|
Due to the many stages of cDNA microarray experiments, it is not surprising that
any single microarray output is subject to substantial variability. Although some
of these errors can be controlled by well-designed protocols and careful operations,
they can by no means be removed completely. Because of that, a replicated experimental design is a
crucial way to assess systematic errors and improve reliability of the results.
One way to represent microarray experimental design is to use directed graphs,
with multiple edges. In this representation, vertices or nodes (for example, A and B)
correspond to target mRNA samples, and edges or arrows correspond to hybridizations between two mRNA samples.
By convention,we place the green-labelled sample at the tail and the red-labelled sample at the
head of the arrow.
Designing a cDNA microarray experiment consists of answering the following question:
Which pairs of samples should be co-hybridized?
There are several widely used principles that can answer this question, such as
common reference design, common reference pool design and circuit design.
Given 8 slides for use in comparisons, panel a at left shows the layout of a common reference
design. In common reference design, all samples are compared to a single reference sample.
That sample may be the sample that is most familiar (e.g. wildtype or at log-phase in a growth
experiment) or may be a pool of cDNA from all samples. Note that using a single reference has
an advantage in terms of easy immediate interpretation of data: all fold-changes are compared to the
reference. However, it has two disadvantages. First, any genes expressed at a low level in the reference
will tend to be poorly measured by this method, because there will be a high variance of the ratios obtained.
Second, and more imporantly, comparisons between non-reference samples are all transitive and will always
inherit the sum of variances from two measures. Direct comparisons are more precise in that sense.
Panel b shows the layout of a circuit design. In a circuit design, samples are compared
to each other in a loop or all-pairwise design. This design has been shown to be he most powerful for
ascertaining differential expression when all samples are of equivalent interest.
Four samples are studied in each design, and each co-hybridization is replicated
twice, with dyes swapped.
There may be technical
replication and biological replication in microarray experiment. The former refers to
replicating the co-hybridization of cDNA from the same reverse transcripted mRNA or biological growth,
while the latter refers to replicating
the co-hybridization of cDNA from wholly independent preparations. Although the technical replication
can help reveal sources of variation in your microarray protocol, it does not provide as much
inference power about the biological question of interest as biological replication does.
Some further general recommendations when designing a microarray experiment are listed below:
Direct comparisons always give more precise results than indirect comparison;
Avoid co-hybridizing two vastly different samples
Keep the whole design dye balanced
|
