Particle Picking and Image Restoration techniques in Cryo-EM

Abstract

With recent advances in Cryo-EM, it has become possible to achieve reconstructions of individual molecules at near atomic levels of resolution (< 4 A). One major roadblock for complete adoption of this technique in lab settings is the lack of automation of the Particle Picking part of the pipeline. This paper serves an introduction to the problem of Particle picking and associated Image restoration techniques. Three major types of particle picking methods are investigated here - Template matching based techniques, Machine Learning based techniques and Edge detection based algorithms, with a brief inspection of their pros and cons and why complete automation has been hard to achieve.

In the second part, we explore a new image restoration technique proposed by Tejal Bhamre and Amit Singer to denoise particles using Covariance estimation. We finally conclude the paper by presenting the experimental results of using the said restoration technique to detect outliers. We conclude that although this is very effective at low SNRs, further refinements need to done for this to be useful in a practical scenario.