2011 Data Analysis

Report
Data Analysis
FSC SSC FITC PE
% Total
Mean
Geometric mean
CV
% of Parent
Mode
Median
% of Grandparent
Data Analysis
Experimental Design
Sample Procurement
Sample preparation
Fix/Perm
Which Fluorophore
Controls
Isotype?
Single color
FMO
Instrumentation
Appropriate Lasers
Appropriate Filters
Instrument Settings
Lin vs Log
Time
A, W, H
Interpretation
Mean, Median
%+
CV
SD
Signal/Noise
Gating
Analysis
Presentation
Histogram
Dot Plot
Density Plot
Overlay
Bar Graph
Objective
This will involve both the “what is” data analysis and the “how to”.
• Instrumentation (Mostly covered in “Flow Basics”)
• Underlying Principles/Concepts
• Proper Technique
First, lets address the problems
• Data analysis incorporates many disciplines
including instrumentation, statistics, biology, and
photonics. Often times knowledge in one of these is
limited
• Many different instruments and software packages
are available.
• Historical precedento Unfortunately there is a large body of work published with poor data and
no clear guidelines
Flow Basics
• A quick look back to Flow Basics to see what we are
analyzing
Photons ElectronsVoltage pulseDigital #
Measurements of the Pulse
Pulse Width
Voltage Intensity
Pulse Height
Pulse Area
Time
1
256
10,000
196
1000
.1
(Volts)
3.54 volts
(Volts)
6.21 volts
128
.01
64
.001
(1mV)
0
1.23 volts
0
Photons=voltage=relative brightness
100
10
1
Relative Brightness
10
Channel Number
10
FCS File
or
List Mode File
FCS File- The flow cytometry data file
standard provides the information
needed to completely describe flow
cytometry data sets within the confines
of the file containing the experimental
data. It is made up of the Header, Text
and Data.
Current version is FSC 3.0
Data
FSC
SSC
FITC
PE
APC
APC-Cy7
FCS file Text
Keywords- value pairs that describe the experiment, the instrument, the
specimen, the data and any other information which the file creator
wishes to include. Some values are user defined, others are
automatically entered
Histograms
120
N=552
Frequency
100
80
60
40
20
0
1
3
5
7
9
11 13 15 17 19
# of Apples picked per tree
Frequency
N=7696
0
101
102
103
104
Arbitrary Fluorescent units
Histogram- is a summary graph showing a
count of the data points falling in various
ranges.
The effect is a rough approximation of the
frequency distribution of the data.
Creation of a Histogram
Listmode File
Event #
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
FSC
500
645
6416
64186
313
354684
654165465
51
5165
5418
654164
3135
684
6541
3546
35416
3154564
1354
3154
164
SSC
PE
6176
6146
3543
315
34
3054
3054
8668
6840
80
9354
35
48979
540
354
354
304
354
68
68
9
2013
3455
251
11
1265
25
659
6145
256
3698
4848
5
100
5468
154
8
7
9852
3652
12
9
6
3
0
101
102
PE
103
104
Dot Plots
Dot plot- Frequency distributions of two
simultaneously displayed parameters
Plots
Contour
Pseudo-color
Histogram
Greyscale Density
Dot plot
Dot plots vs density or Pseudocolor
6,000,000
Events
1,500,000
6,000,000
4,500,000
600,000
300,000
3,000,000
150,000
Gating
Selecting a subset of cells based on a specific fluorescent or scatter pattern.
-Either graph can be OK
-However “gated” may be more
appropriate due to exclusion of
artifact and non T cells
No Gate
CD3+ Gate
11.3%
59.8%
5.8%
30%
Interpretation
11.3% of the WBC’s are positive for CD4
and 5.8% are positive for CD8
Interpretation
59.8% of the CD3+ cells are positive for
CD4 and 30% are positive for CD8
Hierarchical Gating
Compensation
Fluorescent Compensation- Due to broad emission profiles of fluorescent
dyes, a single detector may process light from multiple dyes. The process of
mathematically correcting for this spectral overlap is called compensation
Compensation
Mean/Median/GeoMean
• Mean
• Sum of the “n” individual values of a group divided by n
• NOT GOOD FOR Log DATA
• More easily skewed by outliers
• Geometric Mean
• Multiply the ‘N” values of a cluster together and get the nth
root of this product
• Better for log data
• Median
• Divides your data in half
• Probably the safest for Log data
http://flowjo.typepad.com/the_daily_dongle/2007/10/mean-median-mod.html
Location of mean, median and mode with
different distributions.
Normal
Left skew
Right skew
Driscoll P et al. J Accid Emerg Med 2000;17:274-281
Objective
This will involve both the “what is” data analysis and the “how to”.
• Instrumentation (Mostly covered in “Flow Basics”)
• Underlying Principles/Concepts
• Proper Technique
There is no wrong way to analyze your data
Meaning- Investigators are free to choose:
• Which plot types for display
• Placement of gates for analysis
• Which statistics
• # events to display or collect
• Which software package to use
• How many times you reanalyze
There is definitely a wrong way to analyze
your data
Meaning- Investigators decisions can lead to incorrect
data generation or interpretation:
• Inappropriate gates for analysis (lymphocyte gate
for CD15 staining, or inconsistent gates)
• Misleading or inconsistent plots for display
• Inappropriate controls (e.g. using isotype for gating)
• Inappropriate number of events collected (too few
events for meaningful and accurate statistical
comparison)
Dot plot Vs Histogram
Doublet discrimination
Height
• Cell cycle
• Rare events
• Dim cell lines
Width
Red=Singlets
Blue=Doublets
Fiona Craig
Flow Cytometric Immunophenotyping of
Cerebrospinal Fluid SpecimensAJCP 2011
135:22-34
Cell aggregates were identified
in 29 (16.4%) of 177
CSF specimens and represented
0 to 1,503 events and 0% to
80% of total acquired events
On axis data
Bi-exponential Scaling
•
•
•
•
Approximates log scaling at the upper end
Approximates linear scaling at the lower end
Can display events at or below zero
Improves visual separation of population at the low
end of the scale
http://ucflow.blogspot.com/2011/04/displaytransformation-and-flowjo.html
Log scale
Bi-exponential
Herzenberg et al
Isotypes
• Must be the exact same isotype
• Conjugated to exactly the same degree
Fluorophore/Protein ratio
• Has the same background binding characteristics as your
antibody
• Different vendors use different conjugation protocols
which may cause different characteristics
• Used at the same concentration
• Not used for gating purposes!!!
Rarely is more than the first criterion met.
FMO controls
FMO = Fluorescence minus one
The gating control contains all the fluorophores except the
PE
one of interest
FITC
Perfetto et al
Rare events
The essential feature of Poisson distributions is that if N events are observed the standard deviation (SD) associated
with that count is square root of N . The coefficient of variation (CV) is then given by
CV = 100 X SD/ N or 100 / sqrt N.
How many events to detect is more of a statistics questions
than a flow cytometry question
Statistical Significance of Results
• Determining the Number of Events to Collect
Rare Events Case Study #1
• Number of Events vs Measurement Precision
.1%-.18%
.07%-.1%
Rare Events Case Study #2
• Normal Kappa/Lambda ratio on B cells is .9-3.5 (polytypic)
• When values fall outside this range it’s “clonal”
• Often times in patients undergoing therapy there is a very low
frequency of B cells to enumerate.
Gated on CD19+ Bcells
20
Kappa/Lambda=58/20=2.9(Polytypic)
But, √58=8 and √20=5
So actual values are 58±8/20±5
Which leads to ratio’s between 4.4 and 2.
Is this clonal???????
58
4 decade or 5 decade display
FACSDiVa
•
•
•
The Diva system first converts the analog signal to digital using a
14-bit ADC (16, 384 bins), then when the data is log transformed,
it is multiplied by 16 (or 24) effectively adding 4 bits to the data
making it 18-bit (262,144 bins).
It displays this data on a 5-decade scale, but then does not
display the first decade.
What you’re left with is a 4-decade log scale (actually the scale
goes above the 4th decade a little bit) with bins from 262,144 down
to 262.
Decade
Analog
PMT
Signal
14-bit Bins
Upscaled to
18-bit
Decade Range
5
10V
16384
262144
10001-100000+
4
1V
1638
26214
1001-10000
3
.1V
164
2621
101-1000
2
.01V
16
262
11-100
1
.001V
2
26
1-10
Dead Cell Exclusion
Immunophenotyping
Despite light
scatter
gate
Rare event analysis
Prefetto et al JIM 2006
Dump gate
With dump gate
Without dump gate
Light Scatter
Laser Beam
FSC
Detector
Collection
Lens
SSC
Detector
Original from Purdue University Cytometry Laboratories
Why Look at FSC v. SSC
• Since FSC ~ size and SSC ~ internal structure, a
correlated measurement between them can allow
for differentiation of cell types in a heterogenous
cell population
Granulocytes
Dead
SSC
Lymphocytes
LIVE
Monocytes
RBCs, Debris,
Dead Cells
FSC
Kinetic and Ratio parameters
Automated clustering
algorithms
Aghaeepour N, Nikolic R, Hoos HH, Brinkman RR.
Rapid cell population identification in flow cytometry.
Cytometry A. 79(1):6-13, 2011.
Zare H, Shooshtari P, Gupta A, Brinkman RR. Data
reduction for spectral clustering to analyze high
throughput flow cytometry data. BMC Bioinformatics.
11(1): 403, 2010.
Available Spectra Viewers
•
http://www.bdbiosciences.com/research/multicolor/spectrum_viewer/index.jsp
•
http://www.invitrogen.com/site/us/en/home/support/Research-Tools/FluorescenceSpectraViewer.html
•
http://www.biolegend.com/spectraanalyzer
Data Analysis Feedback
http://www.surveymonkey.com/s/K7WYTH7
References
-Perfetto S, 2006 JIM 313 pg199-208 (live vs dead)
-Keeney et al 1998 Cytometry
-Cytometry 30(5), 1997
http://ucflow.blogspot.com/2011/04/display-transformation-and-flowjo.html (bi-exponential display)
-Cytometry A 783A:384-385
-Seventeen-colour flow cytometry: unravelling the immune system
-Stephen P. Perfetto, Pratip K. Chattopadhyay & Mario Roederer
Nature Reviews Immunology 4, 648-655 (August 2004)
-Interpreting flow cytometry data: a guide for the perplexed
Herzenberg L, 2006 Nature Immunology Vol 7 Num 7
A practical approach to multicolor flow cyometry for immunophenotyping
Baumgarth N, 2000 J. Imm. Methods 243 77-97
-Isotype controls in the Analysis of Lymphocyte and CD34+ stem and progenitor cells by Flow
Cytometry-Time to Let Go!
Keeney, M 1998 Cytomery (Communication in Clincal Cytometry 34: 280-283
# of Parameters
1
2
3
4
5
6
7
8
9
10
# of possible plots
1
1
3
6
10
15
21
28
36
45

similar documents