It is developed in the group of Sjors Scheres at the MRC Laboratory of Molecular Biology. Briefly, the ill-posed problem of 3D-reconstruction is regularised by incorporating prior knowledge: the fact that macromolecular structures are smooth, i.e. they have limited power in the Fourier domain. In the corresponding Bayesian framework, many parameters of a statistical model are learned from the data, which leads to objective and high-quality results without the need for user expertise.