Transcript cellHTS class and workflow components for cell

High-throughput cell-based
assays
Wolfgang Huber
European Molecular Biology Laboratory
European Bioinformatics Institute
Signaling pathways
Drosophila antibacterial
signalling
Drosophila Toll/antifungal
signaling
Interference/Perturbation tools
Small compounds
Full-length cDNA (over-)expression
RNAi
RNAi
Initiation
Execution
He and Hannon, 2004
RNAi as a loss of function perturbator
degradation
protein
gene-sequence
specific reagents
(eg siRNAs)
easy to make for
any gene
(there are
caveats...)
translation
mRNA
transcription
gene
living cells
RNAi as a loss of function perturbator
degradation
protein
gene-sequence
specific reagents
(eg siRNAs)
easy to make for
any gene
(there are
caveats...)
mRNA
transcription
gene
living cells
RNAi as a loss of function perturbator
gene-sequence
specific reagents
(eg siRNAs)
easy to make for
any gene
(there are
caveats...)
mRNA
transcription
gene
living cells
C. elegans
E. coli
Drosophila
Mammals
dsRNA
dsRNA
siRNA
> 200bp
> 200bp
21bp
T7
Feeding
bacteria
Worms
Precursor
dsRNA
Injection
and soaking
Bathing
Cellculture
Transfection
Cellculture
Dicer
siRNAs
Degradation
of target
message
RNAi experiments in different organisms
Differential Expression vs Signaling Function
RIP/IMD
pathways
RNAi
phenotypes
Differentially regulated
genes
R
RIP
Tak1
~ 70
280 genes
IKK
Rel
Targets
Michael Boutros
Most pathway targets are not required for
pathway function
RIP/IMD
pathways
RNAi
phenotypes
Differentially regulated
genes
R
RIP
Tak1
280 genes
~ 70
IKK
Rel
3
Targets
Michael Boutros
What is a phenotype: it all depends on the assay
Any cellular process can be probed.
- (de-)activation of a signaling pathway
- cell differentiation
- changes in the cell cycle dynamics
- morphological changes
- activation of apoptosis
Similarly, for organisms (e.g. fly embryos, worms)
Phenotypes can be registered at various levels of detail
- yes/no alternative
- single quantitative variable
- tuple of quantitative variables
- image
- time course
Monitoring tools for automated phenotyping
Plate reader
96 or 384 well plates, 1…4 measurements per well
FACS; Acumen Explorer
ca. 2000 x 4…8 measurements
per well
Automated Microscopy
practically unlimited.
many MB
Plate reader Assays
Cells are seeded on-top of
pre-aliquoted siRNA pools
2x 68 384-well plates
+/- Compound
treatment (48h pt)
RNAi by reversetransfection
72h
Cell viability
(‘CellTiterGlo’ Assay)
Computational analysis
Secondary assays
cellHTS2
workflows for
analysing a cellbased assay
experiment
NChannelSet
“pheno”Data
(AnnotatedDataframe)
featureData
(AnnotatedDataframe)
assayData can contain N=0, 1, 2, ..., matrices of the same size
D
Physical coordinates
Sample-ID red
R
Sample-ID green
G
Sample-ID blue
B
Array-ID
_ALL_
Sequence
Target gene ID
varMetaData
The data
Numeric values xijk
i = wells (e.g. 20,000)
j = different reporters (e.g. 2)
k = different assays (e.g. 5)
Metadata about wells
pi = plate in which is well i
ri = row (within plate) of well i
ci = column (within plate) of well I
siRNA sequence, target gene, ....
Metadata about reporters
Fluc, Rluc, ...
Metadata about assays k
replicate number
different variants of the assay
date it was done
Between plate effects
kth well
ith plate
package
splots
"Normalisation"
xijk
i = wells (e.g. 20,000)
j = different reporters, dyes
k = different assays
xijk 
xijk
median  xhjk | ph = pi 
Plate median normalisation
- can use other estimators of location, e.g. mean, midpoint of shorth;
or shift and scale according to values of positive and negative
controls
- maintains the dimensions of x
Normalization: Plate effects
Percent of control
x
k
i
x = pos
1
0
0
μ
i
Normalized percent
inhibition
μ
x
k
i
x=

1
0
0
n
e
g
μ
μ
i
z-score
'
k
i
'
k
i
p
o
s
i
p
o
s
i
xki i
x =
σi
'
ki
k-th well
i-th plate
Spatial normalization
B-score:
two-way medianpolish
before
fitted row and column effects


ˆ
ˆ+
ˆ
x

μ
+
R
C
r
c
i
i
r
i
c
i
'
x=
r
c
i
M
A
D
i
rth row
cth column
ith plate
Malo et al., Nat. Biotech. 2006
after
How to estimate the normalization parameters?
From which data points:
• Based on the intensities of the controls
if they work uniformly well across all plates
• Based on the intensities of the samples
invoke assumptions such as "most genes have
no effect", or "same distribution of effect sizes"
Which estimator:
mean vs median vs shorth
standard deviation vs MAD vs IQR
No universally optimal answer, it depends on the data.
In the best case, it doesn't matter.
Estimators of location
100
120
Histogram of x
60
40
20
0
Frequency
80
mean
median
mean
shorth
half.range.mode
-2
0
2
4
x
6
8
10
Estimators of scale
1 n
2
s
ta
n
d
a
rdd
e
via
tio
n:
(
x

x
)

i
n1i1
m
e
d
ia
na
b
s
o
lu
ted
e
via
tio
n
:m
e
d
ia
n{|xi m
e
d
ia
n
{xj }|}
in
te
rq
u
a
rtilera
n
g
e
:Q(
xi })Q(
xi })
7
5{
2
5{
s
h
o
rth
Channel summarisation
xijk
i = wells (e.g. 20,000)
j = different reporters, dyes
k = different assays
xi2k
 xi1k ,xi2k  
xi1k
(log-)ratio
collapses the second dimension of x
Replicate Summarisation
xijk
i = wells (e.g. 20,000)
j = different reporters, dyes
k = different assays
x
ijk1
,...,xijk
n
y
ijr
replicate summarization
changes third dimension of x
Processing steps
xijk
i = wells (e.g. 20,000)
j = different reporters, dyes
k = different assays
normalisation of whole plate effects
normalisation of within plate effects
summarization of channels
replicate summarization
scoring (transformation into z-scores)
contrasts (as in linear models)
dim unchanged
dim unchanged
change 2nd dim
change 3rd dim
can change 1st dim
change 3rd dim
NchannelSet provides a robust and powerful infrastructure for these
operations (and keeping the metadata aligned and intact)

p 
n
Z
'
13

p 
n
Zhang JH, Chung TD, Oldenburg KR, "A Simple Statistical Parameter for
Use in Evaluation and Validation of High Throughput Screening Assays."
J Biomol Screen. 1999;4(2):67-73.
NB:
2p 2n would be more efficient
show example report
Thanks
Ligia Bras
Florian Hahne
Michael Boutros
Thomas Horn, Tina Büchling, Dorothee Nickles, Dierk Ingelfinger
Elin Axelsson
Gregoire Pau
Martin Morgan