Single-cell analysis workshop Sydney Precision Bioinformatics Group - - PowerPoint PPT Presentation

▶
single cell analysis workshop
SMART_READER_LITE
LIVE PREVIEW

Single-cell analysis workshop Sydney Precision Bioinformatics Group - - PowerPoint PPT Presentation

Dec 16, 2022 148 likes •714 views

Single-cell analysis workshop Sydney Precision Bioinformatics Group The University of Sydney Page 1 Sydney Precision Bioinformatics Research Group We share an interest in developing statistical and computational methodologies to tackle the

slide-1
SLIDE 1

The University of Sydney Page 1

Single-cell analysis workshop

Sydney Precision Bioinformatics Group

slide-2
SLIDE 2

The University of Sydney Page 2

Sydney Precision Bioinformatics Research Group

We share an interest in developing statistical and computational methodologies to tackle the foremost significant challenges posed by modern biology and medicine. Meet our senior and junior research leaders and senior research associates, PhD candidates, Honours and TSP students: 25 Find out more: http://www.maths.usyd.edu.au/bioinformatics/ Get interactive: http://shiny.maths.usyd.edu.au/

Jean Yang Samuel Muller John Ormerod PengyiYang Ellis Patrick Rachel Wang Garth Tarr Kitty Lo

slide-3
SLIDE 3

The University of Sydney Page 3

Roadmap for the workshop

  • Setting up: 1:15 – 1:30 Google cloud set up
  • Session 1: 1:30 – 2:00 Single cell analysis overview (scdney)
  • Session 2: 2:00 – 2:45 Quality control and data integration
  • Session 3: 2:45 – 3:45 Cell type identification via cluster analysis
  • Session 4: 3:45 – 4:30 Downstream analysis: identify marker genes & cell type

composition

  • Extension: cell type identification via supervised classification and single cell trajectory

analysis Workshop presenters in each session: Jean Yang, Kevin Wang, Pengyi Yang, Yingxin Lin

slide-4
SLIDE 4

The University of Sydney Page 4

Configuring Google Cloud

–Machine 1: 34.69.169.142 –Machine 2: 34.94.220.230 source("/home/user_setup.R")

slide-5
SLIDE 5

The University of Sydney Page 5

Exponential growth in single cell RNA seq technologies

Svensson et al. Nature Protocols (2018)

slide-6
SLIDE 6

The University of Sydney Page 6

Droplet based technologies are now dominating

Macosko et al. (2015), Cell

10X Genomics is a commercial provider of droplet based scRNAseq platform

slide-7
SLIDE 7

The University of Sydney Page 7

scRNAseq experiments approaching 1 million cells

Saunders et al., (2018) Cell

690,000 individual cells from 9 regions

  • f adult mouse brain
slide-8
SLIDE 8

The University of Sydney Page 8

Number of scRNAseq tools also increasing rapidly

Downloaded from www.scrna-tools.org

slide-9
SLIDE 9

The University of Sydney Page 9

Single-cell RNA-seq analysis

slide-10
SLIDE 10

The University of Sydney Page 10

Components of a typical scRNA-seq analysis process

slide-11
SLIDE 11

The University of Sydney Page 11

Component 1: Data acquisition

Software

  • CellRanger for 10X Genomics data
  • Macosko’s custom scripts for DropSeq data
  • STAR for alignment plus custom scripts (or there

is STAR-solo)

Considerations

  • Single or mix of species? Does it include ERCC

spike-ins? May need to build a custom reference

  • Barcode and/or UMI sequencing errors –

CellRanger takes care of this automatically

  • Align to exon or exon and intron?

Input

  • BCL or fastq file from the sequencer

Output

  • Gene/cell counts matrix
slide-12
SLIDE 12

The University of Sydney Page 12

Component 2: Data preprocessing – Quality control

Software

  • Seurat (all-purpose single cell R package)
  • Scater
  • DropletUtils (R package with a number of handy

utility functions)

  • Your own custom scripts

Considerations

  • Filter out droplets with doublets – may be difficult

to find. Can estimate expected rate by doing species mixture experiment

Croset (2018), eLife

slide-13
SLIDE 13

The University of Sydney Page 13

Component 2: Data preprocessing – Quality control

Software

  • Seurat (all-purpose single cell R package)
  • Scater
  • DropletUtils (R package with a number of handy

utility functions)

  • Your own custom scripts

Considerations

  • Filter out droplets with doublets – may be difficult

to find. Can estimate expected rate by doing species mixture experiment

  • Filter out droplets with no cells
slide-14
SLIDE 14

The University of Sydney Page 14

Component 2: Data preprocessing – Quality control

Software

  • Seurat (all-purpose single cell R package)
  • Scater
  • DropletUtils (R package with a number of handy

utility functions)

  • Your own custom scripts

Considerations

  • Filter out droplets with doublets – may be difficult

to find. Can estimate expected rate by doing species mixture experiment

  • Filter out droplets with no cells
  • Filter out droplets with damaged cells – look for

high mitochondrial gene content or high spike-in

slide-15
SLIDE 15

The University of Sydney Page 15

Component 3: Data integration

Software

  • Seurat (all-purpose single cell R package) for very

basic normalization

  • Batch effect correction
  • mnnCorrect
  • ZINB-Wave
  • scMerge
slide-16
SLIDE 16

The University of Sydney Page 16

scMerge motivation - Liver fetal development time course dataset

GSE87795 Su et al.

E9.5 E10.5 E11.5 E12.5 E13.5 E14.5 E15.5 E16.5 E17.5

slide-17
SLIDE 17

The University of Sydney Page 17

E9.5 E10.5 E11.5 E12.5 E13.5 E14.5 E15.5 E16.5 E17.5

GSE87795 Su et al. GSE90047 Yang et al. GSE87038 Dong et al. GSE96981 Camp et al. N = 320 cells N = 389 cells N = 79 cells N = 448 cells

Liver fetal development time course datasets

slide-18
SLIDE 18

The University of Sydney Page 18

tSNE of liver fetal development time course datasets

Highlighted by cell types Highlighted by batches

Challenge: Strong “batch effect”

slide-19
SLIDE 19

The University of Sydney Page 19

Breaking observed data into components

The data we observe

For n cells with data collected for m genes

Biologically relevant variation cell types p wanted variables Unwanted variation batch and technical effects k unwanted variables Random noise

slide-20
SLIDE 20

The University of Sydney Page 20

scMerge algorithm

RUVIII algorithm Molania et al. (2019), Nuclei Acids Res Estimated with replicates by factor analysis Estimated by stably expressed genes by factor analysis

slide-21
SLIDE 21

The University of Sydney Page 21

scMerge algorithm

Pseudo- replicates

Find Mutual Nearest Clusters as pseudo-replicates Clustering for each batch

(k-means by default)

Frame as pseudo-replicate information

slide-22
SLIDE 22

The University of Sydney Page 22

Coming back to our motivational data – Liver fetal development time course datasets

−20 20 40 −20 20

tSNE1 tSNE2

logcounts

−20 −20 −10

tSNE2

−20 −20 −20 20 −20 −10 10 20 30

tSNE1 tSNE2

scMerge_scSEG

cell_types

cholangiocyte Endothelial Cell Epithelial Cell Hematopoietic hepatoblast/hepatocyte Immune cell Mesenchymal Cell Stellate Cell

batch

GSE87038 GSE87795 GSE90047 GSE96981

Before scMerge After scMerge

slide-23
SLIDE 23

The University of Sydney Page 23

More information

scMerge R package and website: https://sydneybiox.github.io/scMerge/ PNAS:

https://doi.org/10.1073/pnas.1820006116

slide-24
SLIDE 24

The University of Sydney Page 24

We will try this soon … 2:00 – 2:45 Quality control and data integration

slide-25
SLIDE 25

The University of Sydney Page 25

Component 4: Cell type identification

Science questions

  • What cell types are present in the dataset?
  • Can we identify the cell types?
slide-26
SLIDE 26

The University of Sydney Page 26

Phase 3: Cell assignment

Science questions

  • What cell types are present in the dataset?
  • Can we identify the cell types?

Analysis techniques

  • Visualization (dimension reduction)
  • Clustering (unsupervised learning)
  • Classification (supervised learning)
slide-27
SLIDE 27

The University of Sydney Page 27

Dimension reduced plot of our data (tSNE plot)

−20 −10 10 20 −20 −10 10 20

tsne1 tsne2

t−SNE plot

How many cell types are there? What are the cell types?

slide-28
SLIDE 28

The University of Sydney Page 28

k-means clustering

−20 −10 10 20 −20 −10 10 20

tsne1 tsne2

t−SNE plot

How many cell types are there? What are the cell types?

slide-29
SLIDE 29

The University of Sydney Page 32

Clustering algorithms for scRNA-seq

k-means Hierarchical RaceID SC3 CIDR countClust RCA SIMLR

Luke Zappia, et al. PLoS Comp. Bio. 2018

25%+

slide-30
SLIDE 30

The University of Sydney Page 33

Similarity metric is the core of clustering algorithm

k-means Hierarchical RaceID SC3 CIDR countClust RCA SIMLR Spearman Pearson Euclidean Manhattan Maximum

Key question: is there a similarity metric that performs (on average) better for clustering single cells based on their transcriptome?

Correlation-based Distance-based

slide-31
SLIDE 31

The University of Sydney Page 34

k-means Clustering on GSE60361

k-means

Zeisel A, et al. Science 2015

pre-defined cell types

k-means Clustering on GSE60361

slide-32
SLIDE 32

The University of Sydney Page 35

Evaluation framework

Agreement to pre-defined classes: Normalized Mutual Information (NMI) Adjusted Rand Index (ARI) Fowlkes-Mallows Index (FM) Jaccard Index (Jaccard)

Taiyun Kim

slide-33
SLIDE 33

The University of Sydney Page 36

Evaluation results (against the pre-defined cell types)

Multiple datasets

PhD student: Taiyun Kim

slide-34
SLIDE 34

The University of Sydney Page 37

Evaluation results (against the pre-defined cell types)

On average, correlation-based metrics improved on distance-based metrics by 31.5% (NMI), 39.6% (ARI), 16% (FM), 23% (Jaccard)

Evaluation results (against the pre-defined cell types) using other measures

slide-35
SLIDE 35

The University of Sydney Page 38

Linnorm normalisation SAVER imputation Additional processing

Agreement to pre-defined classes: Normalized Mutual Information (NMI) Adjusted Rand Index (ARI) Fowlkes-Mallows Index (FM) Jaccard Index (Jaccard)

Account for data scaling and zero-counts

slide-36
SLIDE 36

The University of Sydney Page 39

Account for normalisation and imputation

slide-37
SLIDE 37

The University of Sydney Page 40

SIMLR

Improving the state-of-the-art clustering method using correlation metric

slide-38
SLIDE 38

The University of Sydney Page 41

Evaluation results of SIMLR with Pearson or Euclidean metrics

slide-39
SLIDE 39

The University of Sydney Page 42

Cells Genes PCA… Cells PCs Problem of PCA is that PCs can only be linear combination of genes: 𝑨𝑗1 = 𝜚11𝑦𝑗1 + 𝜚21𝑦𝑗2 + ⋯+ 𝜚𝑞1𝑦𝑗𝑞

Extension: Methods for accounting high-dimensionality of scRNA-seq

slide-40
SLIDE 40

The University of Sydney Page 43

Input layer Output layer Encode layer Hidden layer Hidden layer

Autoencoder, a deep learning model, allows nonlinear dimension reduction Random projection based ensemble of autoencoders allow multiple views of the scRNA-seq data from different “angles”

Dimension reduction using an ensemble of autoencoders

slide-41
SLIDE 41

The University of Sydney Page 44

Autoencoder input Raw input

Ensemble of autoencoders – does it work (with k-means)?

slide-42
SLIDE 42

The University of Sydney Page 45

More benchmark using advanced clustering algorithm

Geddes T et al., Autoencoder-based cluster ensembles for single-cell RNA-seq data analysis, BMC Bioinformatics (2019)

More benchmark of autoencoder ensemble with PCA using k-means & SIMLR

slide-43
SLIDE 43

The University of Sydney Page 46

We will try this soon…

2:45 – 3:45 Cell type identification via clustering analysis (scClust)

slide-44
SLIDE 44

The University of Sydney Page 47

scClassify: Algorithm

Feature selection at each branch point. Features are selected from :

  • Differential expression analysis;
  • Differential variability analysis;
  • Differential distribution analysis;
  • Chi-squared test,

……

Macrophage Monocytes non−classic−monocyte
  • steoclast
DC immature−DCmature−DC pDC dysf−cd4 NKExhausted Cd8+ cells transitional Regulatory T−cells Tfh Cytotoxicity Memory T−cells naive B cell Plasma cell

DC

Monocyte T cells + NK cells T cells + NK cells + B cell

PhD student: Yingxin Lin

slide-45
SLIDE 45

The University of Sydney Page 48

Component 5: Downstream analysis

Science questions

  • Which genes are differentially expressed between

cell types?

  • What are the marker genes for each cell type?
  • What is the cell type composition?
  • Are the cells transitioning from one state to

another?

slide-46
SLIDE 46

The University of Sydney Page 49

Cell type proportions

Can we conclude that there are more cholangiocytes than mesenchymal cells?

slide-47
SLIDE 47

The University of Sydney Page 50

scDC simulates uncertainty in cell-type proportions via bootstrapping Main components:

  • Sample with replacement from count

matrix, stratified by patient

  • Cell type identification via clustering

(PCA -> Kmeans (Pearson correlation)

  • Calculations of cell – type proportions

standard error from bootstrap samples

  • Calculation of pooled log-linear model

using Rubin’s pooled estimate

Single cell Differential Composition (scDC)

PhD student: Yue Cao

slide-48
SLIDE 48

The University of Sydney Page 51

– Examined two synthetic datasets constructed from two sets of real experimental data — Pancreas (T2D vs healthy) and Neuronal (developing mouse) – In pancreas dataset

  • confirmed the original finding that 1
  • f the 4 subjects has a higher beta

cell value, as IQR non overlap

– In neuronal dataset

  • Revealed new finding that

progenitor cells percentage increase

  • ver time

Single cell Differential Composition (scDC)

Supplementary

slide-49
SLIDE 49

The University of Sydney Page 52

Differences between single cell and bulk RNAseq

– Single cell gene expressions show a bimodal expression pattern – abundant genes are either highly expressed or undetected. – This can be technical (drop-outs) or biological (transcriptional bursts). – Drop-outs lead to technical zeroes in the data. – Technical zeroes are due to low capture efficiency in scRNAseq experiments. – Many methods have been proposed to deal with drop-outs

slide-50
SLIDE 50

The University of Sydney Page 53

Differential expression analysis

– Simple statistical test

– Wilcoxon rank test, t-test

– Methods developed for bulk RNAseq DE – DESeq2

– EdgeR – Voom-Limma

– scRNA specific

– MAST – DECENT – D3E – …. many more!

slide-51
SLIDE 51

The University of Sydney Page 54

Sonesonand Robinson (2018) Nature methods

DE methods comparisons for scRNAseq

slide-52
SLIDE 52

The University of Sydney Page 55

Pseudotime inference

– Why pseudotime?

– Sometimes cells do not occupy discrete states, rather cell states may follow a smooth trajectory – Example: stem cell differentiation

– What is pseudotime?

– Abstract unit of progress along some trajectory

– Typical steps involved in pseudotime inference:

– Reduce the dimensionality of the data – Build some kind of lineage structure – Order the cells in pseudotime

slide-53
SLIDE 53

The University of Sydney Page 56

Comparisons of pseudotime inference methods

Saelens et al., (2019) Nature Biotechnology

slide-54
SLIDE 54

The University of Sydney Page 57

Slingshot example (Street et al., 2018)

Two stages:

1. Inference of the global lineage structure. Uses cluster-based minimum spanning tree 2. Inference of pseudotime variables for cells along each lineage. Fits simultaneous principal curves

slide-55
SLIDE 55

The University of Sydney Page 58

  • We will try this soon…

3:45 – 4:30 Downstream analysis: identify marker genes & cell type composition Extension: cell type identification via supervised classification and single cell trajectory analysis