Biology Reference
In-Depth Information
of MPSS and microarray noise measurements is thus begun by con-
sidering the conversion of Affymetrix GeneChip intensities into units
of tpm.
The intensity of the fluorescence measured at a given probe of
a microarray (i.e., a region containing an oligonucleotide with a given
sequence) increases with the concentration of complementary mRNA
contained in the target sample [5]. However, the relationship between
target concentration and observed fluorescence intensity follows the
Langmuir isotherm function:
nc
nc
p
Ic
()
=
(6)
+
e
where
I
and
c
are the observed intensity and target concetration,
respectively,
n
p
is the saturation value of the intensity, and
n
e
is a probe-
dependent constant [18,19]. It follows from eq. (6) that the observed
intensity is a linear function of concentration at low concentrations
and saturates at high concentrations (
c
n
e
).
The Affymetrix MAS 5.0 software calculates the expression level
of a particular gene from a set of individual probes. The algorithm
followed [14,20] results in the predicted gene expression “signal levels.”
These signal levels exhibit a dependence on target concentration that
follows the same form as the intensity in eq. (6). This may be demon-
strated by computing the signal levels for a series of “spike-in” target
genes—genes that have been added to a solution of mRNA at known
concentration. Specifically, consider the signal levels calculated for a
group of 12 genes that have been spiked into hybridization mixtures
at known concentrations between 0 and 1024 pM and subsequently
hybridized to Affymetrix U95A GeneChips. This data was taken at
Affymetrix and is publicly available (
http://www.affymetrix.com/
support/datasets.affx).
In figure 4.14 is plotted the median expression
levels (as calculated using MAS 5.0) for all of the spike-in genes
reported at each known (i.e., “spike-in”) concentration as a function of
that known concentration. The dashed line is a fit of the data to eq. (6)
with logarithmic weighting (so as to give approximately equal weight
to low- and-high concentration data). It is clear that this functional form
describes the data well.
The comparison between Affymetrix and MPSS data is considerably
simplified if one considers only the data within the concentration range
for which both systems yield expression signal levels that scale linearly
with target concentration. In doing so, the need to nonlinearly rescale
q for all of the s
c
e
<<
>>
(
)
plots computed from the Affymetrix data is
avoided. Specifically, one may fit the data in figure 4.14 with spike-in
concentration values between 1 pM and 100 pM to a straight line
constrained to pass through the origin. This is shown as the solid line
(
q
)
Search WWH ::
Custom Search