Automatic particle selection: results of a comparative study

Manual selection of single particles in images acquired using cryo-electron microscopy (cryoEM) will become a significant bottleneck when datasets of a hundred thousand or even a million particles are required for structure determination at near atomic resolution. Algorithm development of fully automated particle selection is thus an important research objective in the cryoEM field. A number of research groups are making promising new advances in this area. Evaluation of algorithms using a standard set of cryoEM images is an essential aspect of this algorithm development. With this goal in mind, a particle selection "bakeoff" was included in the program of the Multidisciplinary Workshop on Automatic Particle Selection for cryoEM. Twelve groups participated by submitting the results of testing their own algorithms on a common dataset. The dataset consisted of 82 defocus pairs of high-magnification micrographs, containing keyhole limpet hemocyanin particles, acquired using cryoEM. The results of the bakeoff are presented in this paper along with a summary of the discussion from the workshop. It was agreed that establishing benchmark particles and using bakeoffs to evaluate algorithms are useful in promoting algorithm development for fully automated particle selection, and that the infrastructure set up to support the bakeoff should be maintained and extended to include larger and more varied datasets, and more criteria for future evaluations.     

Application of template matching technique to particle detection in electron micrographs

Template matching together with the comprehensive theory of image formation in electron microscope provides an optimal (in Bayesian sense) tool for solving one of the outstanding problems in single particle analysis, i.e., automatic selection of particle views from noisy micrograph fields. The method is based on the assumption that the reference three-dimensional structure is known and that the relevant parameters of the model of the image formation process can be estimated. In the first stage of the procedure, a set of possible particle views is generated using the available reference structure. The template images are constructed as linear combinations of available particle views using a clustering technique. Next, the micrograph noise characteristic is established using an automated contrast transfer function (CTF) estimation procedure. Finally, the CTF parameters calculated are used to construct a matched filter and correlation functions corresponding to the available template images are calculated. In order to alleviate the problem of the biased caused by varying image formation conditions, a decision making strategy based on the predicted distribution of correlation coefficients is proposed. It is demonstrated that due to the inclusion of CTF considerations, the template matching method performed very well in a broad range of microscopy conditions.     

Step-by-step guide to efficient subtomogram averaging of virus-like particles with Dynamo

Subtomogram averaging (STA) is a powerful image processing technique in electron tomography used to determine the 3D structure of macromolecular complexes in their native environments. It is a fast growing technique with increasing importance in structural biology. The computational aspect of STA is very complex and depends on a large number of variables. We noticed a lack of detailed guides for STA processing. Also, current publications in this field often lack a documentation that is practical enough to reproduce the results with reasonable effort, which is necessary for the scientific community to grow. We therefore provide a complete, detailed, and fully reproducible processing protocol that covers all aspects of particle picking and particle alignment in STA. The command line–based workflow is fully based on the popular Dynamo software for STA. Within this workflow, we also demonstrate how large parts of the processing pipeline can be streamlined and automatized for increased throughput. This protocol is aimed at users on all levels. It can be used for training purposes, or it can serve as basis to design user-specific projects by taking advantage of the flexibility of Dynamo by modifying and expanding the given pipeline. The protocol is successfully validated using the Electron Microscopy Public Image Archive (EMPIAR) database entry 10164 from immature HIV-1 virus-like particles (VLPs) that describe a geometry often seen in electron tomography.

This study presents a complete and detailed step-by-step guide for subtomogram averaging using Dynamo software, with a special focus on particle picking and particle averaging; this will enable efficient processing for all experience levels, and lays a foundation for user-specific projects.     

Historical background: Why is it important to improve automated particle selection methods?

A current trend in single-particle electron microscopy is to compute three-dimensional reconstructions with ever-increasing numbers of particles in the data sets. Since manual--or even semi-automated--selection of particles represents a major bottleneck when the data set exceeds several thousand particles, there is growing interest in developing automatic methods for selecting images of individual particles. Except in special cases, however, it has proven difficult to achieve the degree of efficiency and reliability that would make fully automated particle selection a useful tool. The simplest methods such as cross correlation (i.e., matched filtering) do not perform well enough to be used for fully automated particle selection. Geometric properties (area, perimeter-to-area ratio, etc.) and the integrated "mass" of candidate particles are additional factors that could improve automated particle selection if suitable methods of contouring particles could be developed. Another suggestion is that data be always collected as pairs of images, the first taken at low defocus (to capture information at the highest possible resolution) and the second at very high defocus (to improve the visibility of the particle). Finally, it is emphasized that well-annotated, open-access data sets need to be established in order to encourage the further development and validation of methods for automated particle selection.     

A strategy for determining the orientations of refractory particles for reconstruction from cryo-electron micrographs with particular reference to round, smooth-surfaced, icosahedral viruses

Cryo-electron microscopy and three-dimensional image reconstruction are powerful tools for analyzing icosahedral virus capsids at resolutions that now extend below 1 nm. However, the validity of such density maps depends critically on correct identification of the viewing geometry of each particle in the data set. In some cases-for example, round capsids with low surface relief-it is difficult to identify orientations by conventional application of the two most widely used approaches-"common lines" and model-based iterative refinement. We describe here a strategy for determining the orientations of such refractory specimens. The key step is to determine reliable orientations for a base set of particles. For each particle, a list of candidate orientations is generated by common lines: correct orientations are then identified by computing a single-particle reconstruction for each candidate and then systematically matching their reprojections with the original images by visual criteria and cross-correlation analysis. This base set yields a first-generation reconstruction that is fed into the model-based procedure. This strategy has led to the structural determination of two viruses that, in our hands, resisted solution by other means.     

Improved RNA secondary structure prediction by maximizing expected pair accuracy

Free energy minimization has been the most popular method for RNA secondary structure prediction for decades. It is based on a set of empirical free energy change parameters derived from experiments using a nearest-neighbor model. In this study, a program, MaxExpect, that predicts RNA secondary structure by maximizing the expected base-pair accuracy, is reported. This approach was first pioneered in the program CONTRAfold, using pair probabilities predicted with a statistical learning method. Here, a partition function calculation that utilizes the free energy change nearest-neighbor parameters is used to predict base-pair probabilities as well as probabilities of nucleotides being single-stranded. MaxExpect predicts both the optimal structure (having highest expected pair accuracy) and suboptimal structures to serve as alternative hypotheses for the structure. Tested on a large database of different types of RNA, the maximum expected accuracy structures are, on average, of higher accuracy than minimum free energy structures. Accuracy is measured by sensitivity, the percentage of known base pairs correctly predicted, and positive predictive value (PPV), the percentage of predicted pairs that are in the known structure. By favoring double-strandedness or single-strandedness, a higher sensitivity or PPV of prediction can be favored, respectively. Using MaxExpect, the average PPV of optimal structure is improved from 66% to 68% at the same sensitivity level (73%) compared with free energy minimization.     

Robust filtering and particle picking in micrograph images towards 3D reconstruction of purified proteins with cryo-electron microscopy

In order to make a high resolution model of macromolecular structures from cryo-electron microscope (cryo-EM) raw images one has to be precise at every processing step from particle picking to 3D image reconstruction. In this paper we propose a collection of novel methods for filtering cryo-EM images and for automatic picking of particles. These methods have been developed for two cases: (1) when particles can be identified and (2) when particle are not distinguishable. The advantages of these methods are demonstrated in standard purified protein samples and to generalize them we do not use any ad hoc presumption of the geometry of the particle projections. We have also suggested a filtering method to increase the signal-to-noise (S/N) ratio which has proved to be useful for other levels of reconstruction, i.e., finding orientations and 3D model reconstruction.     

Auto-accumulation method using simulated annealing enables fully automatic particle pickup completely free from a matching template or learning data

Single-particle analysis is a 3-D structure determining method using electron microscopy (EM). In this method, a large number of projections is required to create 3-D reconstruction. In order to enable completely automatic pickup without a matching template or a training data set, we established a brand-new method in which the frames to pickup particles are randomly shifted and rotated over the electron micrograph and, using the total average image of the framed images as an index, each frame reaches a particle. In this process, shifts are selected to increase the contrast of the average. By iterated shifts and further selection of the shifts, the frames are induced to shift so as to surround particles. In this algorithm, hundreds of frames are initially distributed randomly over the electron micrograph in which multi-particle images are dispersed. Starting with these frames, one of them is selected and shifted randomly, and acceptance or non-acceptance of its new position is judged using the simulated annealing (SA) method in which the contrast score of the total average image is adopted as an index. After iteration of this process, the position of each frame converges so as to surround a particle and the framed images are picked up. This method is the first unsupervised fully automatic particle picking method which is applicable to EM of various kinds of proteins, especially to low-contrasted cryo-EM protein images.     

An approach to automated particle picking from electron micrographs based on reduced representation templates

Reduced representation templates are used in a real-space pattern matching framework to facilitate automatic particle picking from electron micrographs. The procedure consists of five parts. First, reduced templates are constructed either from models or directly from the data. Second, a real-space pattern matching algorithm is applied using the reduced representations as templates. Third, peaks are selected from the resulting score map using peak-shape characteristics. Fourth, the surviving peaks are tested for distance constraints. Fifth, a correlation-based outlier screening is applied. Test applications to a data set of keyhole limpet hemocyanin particles indicate that the method is robust and reliable.     

Automatic particle selection from electron micrographs using machine learning techniques

The 3D reconstruction of biological specimens using Electron Microscopy is currently capable of achieving subnanometer resolution. Unfortunately, this goal requires gathering tens of thousands of projection images that are frequently selected manually from micrographs. In this paper we introduce a new automatic particle selection that learns from the user which particles are of interest. The training phase is semi-supervised so that the user can correct the algorithm during picking and specifically identify incorrectly picked particles. By treating such errors specially, the algorithm attempts to minimize the number of false positives. We show that our algorithm is able to produce datasets with fewer wrongly selected particles than previously reported methods. Another advantage is that we avoid the need for an initial reference volume from which to generate picking projections by instead learning which particles to pick from the user. This package has been made publicly available in the open-source package Xmipp.     

Choosing haplotype-tagging SNPS based on unphased genotype data using a preliminary sample of unrelated subjects with an example from the Multiethnic Cohort Study

We describe an approach for picking haplotype-tagging single nucleotide polymorphisms (htSNPs) that is presently being taken in two large nested case-control studies within a multiethnic cohort (MEC), which are engaged in a search for associations between risk of prostate and breast cancer and common genetic variations in candidate genes. Based on a preliminary sample of 70 control subjects chosen at random from each of the 5 ethnic groups in the MEC we estimate haplotype frequencies using a variant of the Excoffier-Slatkin E-M algorithm after genotyping a high density of SNPs selected every 3-5 kb in and surrounding a candidate gene. In order to evaluate the performance of a candidate set of htSNPS (which will be genotyped in the much larger case-control sample) we treat the haplotype frequencies estimate above as known, and carry out a formal calculation of the uncertainty of the number of copies of common haplotypes carried by an individual, summarizing this calculation as a coefficient of determination, R2h. A candidate set of htSNPS of a given size is chosen so as to maximize the minimum value of R2h over the common haplotypes, h.     

Biogenic calcite particles from microalgae—Coccoliths as a potential raw material

Synthetic calcite (CaCO₃) particles are found in a broad range of applications. The geometry of particles produced from limestone or precipitation are versatile but limited to basic shapes. The microalga Emiliania huxleyi produces micro‐structured calcite platelets, called coccoliths. This article presents the results of an application‐orientated study, which includes characteristic values also used in the calcite industry for particle evaluation. It is demonstrated that coccoliths are significantly different from all industrial particles produced so far. Coccoliths are porous particles, mainly consisted of calcium carbonate, with further elements such as Mg, Si, Sr, and Fe often embedded in their structure. Their structure is extremely sophisticated, while the overall particle morphology and particle size distribution are homogeneous. This study gives a first inside into the potential of these exceptional objects and may set further impulses for their utilization in specific calcite particle applications.     

Immunoelectron microscopic localization of glutamyl-/ prolyl-tRNA synthetase within the eukaryotic multisynthetase complex

A high molecular mass complex of aminoacyl-tRNA synthetases is readily isolated from a variety of eukaryotes. Although its composition is well characterized, knowledge of its structure and organization is still quite limited. This study uses antibodies directed against prolyl-tRNA synthetase for immunoelectron microscopic localization of the bifunctional glutamyl-/prolyl-tRNA synthetase. This is the first visualization of a specific site within the multisynthetase complex. Images of immunocomplexes are presented in the characteristic views of negatively stained multisynthetase complex from rabbit reticulocytes. As described in terms of a three domain working model of the structure, in "front" views of the particle and "intermediate" views, the primary antibody binding site is near the intersection between the "base" and one "arm." In "side" views, where the particle is rotated about its long axis, the binding site is near the midpoint. "Top" and "bottom" views, which appear as square projections, are also consistent with the central location of the binding site. These data place the glutamyl-/prolyl-tRNA synthetase polypeptide in a defined area of the particle, which encompasses portions of two domains, yet is consistent with the previous structural model.     

Detecting particles in cryo-EM micrographs using learned features

A new learning-based approach is presented for particle detection in cryo-electron micrographs using the Adaboost learning algorithm. The approach builds directly on the successful detectors developed for the domain of face detection. It is a discriminative algorithm which learns important features of the particle's appearance using a set of training examples of the particles and a set of images that do not contain particles. The algorithm is fast (10 s on a 1.3 GHz Pentium M processor), is generic, and is not limited to any particular shape or size of the particle to be detected. The method has been evaluated on a publicly available dataset of 82 cryoEM images of keyhole lympet hemocyanin (KLH). From 998 automatically extracted particle images, the 3-D structure of KLH has been reconstructed at a resolution of 23.2 A which is the same resolution as obtained using particles manually selected by a trained user.     

N-terminal domain of apolipoprotein B has structural homology to lipovitellin and microsomal triglyceride transfer protein: a "lipid pocket" model for self-assembly of apob-containing lipoprotein particles.

The process of assembly of apolipoprotein (apo) B-containing lipoprotein particles occurs co-translationally after disulfide-dependent folding of the N-terminal domain of apoB but the mechanism is not understood. During a recent database search for protein sequences that contained similar amphipathic beta strands to apoB-100, four vitellogenins, the precursor form of lipovitellin, an egg yolk lipoprotein, from chicken, frog, lamprey, and C. elegans appeared on the list of candidate proteins. The X-ray crystal structure of lamprey lipovitellin is known to contain a "lipid pocket" lined by antiparallel amphipathic beta sheets. Here we report that the first 1000 residues of human apoB-100 (the alpha(1) domain plus the first 200 residues of the beta(1) domain) have sequence and amphipathic motif homologies to the lipid-binding pocket of lamprey lipovitellin. We also show that most of the alpha(1) domain of human apoB-100 has sequence and amphipathic motif homologies to human microsomal triglyceride transfer protein (MTP), a protein required for assembly of apoB-containing lipoproteins. Based upon these results, we suggest that an LV-like "proteolipid" intermediate containing a "lipid pocket" is formed by the N-terminal portion of apoB alone or, more likely, as a complex with MTP. This intermediate produces a lipid nidus required for assembly of apoB-containing lipoprotein particles; pocket expansion through the addition of amphipathic beta strands from the beta(1) domain of apoB results in the formation of a progressively larger high density lipoprotein (HDL)-like, then very low density lipoprotein (VLDL)-like, spheroidal lipoprotein particle.     

Automatic cryo-EM particle selection for membrane proteins in spherical liposomes

Random spherically constrained (RSC) single particle reconstruction is a method to obtain structures of membrane proteins embedded in lipid vesicles (liposomes). As in all single-particle cryo-EM methods, structure determination is greatly aided by reliable detection of protein “particles” in micrographs. After fitting and subtraction of the membrane density from a micrograph, normalized cross-correlation (NCC) and estimates of the particle signal amplitude are used to detect particles, using as references the projections of a 3D model. At each pixel position, the NCC is computed with only those references that are allowed by the geometric constraint of the particle’s embedding in the spherical vesicle membrane. We describe an efficient algorithm for computing this position-dependent correlation, and demonstrate its application to selection of membrane-protein particles, GluA2 glutamate receptors, which present very different views from different projection directions.     

Tobacco mosaic virus as an AFM tip calibrator

The study of high-resolution topographic surfaces of isolated single molecules is one of the applications of atomic force microscopy (AFM). Since tip-induced distortions are significant in topographic images the exact AFM tip shape must be known in order to correct dilated AFM height images using mathematical morphology operators. In this work, we present a protocol to estimate the AFM tip apex radius using tobacco mosaic virus (TMV) particles. Among the many advantages of TMV, are its non-abrasivity, thermal stability, bio-compatibility with other isolated single molecules and stability when deposited on divalent ion pretreated mica. Compared to previous calibration systems, the advantage of using TMV resides in our detailed knowledge of the atomic structure of the entire rod-shaped particle. This property makes it possible to interpret AFM height images in term of the three-dimensional structure of TMV. Results obtained in this study show that when a low imaging force is used, the tip is sensing viral protein loops whereas at higher imaging force the tip is sensing the TMV particle core. The known size of the TMV particle allowed us to develop a tip-size estimation protocol which permits the successful erosion of tip-convoluted AFM height images. Our data shows that the TMV particle is a well-adapted calibrator for AFM tips for imaging single isolated biomolecules. The procedure developed in this study is easily applicable to any other spherical viral particles.     

Automatic Peak Selection by a Benjamini-Hochberg-Based Algorithm

A common issue in bioinformatics is that computational methods often generate a large number of predictions sorted according to certain confidence scores. A key problem is then determining how many predictions must be selected to include most of the true predictions while maintaining reasonably high precision. In nuclear magnetic resonance (NMR)-based protein structure determination, for instance, computational peak picking methods are becoming more and more common, although expert-knowledge remains the method of choice to determine how many peaks among thousands of candidate peaks should be taken into consideration to capture the true peaks. Here, we propose a Benjamini-Hochberg (B-H)-based approach that automatically selects the number of peaks. We formulate the peak selection problem as a multiple testing problem. Given a candidate peak list sorted by either volumes or intensities, we first convert the peaks into -values and then apply the B-H-based algorithm to automatically select the number of peaks. The proposed approach is tested on the state-of-the-art peak picking methods, including WaVPeak [1] and PICKY [2]. Compared with the traditional fixed number-based approach, our approach returns significantly more true peaks. For instance, by combining WaVPeak or PICKY with the proposed method, the missing peak rates are on average reduced by 20% and 26%, respectively, in a benchmark set of 32 spectra extracted from eight proteins. The consensus of the B-H-selected peaks from both WaVPeak and PICKY achieves 88% recall and 83% precision, which significantly outperforms each individual method and the consensus method without using the B-H algorithm. The proposed method can be used as a standard procedure for any peak picking method and straightforwardly applied to some other prediction selection problems in bioinformatics. The source code, documentation and example data of the proposed method is available at http://sfb.kaust.edu.sa/pages/software.aspx.     

Functional association between malting quality trait components and cDNA array based expression patterns in barley (Hordeum vulgare L.)

We developed an approach for relating differences in gene expression to the phenotypic variation of a trait of interest. This allows the identification of candidate genes for traits that display quantitative variation. To validate the principle, gene expression was monitored on a cDNA array with 1400 ESTs to identify genes involved in the variation of the complex trait malting quality in barley. RNA profiles were monitored during grain germination in a set of 10 barley genotypes that had been characterized for 6 quality-associated trait components. The selection of the candidate genes was achieved via a correlation of dissimilarity matrices that were based on (i) trait variation and (ii) gene expression data. As expected, a comparison based on the complete set of differentially-expressed genes did not reveal any correlation between the matrices, because not all genes that show differential expression between the 10 cultivars are responsible for the observed differences in malting quality. However, by iteratively taking out one gene (with replacement) and re-computing the correlation, those genes that are positively contributing to the correlation could be identified. Using this procedure between 17 and 30 candidate genes were identified for each of the six malting parameters analysed. In addition to genes of unknown function, the list of candidates contains well-known malting-related genes. Five out of eight mapped candidate genes display linkage to known QTLs for malting quality traits. The described functional association strategy may provide an efficient link between functional genomics and plant breeding.