MicroArray Image Analysis

Microarray analysis

 Array construction, hybridisation,

scanning
 Quantitation of fluorescence signals
 Data visualisation
 Meta-analysis (clustering)
 More visualisation
Technical

sample pseudo-colour
(labelled) image

probe
(on chip)
Experimental design
 Track what’s on the chip
 which spot corresponds to which gene
 Duplicate experimental spots
 reproducibility
 Controls
 DNAs spotted on glass
 positive probe (induced or repressed)
 negative probe (bacterial genes on human chip)
 oligos on glass or synthesised on chip (Affymetrix)
 point mutants (hybridisation plus/minus)
Images from scanner
 Resolution
 standard 10m [currently, max 5m]
 100m spot on chip = 10 pixels in diameter

 Image format
 TIFF (tagged image file format) 16 bit (65’536 levels of grey)
 1cm x 1cm image at 16 bit = 2Mb (uncompressed)
 other formats exist e.g.. SCN (used at Stanford University)

 Separate image for each fluorescent sample

 channel 1, channel 2, etc.
Images in analysis software
 The two 16-bit images (Cy3, Cy5) are compressed into
8-bit images

 Display fluorescence intensities for both wavelengths

using a 24-bit RGB overlay image

 RGB image :
 Blue values (B) are set to 0

 Red values (R) are used for Cy5 intensities

 Green values (G) are used for Cy3 intensities

 Qualitative representation of results

 Images : examples
Pseudo-colour overlay

Cy3

Spot colour Signal strength Gene expression

Cy5
yellow Control = perturbed unchanged
red Control < perturbed induced
green Control > perturbed repressed
Processing of images
 Addressing or gridding
 Assigning coordinates to each of the spots

 Segmentation
 Classification of pixels either as foreground or
as background

 Intensity determination for each spot

 Foreground fluorescence intensity pairs (R, G)
 Background intensities
 Quality measures
 Addressing (I)
 The basic structure of the
images is known (determined by
the arrayer)
 Parameters to address the spots
positions
 Separation between rows and
columns of grids
 Individual translation of grids
 Separation between rows and
columns of spots within each grid
 Small individual translation of spots
 Overall position of the array in the
image
ScanAlyze
Addressing (II)
 The measurement process depends on the
addressing procedure

 Addressing efficiency can be enhanced by

allowing user intervention (slow!)

 Most software systems now provide for

both manual and automatic gridding
procedures
Segmentation (I)

 Classification of pixels as foreground

or background
-> fluorescence intensities are
calculated for each spot as measure
of transcript abundance

 Production of a spot mask : set of

foreground pixels for each spot
 Segmentation (II)
 Segmentation methods :
 Fixed circle segmentation
 Adaptive circle segmentation
 Adaptive shape segmentation
 Histogram segmentation

Fixed circle ScanAlyze, GenePix, QuantArray

Adaptive circle GenePix, Dapple
Adaptive shape Spot, region growing and watershed
Histogram method ImaGene, QuantArray, DeArray and adaptive thresholding
Fixed circle segmentation
 Fits a circle with a constant diameter
to all spots in the image
 Easy to implement
 The spots need to be of the same
shape and size

May not be good

for this example !
 Adaptive circle segmentation
 The circle diameter is
estimated separately
for each spot
 Dapple finds spots by
detecting edges of
spots (second
derivative)

 Problematic if spot
exhibits oval shapes
Adaptive shape segmentation
 Specification of starting points or seeds
 Bonus: already know geometry of array!
 Regions grow outwards from the seed points
preferentially according to the difference
between a pixel’s value and the running mean of
values in an adjoining region.
 Histogram segmentation
 Uses a target mask chosen to be
larger than any other spot

 Foreground and background intensity

are determined from the histogram
of pixel values for pixels within the
masked area

 Example : QuantArray
 Background : mean between 5th and
20th percentile
 Foreground : mean between 80th and
95th percentile

 May not work well when a large

target mask is set to compensate for
variation in spot size Bkgd Foreground
Spot ‘foreground’ intensity
 The total amount of hybridization for a
spot is proportional to the total
fluorescence generated by the spot
 Spot intensity = sum of pixel intensities
within the spot mask
 Since later calculations are based on ratios
between Cy5 and Cy3, we compute the
average* pixel value over the spot mask
*alternative : use ratios of medians instead of
means – may be better if bright specks present
Background intensity
 Spot’s measured intensity includes a contribution
of non-specific hybridization and other chemicals
on the glass

 Fluorescence from regions not occupied by DNA

should by different from regions occupied by DNA
-> one solution is to use local negative controls
(spotted DNA that should not hybridize)

 Different background methods :

 Local background
 Morphological opening
 Constant background
 No adjustment
Local background
 Focusing on small regions surrounding the spot mask.
 Median of pixel values in this region
 Most software package implement such an approach

ScanAlyze ImaGene Spot, GenePix

 By not considering the pixels immediately surrounding

the spots, the background estimate is less sensitive to
the performance of the segmentation procedure
 Morphological opening
 Non-linear filtering, used in Spot
 Use a square structuring element with
side length at least twice as large as
the spot separation distance
 Compute local minimum filter, then
compute local maximum filter
 This removes all the spots and
generates an image that is an estimate
of the background for the entire slide
 For individual spots, the background is
estimated by sampling this background
image at the nominal center of the spot
 Lower background estimate and less
variable
Constant background
 Global method which subtracts a constant
background for all spots

 Some evidence that the binding of fluorescent

dyes to ‘negative control spots’ is lower than the
binding to the glass slide

 -> More meaningful to estimate background based

on a set of negative control spots
 If no negative control spots :
approximation of the average background =
third percentile of all the spot foreground values
No background adjustment

 Do not consider the background

 Probably not accurate, but may be

better than some forms of local
background determination!
Quality control (-> Flag)
 How good are foreground and background
measurements ?
 Variability measures in pixel values within each
spot mask
 Spot size
 Circularity measure
 Relative signal to background intensity
 Dapple:
 b-value : fraction of background intensities less than
the median foreground intensity
 p-score : extend to which the position of a spot
deviates from a rigid rectangular grid
 Flag spots based on these criteria
 Summary M = log2 R/G
A = log2 √(R•G)

 The choice of background

correction method has a larger
impact on the log-intensity ratios
than the segmentation method
used

 The morphological opening method

provides a better estimate of
background than other methods
 Low within- and between-slide
variability of the log2 R/G
Spot, GenePix
 Background adjustment has a
larger impact on low intensity
spots

ScanAlyze
 Selected references

 Yang, Y. H., Buckley, M. J., Dudoit, S. and

Speed, T. P. (2001), ‘Comparisons of methods
for image analysis on cDNA microarray data’.
Technical report #584, Department of
Statistics, University of California, Berkeley.
[Link]

 Yang, Y. H., Buckley, M. J. and Speed, T. P.

(2001), ‘Analysis of cDNA microarray images’.
Briefings in bioinformatics, 2 (4), 341-349.
Excellent review in concise format!