Comp flow lec 081119 - University of Pittsburgh · 160914_B_ST0_ST_new.jo Layout_cells_D256_SP...

Lisa BorghesiAssociate Professor of Immunology

Director, Unified Flow Core

FLOWvergnügenthe power and pleasure of flow cytometry!

Intro to Computational Flow Outline

Cytometry technologiesconventional

spectralCyTOF

Power of algorithmic approachesunbiased, high-dimensional

access algorithms at Pittprinciples of algorithmic operations

Unified Flow Core Website

Conventional flow cytometry“one detector, one color”

PMT 1

PMT 2

PMT 5

PMT 4

DichroicFilters

BandpassFilters

Laser

Flow cell

PMT 3

forwardscatter

Sample

From Practical Flow Cytometry Third Edition, Howard M. Shapiro p.164.

Spectral Overlap

Our high speed analyzers

3 Cytek Auroras= spectral

3 BD LSRFortessa, 2 BD LSRII= conventional

Spectral flow cytometry: Cytek Aurora

collect ALL of the light àincreased sensitivity

discriminate fluors with overlapping spectra à not

possible with conventional flow cytometry

48 channels

FACSAntibodies labelled

with fluorochromes

CyTOFAntibodies labelled with

elemental isotopes

Conventional cytometry, 10-15 colors; beyond that

gets tough

Spectral technologies = game changer, 48

channels

40-50 parameters in a single tube

Ex. gadolinium (Gd), samarium (Sm), terbium

(Tb), thorium (Th), ytterbium (Yb)

Spectral cytometry rivals CyTOF

Advantages of Full Spectrum Flow Cytometry over CyTOF

• SPEED, flow cytometry is high-throughput at >35,000 cell/s while CyTOFprocesses at one-tenth this rate (

Power of Algorithmic Analysis

Power of Algorithmic Analysis

High-dimensional analysis – Human brain is not so hot at >3 dimensions. What does 18 dimensional data (18 color) even look like in high dimensional space?

Highlight patterns in the data – Manual analysis with a plethora of iterative bivariate plots is highly inefficient.

Facilitates population discovery – Meaningful populations can overlooked b/c biases and a priori knowledge dictate analysis.

Automated/more unbiased – Manual analysis is subjective and biased.

Fun!

spleen

blood

ILN

MLN

colon

LLN

jejunum

Creating a Human B cell Atlas

BM

CD19CD27CD38CD138IgMIgDIgGCD10CD95

CD45RBCD69CD21CD200CD218aCD370etc.

ERK/MAPKS6AKTmTORstat3zbtb32bach2

30+ donors

Courtesy: Florian Weisel

A day in the life of a FlowJo analyst

Donor 1, Tissue 1

D256 various samples – subsets CD45RB/ CD69

Naive B cellsCD21+ CD27-

atypical B cellsCD21- CD27-

classical MBCCD21+ CD27+

activated MBCCD21- CD27+

CD27

CD21

D256_SP D256_B D256_Col D256_T

160914_B_ST0_ST_new.jo Layout_cells_D256_SP

20.09.2016 11:34 Uhr Page 1 of 1 (FlowJo v9.7.6)

0 102 103 104 105

0102

103

104

105

6.23

89.2

0 102 103 104 105

0102

103

104

105

8.34

88.8

160914_B_ST0_ST_new.jo Layout_cells_D256_B


0 102 103 104 105: CD27

0102

103

104

105

: C

D21

2.93

96.4

0 102 103 104 105: CD27

0102

103

104

105

: C

D21

7.96

90.4

160902_gut_ST0_STnew.jo Layout_cells_D256_Col


0 102 103 104 105: CD27

0102

103

104

105

: C

D21

3.4

96.3

0 102 103 104 105: CD27

0102

103

104

105

: C

D21

5.36

94.3

160713_LS_screen.jo Layout: diff B cells


0 102 103 104 105

0102

103

104

105

2

97.3

0 102 103 104 105

0102

103

104

105

7.35

91.4

160914_B_ST0_ST_new.jo Layout: D256_SP_CD45RB_CD69_diffBcells


0 102 103 104 105

0102

103

104

105

2.61 1.84

11.284.3

0 102 103 104 105

0102

103

104

105

3.05 3.71

7.3885.8

0 102 103 104 105

0102

103

104

105

22.6 32.6

2420.7

0 102 103 104 105

0102

103

104

105

48.3 13.3

7.131.4

160914_B_ST0_ST_new.jo Layout: D256_B_CD45RB_CD69_diffBcells


0 102 103 104 105: CD69

0102

103

104

105

: CD4

5RB

4.08 0.0227

0.22795.7

0 102 103 104 105: CD69

0102

103

104

105

: C

D45R

B

2.98 0

1.9995.3

0 102 103 104 105: CD69

0102

103

104

105

: CD4

5RB

82.9 0.286

0.15916.6

0 102 103 104 105: CD69

0102

103

104

105

: CD4

5RB

40.1 0.361

0.36158.5

160902_gut_ST0_STnew.jo Layout: CD45RB_CD69_diffBcells_D256_Col


0 102 103 104 105: CD69

0102

103

104

105

: CD4

5RB

5.45 2.89

9.2282.4

0 102 103 104 105: CD69

0102

103

104

105

: CD4

5RB

14.8 22.7

18.244.3

0 102 103 104 105: CD69

0102

103

104

105

: CD4

5RB

11.2 73.7

11.63.55

0 102 103 104 105: CD69

0102

103

104

105

: CD4

5RB

16.4 57.2

15.410.9

160713_LS_screen.jo Layout: D256_T_CD45RB_CD69


0 102 103 104 105

0102

103

104

105

6.13 3.22

12.678.1

0 102 103 104 105

0102

103

104

105

8.57 5.99

9.0876.5

0 102 103 104 105

0102

103

104

105

30.1 39

18.112.8

0 102 103 104 105

0102

103

104

105

36.1 41.7

11.310.9

CD69

CD45

RB

Courtesy:Nadine Weisel

D256 various samples – IgG/ IgM distribution in B cell populations160914_B_ST0_ST_new.jo Layout_cells_D256_SP


0 102 103 104 105

0102

103

104

105

6.23

89.2

0 102 103 104 105

0102

103

104

105

8.34

88.8

160914_B_ST0_ST_new.jo Layout_cells_D256_B


0 102 103 104 105: CD27

0102

103

104

105

: C

D21

2.93

96.4

0 102 103 104 105: CD27

0102

103

104

105

: C

D21

7.96

90.4

160902_gut_ST0_STnew.jo Layout_cells_D256_Col


0 102 103 104 105: CD27

0102

103

104

105

: C

D21

3.4

96.3

0 102 103 104 105: CD27

0102

103

104

105

: C

D21

5.36

94.3

160713_LS_screen.jo Layout: diff B cells


0 102 103 104 105

0102

103

104

105

2

97.3

0 102 103 104 105

0102

103

104

105

7.35

91.4

160914_B_ST0_ST_new.jo Layout: D256_SP_IgG_IgM_diffBcells


0 102 103 104 105

0102

103

104

105

5.48 1.38

71.821.2

0 102 103 104 105

0102

103

104

105

2.06 2.82

74.220.9

0 102 103 104 105

0102

103

104

105

43.4 0.684

26.829.1

0 102 103 104 105

0102

103

104

105

40.8 3.79

34.321.1

160914_B_ST0_ST_new.jo Layout: D256_B_IgG_IgM_diffBcells


0 102 103 104 105: IgM

0102

103

104

105

: I

gG

0.128 1.31

95.82.75

0 102 103 104 105: IgM

0102

103

104

105

: I

gG

1.24 7.44

81.69.93

0 102 103 104 105: IgM

0102

103

104

105

: I

gG

8.01 1.4

68.522.1

0 102 103 104 105: IgM

0102

103

104

105

: I

gG

20.6 14.1

32.132.5

160902_gut_ST0_STnew.jo Layout: IgG_IgM_diffBcells_D256_Col


0 102 103 104 105: IgM

0102

103

104

105

: I

gG

1.04 0.12

79.219.7

0 102 103 104 105: IgM

0102

103

104

105

: I

gG

9.09 0

31.859.1

0 102 103 104 105: IgM

0102

103

104

105

: I

gG

9.47 0.0999

29.660.8

0 102 103 104 105: IgM

0102

103

104

105

: I

gG

15 0

11.773.2

160713_LS_screen.jo Layout: D256_T_IgG_IgM


0 102 103 104 105

0102

103

104

105

2.5 2.39

64.530.6

0 102 103 104 105

0102

103

104

105

8.46 1.7

61.128.9

0 102 103 104 105

0102

103

104

105

34.9 1.89

30.432.8

0 102 103 104 105

0102

103

104

105

44.5 0.655

17.237.6

D256_SP D256_B D256_Col D256_T

Naive B cellsCD21+ CD27-

atypical B cellsCD21- CD27-

classical MBCCD21+ CD27+

activated MBCCD21- CD27+

CD27

CD21

IgM

IgG

Courtesy:Nadine Weisel

Problem: how do you visualize millions of cells marked by a rainbow of colors in order to establish the underlying structure or principles of organization?

Solution: take high-D spaces and embed this information into low-D spaces that our brains can understand

à algorithms identify local patterns and global patterns that our brains may miss when performing manual analysis via iterative pairwise plots

D182 D205 D243D256 D269

Blood

D181D182 D205 D243D256

SPL

BM

merge

merge

merge

manual gateoverlaysindividual donors

D193 D260D182 D205 D215 D246D256

D164

2K10K 2K 2K 2K 2K

2K12K 2K 2K 2K 2K

2K14K 2K 2K 2K 2K 2K 2K

2K

Donor distance:

D181

D182

D205

D243

D256

D18

1

D18

2

D20

5

D24

3

D25

6

D164D182D205D243D256D269

D16

4D

182

D20

5D

243

D25

6D

269

D182D193D205D215D246D256D260

D18

2D

193

D20

5D

215

D24

6D

256

D26

0

BMBloodSPL

Naive BASCMBC IgM+MBC IgG+MBC swAg-exper. IgG+Ag-exper. sw

Unbiased analysis of across-donor heterogeneity

-40 -30 -20 -10 0 10 20 30 40bh-SNE1

-40

-30

-20

-10

0

10

20

30

40

bh-SNE2

-40 -30 -20 -10 0 10 20 30 40bh-SNE1

-40

-30

-20

-10

0

10

20

30

40

bh-SNE2

-40 -30 -20 -10 0 10 20 30 40bh-SNE1

-40

-30

-20

-10

0

10

20

30

40

bh-SNE2

-40 -30 -20 -10 0 10 20 30 40bh-SNE1

-40

-30

-20

-10

0

10

20

30

40

bh-SNE2

-40 -30 -20 -10 0 10 20 30 40bh-SNE1

-40

-30

-20

-10

0

10

20

30

40

bh-SNE2

-40 -30 -20 -10 0 10 20 30 40bh-SNE1

-40

-30

-20

-10

0

10

20

30

40

bh-SNE2

Blood BM SPL T LLN Col

D260

-40 -30 -20 -10 0 10 20 30 40bh-SNE1

-40

-30

-20

-10

0

10

20

30

40

bh-SNE2

-40 -30 -20 -10 0 10 20 30 40bh-SNE1

-40

-30

-20

-10

0

10

20

30

40

bh-SNE2

-40 -30 -20 -10 0 10 20 30 40bh-SNE1

-40

-30

-20

-10

0

10

20

30

40

bh-SNE2

-40 -30 -20 -10 0 10 20 30 40bh-SNE1

-40

-30

-20

-10

0

10

20

30

40

bh-SNE2

-40 -30 -20 -10 0 10 20 30 40bh-SNE1

-40

-30

-20

-10

0

10

20

30

40

bh-SNE2

-40 -30 -20 -10 0 10 20 30 40bh-SNE1

-40

-30

-20

-10

0

10

20

30

40

bh-SNE2

-40 -30 -20 -10 0 10 20 30 40bh-SNE1

-40

-30

-20

-10

0

10

20

30

40

bh-SNE2

-40 -30 -20 -10 0 10 20 30 40bh-SNE1

-40

-30

-20

-10

0

10

20

30

40

bh-SNE2

D256

merge

Naive BASCMBC IgM+IgD-MBC IgD+MBC swAg-exper. sw

manual gating

Tissue-specific B cell signatures

t-SNE FlowSOMSPADE

Algorithm Overview

CD19CD27IgDIgMCD38CD21

CD45RBCD69CD200CD218aCD370

star chart

CITRUS

PhenoGraph

Single cell resolution Tree of relationships

Automated population identification

Tree of relationships

Computational algorithms

Kimball 2017 JI

Choosing an Algorithm

Algorithms on site, actually free or basically free

1) FlowJo v.10+: t-SNE, SPADE, FlowMeans ($200/yr site license)

2) Cyt package: viSNE, PhenoGraph, Wishbone, Wanderlust, PCA, K-means, other statistical tools (free for faculty and students and $100/yr for everyone else)

3) SPADE http://pengqiu.gatech.edu/software/SPADE/ (free)

4) ExCYT https://github.com/sidhomj/ExCYT (free)

5) High performance cluster: viSNE, SPADE (free)

see supplemental slides at end of ppt for instructions on #2-5

Commercial Options

Cytobank: viSNE, SPADE, CITRUS, Sunburst; FlowSOM beta ($1,500/yr/user) www.cytobank.org

FCS Express: t-SNE, SPADE, K-means, PCAalso handles raw image files from the ImageStream($234-500/yr/license) www.denovosoftware.com

t-SNE – dimensionality reduction algorithm

Goal: find a low dimensional visualization that best reflects population structure in high dimensional space

à colloquially, get a feel for how objects are arranged in data space

tSNE_X

tSN

E_Y

Laurens van der Maaten explains t-SNE (UCSD seminar) – fun and informative!!

https://www.youtube.com/watch?v=EMD106bB2vY

PCA Swiss roll problem

• Principal Components Analysis (PCA) preserves large pairwise distances• Euclidean distance between two points on the Swiss roll does not

accurately reflect local structure

Why not just use PCA?

t-SNE preserves local distances and global distances

Amir 2013 Nature Biotech, Suppl.

PCA tSNE

High-D data space. Draw Gaussian bell (circle) around data point. Measure density of all other points relative to that Gaussian bell, and establish probability distribution that represents their similarity. Computes local densities to get a distribution of pairs of points. à Pij

Low-D 2D map. Repeat above. à Qij

Mathematically minimize P||Q difference. Zero would be if two points were the same.

t-SNE operation

Barnes-Hut Modification of t-SNE (bh-SNE)

t-SNE

Advantages

• single cell information

• non-linear assumptions (as opposed to PCA)

• preserves local and global structure

Limitations• computationally expensive; obligate downsampling means data are discarded

• plot axes are arbitrary and have no intrinsic meaning

• no population identification; follow up approaches required to assign identity

to clusters and cells

• distance between clusters is not meaningful; no hierarchy

tSNE_X

tSN

E_Y

SPADE: Hierarchical clustering algorithm

Goal: organize cells into a hierarchy using unsupervised approaches

à colloquially, generate a tree of relationships

spanning tree progression of density normalized events

Output minimal spanning tree (MST) highlights the relationships between most closely related cell type clusters

SPADE

SPADE views data as a cloud of points (cells) where the dimensions = # markers

Density-dependent downsamplingto equalize density in different parts of cloud, ensures rare cells not lost

Agglomerative clustering based on marker intensity

Connect clusters in minimal spanning tree that best reflects geometry of the original cloud

Upsampling, map each cell in the original data set to the clusters

SPADE

Advantages

• rare pops preserved through density-dependent downsampling

• enables visualization of continuity of phenotypes

• can combine data sets that share common markers, and then co-map

any markers unique to each data set (see orig. paper)

Limitations

• loss of single cell information

• user chooses cluster number

• MSTs are non-cyclic and paths can be artificially split

SPADE

t-SNE

PhenoGraph

Goal: automated partitioning of high-dimensional single-cell data into subpopulations

à colloquially, map nearest neighbors

PhenoGraph

First order relationship – find the k nearest neighbors for each cell using Euclidean distance

Second order relationships – cells with shared neighbors should be placed near one another

Third, identify communities – Louvain method that measures the density of edges inside communities to edges outside communities

PhenoGraph: Number of neighbors

Neighbors = 5 Neighbors = 30 Neighbors = 200

Manual gate overlays PhenoGraph

Naive BASCMBC (total)Ag-exper.

PhenoGraph – population discovery

PhenoGraph

Advantages

• opportunity for population discovery

• can resolve subpopulations as rare as 1 in 2000 cells

• robust to cluster shape (e.g., need not be spherical)

Limitations• user specifies number of neighbors

• ideal cluster number, or biologically relevant cluster number, is largely

unconstrained

Harinder Singh, PhD

• 2 Cytek Aurora instruments + computational flow toolset

• algorithms installed on HPC

• Center for Systems Immunology (CSI)

• harness newly emerging experimental and computational approaches

• programmatic analysis pipelines

Looking Forward

2019 Cytometry Bootcamp

AugustIntro to Flow Cytometry - Tue, August 6, 10-12pm E1095Intro to Computational Flow Cytometry - Mon, August 12, 10-11am E1095Algorithm Installation - Tue, August 20, 10-11am E1095

SeptHelp Session for Computational Flow Cytometry - Tue, Sept 10, 10:30-11:30am E1095Basic Flow Cytometry Panel Design - Thu, September 26, 9:30-10:30am E1095

OctAurora Advanced Panel Design - Wed, October 9, 9-10am W1039Intro to Computational Flow Cytometry - Thu, November 7, 9-10am E1095

NovAlgorithm Installation - Mon, November 11, 3-4pm E1095Help Session for Computational Flow Cytometry - Thu, November 21, 9-10am E1095

Acknowledgements

Dewayne FalknerDirector of Operations

Aarika MacIntyreSenior Technologist

Ailing LiuApplications Specialist

Unified Core, BoardMark ShlomchikFadi LakkisMark GladwinLarry MorelandJohn McDyer

Center for Research ComputingKim Wong

Nan ShengCore Technician

Heidi GunzelmanSenior Flow Cytometry Specialist

ResourcesUseful starting places - reviews1. Kimball AK, Oko LM, Bullock BL, Nemenoff RA, van Dyk LF, Clambey ET. A Beginner's Guide to Analyzing and

Visualizing Mass Cytometry Data. J Immunol. 2018 Jan 1;200(1):3-22.2. Saeys Y, Gassen SV, Lambrecht BN. Computational flow cytometry: helping to make sense of high-dimensional

immunology data. Nat Rev Immunol. 2016 Jul;16(7):449-62.3. Mair F, Hartmann FJ, Mrdjen D, Tosevski V, Krieg C, Becher B. The end of gating? An introduction to automated

analysis of high dimensional cytometry data. Eur J Immunol. 2016 Jan;46(1):34-43.4. Chester C & Maecker HT. J Immunol. 2015 Aug 1;195(3):773-9. doi: 10.4049/jimmunol.1500633. Algorithmic Tools

for Mining High-Dimensional Cytometry Data. J Immunol. 2015 Aug 1;195(3):773-9.

Original application of algorithms1. Qiu P, Simonds EF, Bendall SC, Gibbs KD Jr, Bruggner RV, Linderman MD, Sachs K, Nolan GP, Plevritis SK.

Extracting a cellular hierarchy from high-dimensional cytometry data with SPADE. Nat Biotechnol. 2011 Oct 2;29(10):886-91. SPADE

2. Amir el-AD, Davis KL, Tadmor MD, Simonds EF, Levine JH, Bendall SC, Shenfeld DK, Krishnaswamy S, Nolan GP, Pe'er D. viSNE enables visualization of high dimensional single-cell data and reveals phenotypic heterogeneity of leukemia. Amir el-AD, Davis KL, 3. Tadmor MD, Simonds EF, Levine JH, Bendall SC, Shenfeld DK, Krishnaswamy S, Nolan GP, Pe'er D. Nat Biotechnol. 2013 Jun;31(6):545-52. viSNE

3. Levine JH, Simonds EF, Bendall SC, Davis KL, Amir el-AD, Tadmor MD, Litvin O, Fienberg HG, Jager A, ZunderER, Finck R, Gedman AL, Radtke I, Downing JR, Pe'er D, Nolan GP. Data-Driven Phenotypic Dissection of AML Reveals Progenitor-like Cells that Correlate with Prognosis. Cell. 2015 Jul 2;162(1):184-97. PhenoGraph

4. Bruggner RV, Bodenmiller B, Dill DL, Tibshirani RJ, Nolan GP. Automated identification of stratifying signatures in cellular subpopulations. Proc Natl Acad Sci U S A. 2014 Jul 1;111(26):E2770-7. CITRUS

5. Setty M, Tadmor MD, Reich-Zeliger S, Angel O, Salame TM, Kathail P, Choi K, Bendall S, Friedman N, Pe'er D. Wishbone identifies bifurcating developmental trajectories from single-cell data. Nat Biotechnol. 2016 Jun;34(6):637-45. Wishbone

6. Bendall SC, Davis KL, Amir el-AD, Tadmor MD, Simonds EF, Chen TJ, Shenfeld DK, Nolan GP, Pe'er D. Single-cell trajectory detection uncovers progression and regulatory coordination in human B cell development. Cell. 2014 Apr 24;157(3):714-25. Wanderlust

Install Cyt3 on Your Personal ComputerviSNE, PhenoGraph, (Wanderlust, Wishbone, and more)

1. Install Matlab from my.pitt.edu software downloads• Free for students and teaching faculty• Everyone else, $100/yr license

2. Install Cyt3, obtain installer from Unified Flow CoreMac People – Cyt3 installer for MacWindows People – Master Cyt3 installer, bh_tsne_win64 file

• Then, in the Cyt3 à src à 3rdparty à bh_tsne folder remove bh_tsne_mac64 and replace with bh_tsne_win64

3. Follow CyTutorial ppt instructions provided by installer to perform analysis• CyTutorial instructions lean toward the skeletal• implementation can be tricky• no longer supported by developer Dana Pe’er

viSNE

Amir et al. Nat Biotechnol. 2013 Jun;31(6):545-52

Install SPADE 3.0 on Your Personal Computer

Easy to install and use!

1. For Mac or Windows2. Stand-alone (or via Matlab)3. Great instructions provided at URL4. Supported by developer Peng Qiu

http://pengqiu.gatech.edu/software/SPADE/

Qiu et al. Nat Biotechnol. 2011 Oct 2;29(10):886-91

ExCYT

• interactive gating• gate directly on tSNE plots• tSNE and then re-tSNE on gated subsets• novel high-dimensional flow plots

Manuscript & video:https://www.jove.com/video/57473/excyt-graphical-user-interface-for-streamlining-analysis-high?status=a59479k&fbclid=IwAR2i0i-M51YhAkrEermEwtpinvPLPccc_n88pXmrfZPwhhu9tZLbZTzvQuA

Software:https://github.com/sidhomj/ExCYT

https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.jove.com%2Fvideo%2F57473%2Fexcyt-graphical-user-interface-for-streamlining-analysis-high%3Fstatus%3Da59479k%26fbclid%3DIwAR2i0i-M51YhAkrEermEwtpinvPLPccc_n88pXmrfZPwhhu9tZLbZTzvQuA&data=02%7C01%7Cborghesi%40pitt.edu%7C90fd12c6477440e018b808d65addcbc0%7C9ef9f489e0a04eeb87cc3a526112fd0d%7C1%7C0%7C636796305975445391&sdata=X6LbX0ZaywG0l2eJMzwbtZ2YyX0nqH9z3bdM8%2F0U9xE%3D&reserved=0https://github.com/sidhomj/ExCYT

Comp flow lec 081119 - University of Pittsburgh · 160914_B_ST0_ST_new.jo Layout_cells_D256_SP...

Documents

Transcript of Comp flow lec 081119 - University of Pittsburgh · 160914_B_ST0_ST_new.jo Layout_cells_D256_SP...