80
When selecting 219 pairs, our model and selecting criterion were able to choose
the pairs with high increasing counts as well as the ones with oscillating and/or small
counts over the three stages but exhibiting large counts on the second and/or the
third stages in relation to the previous stage count. These pairs are candidates for a
deeper study of their binding behavior.
In consulting with the biology investigators, they would expect a small count if
there were no tripeptide enrichment of the library from the previous stage. In agree-
ment with its interpretation given above, we initially assume that the parameter μi is
small by choosing a prior distribution with small mean and variance. It is necessary
to assume a prior small variance since we have just one three-count observation for
every tripeptide-tissue pair. The other two parameters βi and δi can be assumed to
have diffuse priors.
Our model is not considering that different tissues can have different binding be-
haviors. For example, there may be a tissue that absorbs more tripeptides, or present
different count variances. Although this problem is ameliorated considering the base-
line mean count parameter μi, the model could be improved by priori incorporating
a correlation structure of the pairs sharing the same tissue.