In this communication, we address the problem of robust classification of proteomic serum samples. We propose coupling classification with the inverse problem methodology. The analytical chain comprising a liquid chromatograph and a mass spectrometer in Selected Reaction Monitoring mode is modelled, integrating an implicit hierarchy. We solve the inverse problem by the means of full-Bayesian statistics, resorting to stochastic sampling algorithms for the numerical computations. We compare our joint Inversion-Classification to state-of-the-art methods (Naïve Bayes, logistic regression, fuzzy c-means) using sequential estimations and show very encouraging results on simulated multi-class data.