Hotel Estoril Eden, Monte Estoril, 



Terry Speed, Frédéric Schütz Over the last couple of years interest in what one might call fully specified probabilistic models for tandem mass spectometry (MS/MS) data has increased greatly. It is safe to say that a driving force in this development has been the desire of many people to associate valid "pvalues" or "confidences" with peptide identifications. In an ideal world, these models would incorporate all the information available about a particular spectrum, including not only the direct measurements of the mass spectrometer (mass of the precursor and lists of peaks and intensities of the product ion spectrum), but also any other experimental information such as the instrument parameters (collision energy, etc), the separation methods that were used (LC retention time), etc. In the real world, it is not possible to take into account all those parameters and simplifying assumptions must be made, such as assuming the independence of many of the variables. This problem is especially acute in the context of database searching, where such a model would be used to calculate probabilities for all the peptides entries in the database in the shortest possible time. Several different research groups have tried to solve this problem and define good models and in this process they have followed different path and used different assumptions. We will present the approaches taken by those methods and compare the resulting models
