73
terion chooses pairs with an oscillating count pattern but with a substantially large
increment from the first to the second stages or from the second to the third stages.
Besides, it picks pairs that have small counts over the three stages but with a large
increment in the second or∕and third stage count in comparison to the previous count.
On the contrary, there are some other pairs that despite having relatively large and
nondecreasing counts over the three stages are not selected; the selecting method is
detecting that this event happens because the corresponding tripeptide has a large
base-line count to start with and not necessarily a strongly binding behavior to the
respective tissue. The expected values of the FDR and FNR are 0.534 and 0.089
respectively. Due to space limitation these pairs are not listed in this manuscript.
In Figure 3.7, we plotted the histograms of the simulated samples of the posterior
distributions of βi and <5i for three specific pairs. We notice that the bimodal behavior
of the first two pairs considered in this plot cannot be detected by the parametric
version of our model. The third pair is an example in which there is no statistical
evidence of an increasing pattern (since βi < 1). Nevertheless, the decision rule (3.35)
tends to pick this pair due to its large count observed at the third stage.