Reference: Youngs N, et al. (2013) Parametric Bayesian priors and better choice of negative examples improve protein function prediction. Bioinformatics 29(9):1190-8