Biology Reference
In-Depth Information
around the diagonal is a measure of noise. In both figure 4.8a and
4.8b, the
x
-axes are the log
10
-tpm value of the signatures in experi-
ment FS.A1.a (see figure 4.7), while the
y
-axes correspond, respectively,
to the log
10
-tpm of the signatures in experiments FS.A1.b (figure 4.8a)
and FS.B.a (figure 4.8b). The spread due to the combined variance intro-
duced by library creation, bead loading, and sequencing (figure 4.8b)
is much larger than that due to sequencing alone (figure 4.8a).
Adopting the definitions for dq,
-
, and
P
k
(dq
q) from the microarray
analysis above, one may compute the function s
|
(
q
0
)
using eq. (1). Plots
of s
derived from the data of figures 4.8a and 4.8b show that
s decreases with the expression intensity q
(
q
0
)
, with the overall s (filled
circles, figure 4.8c) being about twice as large as the s arising from the
bead loading and sequencing (plus signs, figure 4.8c). Note that the
noise from the combination of bead loading and sequencing is almost
indistinguishable from that of sequencing alone (diamonds, figure 4.8c),
demonstrating that noise stemming from bead loading is negligible.
0
THE STATISTICS OF THE ZERO COUNTS
Thus far, the analysis presented has dealt only with signatures whose
bead counts are at least unity in each of the replicate experiments
under consideration. However, one also observes numerous signatures
with a finite bead count for one replicate experiment and zero for
the other. This situation is unique to technologies that generate a truly
digital measure of transcript expression (MPSS as opposed to, say,
DNA microarrays). Signatures with a finite bead count for one repli-
cate experiment and zero for the other appear in figures 4.8a and 4.8b
as the sets of points forming linear structures at the left and bottom of
these figures. (As these plots are in log-log scale, the value of zero
counts, i.e. zero tpm, has been arbitrarily given a log-tpm value of 0.)
It is clear from these figures that the statistics of the signatures with
low but positive counts in both runs are significantly different from
the statistics of the signatures measured as zero in one of the replicates.
Similar result has been observed in other data sets, for example, in
MPSS experiments on
A. thaliana
[17].
To investigate the significance of zero count measurements in an
MPSS data set, expression data taken on macrophages 8 hours after
LPS stimulation is studied. (This data was chosen because four TS
and four FS MPSS runs were taken on this sample.) First, the signa-
tures with exactly four, three, and two nonzero bead counts within
the four replicates are identified (see ref. [13] for the method of deter-
mining whether the TS or FS data was used for a given signature).
Next, one computes the function s
two of the
nonzero replicate measurements are chosen at random when dealing
with signatures with more than two nonzero values) separately for
the data sets in which zero, one, or two out of the four sequencing
(
q
0
)
(in computing s
(
q
0
),
Search WWH ::
Custom Search