figure 3, S5, S6, S8), run notebook 03 to compute t-SNEs etc. and notebook 04 to make the figures
• to reproduce Figure S7 for the Census/qUMI comparison, run Census and qUMI with 05_compute_tasic_qumis_census.R (using our separate R environment, see below for setup instructions), and then run notebooks 06-08 to load the data, process and plot
• to reproduce Figure S9 for the Tasic-like simulated data, use notebooks 09-11 to simulate, process and plot
• to reproduce Figure 6, run notebook 16 to prepare simulated data and notebook 17 to plot

All plots based on the reads-per-UMI tables from the Ziegenhain/Hagemann-Jensen datasets requires 12_prepare_ziegenhain to run first. Then,

• to reproduce the main Figures 4 and 5, run notebook 13
• to reproduce Figure S3 on Pseudogenes, run notebook 14
• to reproduce Figure S10 and S11 on per-cell amplification parameter estimates, run notebook 15

Datasets

• Download the reads-per-UMI tables from zenodo and save them to .data/reads_per_umi_tables/. R code to obtain the same tables from the public raw data is available in data/reads_per_umi_tables/prepare_data.R.
• Download the Tasic raw count data from brain-map.org via the Gene-level (exonic and intronic) read count values for all samples (zip) link. From these *.zip files, extract the mouse_ALM_2018-06-14_exon-matrix.csv and mouse_VISp_2018-06-14_exon-matrix.csv to .data/tasic/.
• All required metadata tables are contained in this repository for convenience.

Compute environment

We ran all notebooks in Python 3.8.10 on an Ubuntu machine with 40 CPUs and 440 GB RAM. The following package versions were used:

• scanpy 1.9.0
• anndata 0.8.0
• sklearn 1.0.2
• numpy 1.21.5
• matplotlib 3.5.1
• openTSNE 0.6.0
• pandas 1.4.1
• seaborn 0.11.2
• mygene 3.2.2.
• scipy 1.8.0

Census and qUMI where run in a separate R conda environment specified in r41_env.yml. To install it, create the environment from that file with

conda env create -f r41_env.yml

Then, to install qUMI, activate the environment with conda activate r41_env_full, start R and run

remotes::install_github("willtownes/quminorm")