Biomarker detection and prognostic classification are common steps in the analysis of proteomics mass spectrometry data. However, many existing classifiers do not incorporate the spectral nature of the data properly which may result in poor classification performance. In this talk, I will describe a newly developed Gaussian process discriminant analysis that is suitable for classifying mass spectrometry data. The proposed model incorporates feature selection and classification within a unified framework. The spectral nature of the data is accounted for with an appropriate covariance function. The computational efficiency of the model is kept within a reasonable range through variational Bayes and computational shortcuts. I will conclude the talk with numerical results based on simulated and real datasets.