Donor inference

• Vireo
• Souporcell
• scsplit
• Genetic demultiplexing of pooled single-cell RNA-sequencing samples in cancer facilitates effective experimental design + code: https://github.com/lmweber/snp-dmx-cancer
• Can mention demuxlet, but it requires additional information

Dataset: Maybe the Souporcell test dataset mentioned in their README? Would restrict our notebook to Souporcell because Souporcell has a different input than e.g. Vireo.

Advanced: Interactive visualization

• host our preprocessed example dataset on the cellxgene data portal
• add chan zuckerberg lab to the acknowledgements in paper
• embedd link to cellxgene data portal in the tutorial

TODO

GRNs
Explore concepts in ATAC-RNA matching (and potentially link to multi-model data). Define current best practices in gene regulation and regulatory patterns.

Potential starting points:
SCENIC
cellOracle

Datasets:
GRN specific dataset, check for dimensions

PAGA and much much more

https://jupyterbook.org/advanced/pdf.html

Solely for fun

Provide an updated version of current best-practices in Quality Control, data normalization, …

(Technical) requirements engineering:

• Visual design and functionality of https://inria.github.io/scikit-learn-mooc/ <- https://github.com/INRIA/scikit-learn-mooc
• Custom domain
• Trivial command to build the complete book while rerunning all jupyter notebooks
• Command to selectively build a single chapter
• Caching already built jupyter notebooks that were not edited. Don't rerun them when building the book!
• Interactive sessions for specific chapters with runnable instances with sufficent power with Binder. #1
• Environment(s) to build the complete book. Identify whether a single environment for the complete book works or we need to somehow build one environment for one chapter.
• Jupyter lab + a couple of extensions (black, [...]) must be part of the environments to format them cleanly
• A way of comparing Jupyter Notebook diffs on PRs -> likely just reviewnb
• A way to build a single PDF from the book.
• Well written documentation explaining how to build (parts of) the book, exploring the content on your own, [...]
• Build the book via CI?

cellrank
Palantir
trajectoryNet
Schiebinger et al optimal transport

Perturbations

• Tools: https://www.scrna-tools.org/tools?sort=name&cats=Perturbations
• Quantifying the effect of experimental perturbations at single-cell resolution
• scGen
• scMAGeCK links genotypes with multiple phenotypes in single-cell CRISPR screens <- really unsure about this one
• Conditional out-of-distribution generation for unpaired data using transfer VAE
• CPA
• Comprehensive benchmarking of single cell RNA sequencing technologies for characterizing cellular perturbation <- this is from an experimental point of view!
• Machine learning for perturbational single-cell omics

1. Discern many perturbations (>=5 maybe)
2. Differentiate between biological and confounded effects
3. Detect perturbed and non-perturbed cells -> how to remove cells that escaped the perturbations
4. Visualizing similarities and differences across different perturbations
5. Applications of tools like ScGen to predict the effect of unseen perturbations

Feature selection for marker genes
Provide updated best practices for feature selection for marker genes and introduce feature selection methods for multi-sample scRNA-seq data.

Newer methods for DE gene detection combat p-value inflation by learning latent space, clustering, and learning marker genes in one go.

Potential starting points:
(maybe) Jan Hasenauer wilcoxon test method (no separate publication for this method, check/discuss if worth adding)
Plus: marker gene methods which aren’t based on differential expression (check for benchmarks)

https://www.biorxiv.org/content/10.1101/2022.05.09.490241v1

A few recent benchmarks on transcription factor and pathway analysis came out. It might be time to update what we mentioned in v1.

Potential starting points:

• Robustness and applicability of transcription factor and pathway analysis tools on single-cell RNA-seq data
• Benchmarking algorithms for pathway activity transformation of single-cell RNA-seq data

We should write a preamble chapter.

Possible content:

1. What this book is.
2. Who the book is aimed at and who is not the target audience
3. People and experts who contributed to this book
4. Links to both papers
5. How to use this book -> the "traffic light" system and more

We need to ensure that it does not overlap with the introduction chapter.

• References to other existing single-cell books and tutorials
• References to best-practice papers and books (aka our earlier work)
• [...]

https://github.com/QuantEcon

Pointed out by Isaac. May help to make this efficient.

Differential Gene Expression

Potential topics for updated best practices are DE test on conditions and mixture/covariate models

Potential starting points:

• Muscat
• MAST
• limma
• Comparison of methods to detect differentially expressed genes between single-cell populations
• Comparative analysis of differential gene expression analysis tools for single-cell RNA sequencing data
• aggregateBioVar
• Confronting false discoveries in single-cell differential expression
• A practical solution to pseudoreplication bias in single-cell studies
• Reproducibility of Methods to Detect Differentially Expressed Genes from Single-Cell RNA Sequencing
• A Comprehensive Survey of Statistical Approaches for Differential Expression Analysis in Single-Cell RNA Sequencing Studies
• The paper which contests many of the results: https://www.biorxiv.org/content/10.1101/2022.02.16.480662v1
• A rant about why pseudobulk methods are good and Zimmermann et al. is wrong: https://www.biorxiv.org/content/10.1101/2022.02.16.480517v1

Experts

• Mark Robinson
• John Marioni

Host own Binder service https://binderhub.readthedocs.io/en/latest/ and then point Jupyter Book to that BinderHub's URL

Pathway analysis

TODO

Experts

Julio Saez-Rodriguez

Dealing with conditions

Dataset

https://singlecell.broadinstitute.org/single_cell/study/SCP548/an-immune-cell-signature-of-bacterial-sepsis-patient-pbmcs?cluster=All%20Cells&spatialGroups=--&annotation=Cohort--group--study&subsample=100000#study-visualize

https://www.nature.com/articles/s41591-020-0752-4

• UTI = urinary-tract infection
• Leuk-UTI = UTI with leukocytes, but no organ disfunction
• Int-URO = UTI with mild or transient organ disfunction
• URO = UTI with clear or persistent organ disfunction
• Bac-SEP = bacteremic individuals with sepsis in hospitals

Figure

• From top to bottom with arrows
• Start with dataset with several perturbations/conditions
• Split into the two major schools of thought -> DE (volcano plot) and perturbation analysis (something like this?
  
  )
• Split the two major schools of thought again up into

1. DE: pseudobulk vs cell wise
2. Perturbation analysis: ScGen aka DNN(VAE) approaches vs something else (Mixscape style)?

Experts

Yuge

Discusses how transition from the R to the Python ecosystem and vice versa.
Introduces packages like anndata2ri, zellkonverter, Seurat functions, AnndataR

• Living book -> will change
• Open problems will benchmark stuff and we will update
  -(End-to-end pipelines)
• Map onto reference which include low quality stuff as well
• Filter low quality stuff out on the fly
• Embed based on references
• Technical method outlook
• Longitudinal data modelling

Clustering
BP1.0 included louvain clustering as best practice. New best practices suggest to use leiden as a clustering method for scRNA-seq data. We want to keep this short.

Potential starting points:
Leiden
Leiden issue on Scanpy
Benchmark (2019-12)
Benchmarking clustering algorithms on estimating the number of cell types from single-cell RNA-sequencing data

adding this issue as an reminder for the interactive visualization section, which requires the following steps

• finalization of preprocessing section #10
• host our preprocessed example dataset on the cellxgene data portal
• add chan zuckerberg lab to the acknowledgements in paper
• embedd link to cellxgene data portal in the tutorial

TODO

1.0 Prior art (bioconductor book, etc.)
1.1 Experimental data collection
1.2 Raw data processing
1.3 Data infrastructure (maybe as separate chapter)
1.4 Interoperability (anndata2ri, zellkonverter, Seurat functions, AnndataR)

Details to follow.

As discussed with @le-ander

• Based on @le-ander existing container, but our own container for the book. Eventually copy the counter source into our repository
• Python 3.8
• Scanpy 1.8.2
• R 4.1
• Maybe also pin Pandas to 1.2.5. Not sure anymore, feel free to ignore for now.
• The rest of the packages that you already have in the existing container
• jupyter-book
• https://jupyterlab-code-formatter.readthedocs.io/en/latest/ <- configured with black (if we can pre configure this)

@Zethson will provide a book to start from scratch with in 2 weeks.

Dimensionality reduction
TODO

Variant calling & SNP analysis

• Maybe a bit tangential, but interesting nevertheless: Multiplexed single-cell transcriptional response profiling to define cancer vulnerabilities and therapeutic mechanism of action

• Good eQTL intro that has no relevance here, but helps @Zethson: https://elifesciences.org/articles/52155

• Optimizing expression quantitative trait locus mapping workflows for single-cell studies

• Red panda: a novel method for detecting variants in single-cell RNA sequencing <- a method that focuses on VC using scRNA seq data. Compares to the "usual" WGS and WES VC

Write introduction TODO

Ideally we could use one environment per notebook, but this may not be feasible. I shall think of approaches to do this and evaluate them briefly. Would evolve partially executing the scripts.

We should start out by developing a new R >= 4 based container (look at @le-ander work) and trying to run and integrate @LuckyMD old notebook first. Ideally I could also test the notebook to script and vice versa conversion here. While doing so I shall ensure that as many dependencies as possible are defined in the Mamba environment file.

• How to best annotate cells.
• How to get good markers
• Tips on color scale when plotting UMAPs (red on white instead of blue on green)
• Automated cell type annotation
• Dendrograms to QC cell type annotations
• [...]

Includes:

Feature selection and Manual annotation
Automated cell type annotation
Reference mapping

Cell type composition analysis is difficult to write best-practices for because there are no reviews or benchmarks yet for it. Therefore, we should focus on which tools are available, 1-2 comments on strengths and weaknesses and what should be kept in mind when conducting cell type composition analysis.
Potential starting points:

• scCODA
• Bisque
• scDC or scdney
• speckle
• Milo
• sccomp

Experts: Maren

0. Preamble

1. Intro
   1.0 Prior art (bioconductor book, etc.) (LH)
   1.1 Experimental data collection (LH)
   1.2 Raw data processing (AS)
   1.3 AnnData & Scanpy intro for newbies (LH)
   1.4 Data infrastructure (maybe as separate chapter) (AS)
   1.5 Interoperability (anndata2ri, zellkonverter, Seurat functions, AnndataR, https://github.com/cellgeni/sceasy) (AS)

X. Datasets

2. Preprocessing and visualization
   2.1 QC
   2.2 Normalization
   2.3 Feature selection
   2.4 Dimensionality reduction
   2.5 Advanced: Ambient RNA
   2.6 Advanced: single-cell vs single nuclei
   2.7 Advanced: Interactive visualization

3. Dealing with batch effects
   3.1 Batch-aware feature selection
   3.2 Data integration
   3.2 Evaluation

4. Interpreting cellular structure
   4.1 Clustering
   4.2 Annotation
   4.2.1 Feature selection and Manual annotation
   4.2.2 Automated cell type annotation
   4.2.3 Reference mapping
   4.3 Trajectory inference
   4.4 RNA velocity
   4.5 Fate mapping

5. Dealing with conditions
   5.1 DE testing
   5.2 Compositional analysis
   5.3 Perturbation modeling
   5.4 Pathway analysis

6. Modeling mechanisms
   6.1 GRNs
   6.2 Cell-cell communication
   6.3 Variant calling (?)

7. Deconvolution
   7.1 Bulk deconvolution
   7.2 Donor deconvolution

8. Spatial omics

9. Multi-omics

10. Outlook

Optional topic.

Refer reviews of spatial omnics methods and give an overview of analysis pipeline. Potentially, discuss benchmarking of spot deconvolution approaches in Visium data and segmentation algorithms. Depending on scope, we might want to discuss the different spatial transcriptomics technologies.

Potential starting points:
Tangram (for reference-based deconvolution)
Cellpose (segmentation)
StarDist (segmentation)

Theislab experts:
Giovanni, Hannah, Leon

Bulk deconvolution is still popular to extract as much knowledge as possible from bulk data. Many or even too many methods exist, but there have been some benchmarks and evaluations discussing them. Let us give some pointers.
Potential starting points:

• SCaden
• DWLS
• CPM
• AutoGeneS
• Benchmarking of cell type deconvolution pipelines for transcriptomics data
• A benchmark for RNA-seq deconvolution analysis under dynamic testing environments

Experts: Amit & Hana

• Introduction - scope 0.5-1 page

• look into reviews

• broad overview into the field

• mention nature methods "method of the year 2020"

• define axis: targeted/high-resolution vs unbiased/cell-aggregates

• Technologies (depends heavily on what we show in the tutorial) - scope 0.5 page per method

• ST, visium

• MERFISH

• IMC (Imaging Mass Cytometry)

• tools (plus tutorial) - check current reviews - 1 page overall

• squidpy

• spectreMAP

• giotto

• ??

• spot deconvolution (show two methods and tutorials for this) - 0.5 page

• RCTD

• SpotLight

• Stereoscope

• DestVI

• cell2location

• Identification of Cells from mRNA spot - 0.5 page

• Baysor

• SSAM

• Sparcle

• ClusterMap

• other (check where to put those, if even include)

• tangram

• cell type annotation

• imputation

• downstream tools

• ncem

Should explain how to get from base calling signals up to count tables.

This issue tracks all things related to cell hashing. This will be an advanced topic.

Content

Dataset

Experts

Content of the dealing with batch effects chapter.

Please edit this issue if you want to @LuckyMD @lazappi

Would also be good if you could add a few details to your chapter here: #9

Data integration & automated cell type annotation
Data integration methods have now been tested extensively and there is (as always) no clear winner. However, some methods are better at certain tasks or problem sizes and we should emphasize that and discuss that.
Dataset: Haber et al 2018 + new mouse dataset from Malte

• Benchmarking atlas-level data integration in single-cell genomics
• Building the Mega Single Cell Transcriptome Ocular Meta-Atlas
• Computational principles and challenges in single-cell data integration
• Computational methods for the integrative analysis of single-cell data
• A benchmark of batch-effect correction methods for single-cell RNA sequencing data
• Flexible comparison of batch correction methods for single-cell RNA-seq using BatchBench

Cell type annotation can be conducted in an either automated or expert driven fashion. Automated cell type annotation is improving over time as the number of annotated datasets and marker genes grows. Especially, the mapping against reference atlas datasets is becoming more and more interesting.

• A comparison of automatic cell identification methods for single-cell RNA sequencing data
• Automated methods for cell type annotation on scRNA-seq data <- overview of available methods
• Tutorial: guidelines for annotating single-cell transcriptomic maps using automated and manual methods
• Evaluation of Cell Type Annotation R Packages on Single-cell RNA-seq Data <- R only eww
• Mapping single-cell data to reference atlases by transfer learning <- not sure about this one here; LH will revisit
• Lisa's atlas paper?

Answers questions like

1. How to get data
2. Where to get data from
3. Sfaira
4. ....

QC
Current best-practices suggest cell QC based on three QC covariates: count depth, number of genes per barcode, and fraction of counts from mitochondrial genes per barcode. We want to provide an update if these QC measures are still state of the art or if there are QC tools that function as best practices.

Potential starting points:
goals and approaches
miQC
benchmarks QC, normalisation

Doublet detection & ambient RNA
Benchmark doublet detection

Cell-cell communication
Define current best practices in cell-cell communication in terms of between cell communication and prediction of ligand–target links between interacting cells. (Potential link to spatial omnics.)

Potential starting points:
NicheNet
Cellchat
cellphoneDB
omnipath

Normalization
BP1.0 suggested scran for normalization of non-full-length datasets. Scaling to a zero mean and unit variance was not preferred. Normalized data should be log(x+1) transformed to obtain normally distributed data. (Side note: scVI etc. uses raw counts for integration, mention this here as well.)

Newer approaches recommend to use GLM-PCA for non-normal distributions, Pearson residuals from regularized negative binomial regression and analytic Pearson residuals. These methods claim to outperform other normalization tasks in downstream analysis tasks like identifying biologically variable genes or dimensionality reduction. Goal is to provide an overview of these methods and check for potential independent benchmarks.

Potential starting points:
GLM-PCA
scTransform
statistical error models for scRNA-seq
Pearson residuals

We will split up the current notebook into several chapters and add more chapters later. However, for this to become easily readable we should add some a few optional introductory chapters like https://www.singlecellcourse.org/.
Should also emphasize that contrary to their book our content is NOT biased by their lab, but by best-practices

All chapters should have an overview.
All subchapters should have something like:

1. Overview/Introduction
2. Content
3. Quiz
4. Solutions
5. Main takeaways

This is STC

Should explain the complete wetlab workflow up to raw base calling.

A couple of figures will help.

Potential starting points:
General
Scvelo

Datasets:
Pancreas

Theislab experts:
Philipp, Marius, (Volker)

Interpreting cellular structure

Introduction for this chapter TODO.

• Multi omics paper by Fabi & Carlos is the paper to cite
• multi omics best practices comes in phase 2 later