Robust ab initio solution of the cryo-EM reconstruction problem at low resolution with small data sets

Aaditya V. Rangan,Leslie Greengard

2023-07-18

Abstract:Single particle cryo-electron microscopy has become a critical tool in structural biology over the last decade, able to achieve atomic scale resolution in three dimensional models from hundreds of thousands of (noisy) two-dimensional projection views of particles frozen at unknown orientations. This is accomplished by using a suite of software tools to (i) identify particles in large micrographs, (ii) obtain low-resolution reconstructions, (iii) refine those low-resolution structures, and (iv) finally match the obtained electron scattering density to the constituent atoms that make up the macromolecule or macromolecular complex of interest. Here, we focus on the second stage of the reconstruction pipeline: obtaining a low resolution model from picked particle images. Our goal is to create an algorithm that is capable of ab initio reconstruction from small data sets (on the order of a few thousand selected particles). More precisely, we propose an algorithm that is robust, automatic, and fast enough that it can potentially be used to assist in the assessment of particle quality as the data is being generated during the microscopy experiment.

Numerical Analysis,Quantitative Methods

What problem does this paper attempt to address?

The paper is primarily dedicated to addressing the reconstruction problem at low resolution in single-particle cryo-electron microscopy (cryo-EM), particularly achieving this goal with smaller datasets. Specifically, the research focuses on the second stage of obtaining a low-resolution model from selected particle images, which involves initial model construction without knowing the particle orientations.

The method proposed in the paper aims to create an algorithm capable of robustly performing ab initio reconstruction on small datasets (a few thousand selected particles). Additionally, the algorithm needs to be automated and fast to evaluate particle quality during the experiment. To achieve these goals, the authors propose an improved Alternating Minimization (AM) method called "Alternating Minimization with Entropy Maximization and Principal Modes" (EMPM). This method includes two main improvements:

1. **Maximum Entropy Alignment**: In addition to the traditional maximum likelihood alignment strategy, a "maximum entropy" alignment strategy is also employed. This means finding the experimental image that best matches the projection corresponding to the current structure for each uniformly chosen observation angle. Although this may sound counterintuitive, this approach helps stabilize the reconstruction process based on a small number of images at low resolution.

2. **Principal Mode Projection**: Each image is transformed onto a polar grid in the Fourier space. In the Fourier transform domain, all data points with the same modulus form a Fourier ring. Instead of considering all Fourier rings, only some carefully selected combinations, known as "principal modes," are chosen. This can significantly compress the data, improve computational efficiency, and retain useful information for alignment.

With these improvements, the EMPM method not only achieves faster and more accurate results but is also more reliable than many existing ab initio reconstruction algorithms. Experimental results show that the method performs well on several published datasets, with reconstruction results often comparable to, or even better than, the best results expected from the "oracle-guided" AM method (i.e., the standard AM method using true observation angles and displacements as initial values). Additionally, the method runs approximately 20 times faster than the standard Bayesian inference or the widely used open-source package Relion for ab initio reconstruction.

In summary, the goal of this paper is to develop a new algorithm to overcome the limitations of existing single-particle cryo-EM reconstruction workflows, particularly when the available data is limited, enabling more efficient and accurate low-resolution model acquisition.