The model bias: a pitfalls of single particle of cryo-electron microscopy

Abstract

In structural biology, single particle of cryo-electron microscopy (cryo-EM) has been a popular method of determining biological structures. In particular, computation analysis plays an important role in solving structure from a highly noisy cryo-EM image set. Henderson (2013) pointed out that the computer-based image process needs a careful check to ensure the valid of the progress. Otherwise, it can cause an over-fitting problem of single particle cryo-EM in which the solved structure from a highly noise cryo-EM image set can be strongly biased toward the reference model. It is referred as the model bias. An interesting example of the model bias is ``Einstein from noise'', that is, the face of Einstein can be reconstructed from pure noise. This surprising example prompted our interest in investigating the phenomenon from a theoretical point of view. In this talk, we will present a series of simulation results in order to illustrate this phenomenon. Further, we will also give clear mathematical explanations of these results.