Oliver Serang: A graphical, nonparametric Bayesian approach for the robust decomposition of mixtures

A novel, non-parametric Bayesian alternative to the Kolmogorov-Smirnov test and the Kullback-Leibler divergence is proposed and used to decompose mixtures in the presence of arbitrary graphical dependencies. This method can be applied to estimate hyperparameters commonly thought to be inestimable. This is illustrated by using the method to estimate an ideal false discovery rate (FDR) threshold for mass spectrometry-based protein identification.