Higher Resolution with the Same Data - Tuning cryo-EM Refinement in Relion

Abstract

Structure refinement used to be limited by crystallization or computational time. We have eliminated the computational burden of cryo-em by accelerating data processing through development of a new, GPU-enabled version of RELION, in collaboration with its ongoing development. The iterative refinement of bio-molecular structures from cryo-em image data, is therefore faster than ever before. However no systematic analysis has been published on the optimal use of input data in the final stages of high-resolution reconstruction. As a consequence, selective exclusion of available image data (at the discretion of the experimenter) represents the currently most efficient (and often used) way to increase the attained resolution. We show that a higher level of reconstructed detail can be achieved by incorporating new methods into existing cryo-em refinement algorithms, which aim to increase the density of high-resolution data used in the final reconstruction steps. Using a combination of tunable parameters and guided data exclusion, we show that these methods can both increase resolution, as well as reduce sensitivity on the input data. These measures facilitate an improvement to the cryo-em refinement method by a) making reconstructed detail more robust to the used data, and b) rendering the process less subjective. Since cryo-em is rapidly becoming a common tool in structural biology, this represents a necessary development for an enormously influential method over the coming years.