Let me try again...though I don't know if I'll make it any better (my biology isn't so good...I'll probably up the explanation)...
Proteins are created in the body due to the interaction of genes in the presence of environmental factors. Certain genes directly control the (eventual) creation of certain proteins. Some genes act as controllers for other genes...turning them on/up, or off/down.
Even though the majority of the human genome has been decoded, that doesn't mean we know what they all do or how they interact with each other.
Experiments can be conducted to measure what's known as "gene expression" due to certain environmental factors. Even then, all the measurement shows is a series of levels for certain substances. It doesn't show how they got there...which gene expressed/controlled another. Also, the measurements are not perfect. In fact, they're quite noisy. Further, even if we knew piece-wise which gene did what, that doesn't mean we could deterministically figure out a particular sequence of input to output...there's just too many combinations. So we try to figure things out probabilistically.
Bayesian theory is a way to use prior estimates and data with current estimates and data to produce (hopefully) better estimates. In this case, the end product of the analysis would be estimates of the network structure of a particular gene interaction (in other words...given a certain input, an estimate of which genes, in which order, at what magnitude, interacted to produce that particular output).
In my case, my first task is to find a way to combine several sets of dissimilar data into a model which would (hopefully) predict network paths better than current models do.