Thresholding of cryo-EM density maps by false discovery rate control.

Maximilian Beckers,Carsten Sachse,Arjen J Jakobi

doi:10.1107/s2052252518014434

Abstract

Cryo-EM now commonly generates close-to-atomic resolution as well as intermediate resolution maps from macromolecules observed in isolation and in situ. Interpreting these maps remains a challenging task owing to poor signal in the highest resolution shells and the necessity to select a threshold for density analysis. In order to facilitate this process, a statistical framework for the generation of confidence maps by multiple hypothesis testing and false discovery rate (FDR) control has been developed. In this way, three-dimensional confidence maps contain signal separated from background noise in the form of local detection rates of EM density values. It is demonstrated that confidence maps and FDR-based thresholding can be used for the interpretation of near-atomic resolution single-particle structures as well as lower resolution maps determined by subtomogram averaging. Confidence maps represent a conservative way of interpreting molecular structures owing to minimized noise. At the same time they provide a detection error with respect to background noise, which is associated with the density and is particularly beneficial for the interpretation of weaker cryo-EM densities in cases of conformational flexibility and lower occupancy of bound molecules and ions in the structure.

Highlights

Cryo-EM-based structure determination has undergone remarkable technological advances over the past few years, leading to a sudden multiplication of near-atomic resolution structures (Patwardhan, 2017)
It is demonstrated that confidence maps and false discovery rate (FDR)-based thresholding can be used for the interpretation of near-atomic resolution single-particle structures as well as lower resolution maps determined by subtomogram averaging
We generated a test image with an signalto-noise ratio (SNR) of 1.2 and noted that signal from high-resolution features cannot be detected in the power spectrum computed from the simulated noise images, it is present in the noise-free power spectrum (Fig. 1c, right)

Summary

Introduction

Cryo-EM-based structure determination has undergone remarkable technological advances over the past few years, leading to a sudden multiplication of near-atomic resolution structures (Patwardhan, 2017). The highest resolution structures have become available at $2 Aresolution (Merk et al, 2016; Bartesaghi et al, 2018, 2015) and sub-4 Aresolution structures of molecules below 100 kDa have been resolved from images obtained with and without an optical phase plate (Merk et al, 2016; Khoshouei et al, 2017) These studies established technical routines for the determination of atomic models of structures that it was previously thought to be impossible to resolve by cryo-EM or any other technique (Bai et al, 2015; Galej et al, 2016; Fitzpatrick et al, 2017; Gremer et al, 2017).

Objectives

Methods

Results

Conclusion