The model presented above seems biological sensible, but how on earth are we going to parameterise it? Are we honestly going to let a user try to juggle the forty odd parameters inherent to this model? Clearly not. The approach we have taken to this is to provide set statistics derived from a maximum likelhood approach from known genes - this requires virtually no training - and then give switches to the user to turn on and off a variety of different parts of the algorithm.
The model is parameterised as probabilities, but actually calculated in log space. If you look in the code you would find that there is alot of switching between the two spaces: these are provided by the functions Probability2Score and Score2Probability (notice that the 'Score' here is very specific to the Wise2 package - you can't put any old score into Score2Probability to get a probability out as it depends on how that Score was converted into Log space).