On the parameters affecting dual-target-function evaluation of single-particle selection from cryo-electron micrographs

TL;DR

This study systematically investigates how signal-to-noise ratio and reference choices affect the dual-target-function approach in cryo-EM particle selection, demonstrating conditions for optimal performance and bias suppression.

Contribution

It provides a quantitative analysis of the parameters influencing dual-target-function evaluation in cryo-EM, including the critical SNR threshold and effective reference models.

Findings

Identifies the critical SNR threshold where DTF performance declines.

Shows Gaussian models are as effective as single projections for particle picking.

Using Gaussian models reduces reference bias in particle selection.

Abstract

In the analysis of frozen hydrated biomolecules by single-particle cryo-electron microscopy, template-based particle picking by a target function called fast local correlation (FLC) allows a large number of particle images to be automatically picked from micrographs. A second, independent target function based on maximum likelihood (ML) can be used to align the images and verify the presence of signal in the picked particles. Although the paradigm of this dual-target-function (DTF) evaluation of single-particle selection has been practiced in recent years, it remains unclear how the performance of this DTF approach is affected by the signal-to-noise ratio of the images and by the choice of references for FLC and ML. Here we examine this problem through a systematic study of simulated data, followed by experimental substantiation. We quantitatively pinpoint the critical signal-to-noise ratio (SNR), at which the DTF approach starts losing its ability to select and verify particles from cryo-EM micrographs. A Gaussian model is shown to be as effective in picking particles as a single projection view of the imaged molecule in the tested cases. For both simulated micrographs and real cryo-EM data of the 173-kDa glucose isomerase complex, we found that the use of a Gaussian model to initialize the target functions suppressed the detrimental effect of reference bias in template-based particle selection. Given a sufficient signal-to-noise ratio in the images and the appropriate choice of references, the DTF approach can expedite the automated assembly of single-particle data sets.