Mitigating Local Over-fitting During Single Particle Reconstruction with SIDESPLITTER

Abstract

Single particle analysis of cryo-EM images enables macromolecular structure determination at resolutions approaching the atomic scale. Experimental images are extremely noisy, however, and during iterative refinement it is possible to stably incorporate noise into the reconstructed density. Such "over-fitting" can lead to misinterpretation of the structure, and thereby flawed biological results. Several strategies are routinely used to prevent the spurious incorporation of noise within reconstructed volumes, the most common being independent refinement of two sides of a split dataset. In this study, we show that over-fitting remains an issue within regions of low local signal-to-noise in reconstructed volumes refined using the half-set strategy. We propose a modified filtering process during refinement through the application of a local signal-to-noise filter, SIDESPLITTER, which we show to be capable of reducing over-fitting in both idealised and experimental settings, while maintaining independence between the two sides of a split refinement. SIDESPLITTER can also improve the final resolution in refinements of structures prone to severe over-fitting, such as membrane proteins in detergent micelles.