Abstract

We use three-dimensional electron microscopy to visualise molecular machines, which are protein and/or RNA complexes that fulfill vital processes in any living organism. Despite their small size molecular machines often function in ways that are strikingly similar to machines from daily life, forming the nano-scale equivalents of water mills, fuel-driven motors or even walking legs. In the electron microscope two-dimensional projection images of individual complexes are recorded. Multiple images from complexes in different orientations are then combined to obtain a three-dimensional reconstruction. However, the recorded images are extremely noisy (with typical signal-to-noise ratios below 0.1) and their relative orientations are unknown. Consequently, 3D-reconstruction of electron microscopy images is a severely ill-posed problem with incomplete data. I will discuss statistical image processing approaches that address this problem, in particular a maximum-likelihood approach that marginalises over the unknown orientations [1], and a maximum-a-posteriori approach that prevents overfitting through the use of a regularised likelihood function (in preparation).

[1] F.J. Sigworth, P. Doerschuk, J.M. Carazo, S.H.W. Scheres (2010) “An introduction to maximum-likelihood methods in cryo-EM”, Methods in Enzymology, 482, 263-294. http://dx.doi.org/10.1016/S0076-6879(10)82011-7