Microscopy Conference 2023

Prof. Dr.

Session chairs

Image analysis is an integral part of the single-particle as well as the tomography cryo-EM workflows. In single-particle cryo-EM, automated image acquisition procedures generate large data sets that enable the ensemble structure determination of increasing number of conformations at higher resolution. This procedure also requires algorithms that can detect the variability in large data sets and distinguish different conformations from each other. Similarly, using subtomogram averaging the analysis of structural variability is driven by improved classification techniques of larger data sets. Due to the processing of volumes instead of projections, this method is more computationally intensive than single particle analysis and therefore requires new software solutions. Moreover, image processing of large numbers of cellular cryotomograms by new segmentation and template matching methods will enable a quantitative image analysis of the cellular environment.

DeepHEMNMA approach for analyzing continuous conformational heterogeneity in single-particle cryo-EM images

Abstract text (incl. figure legends and references) Introduction: Single-particle cryo electron microscopy (cryo-EM) allows 3D reconstruction of multiple conformations of purified biomolecular complexes from their 2D images. The elucidation of different conformations is the key to understand the molecular mechanisms behind the biological functions of the complexes and the key to novel drug discovery. The standard cryo-EM data analysis procedures involve many rounds of 2D and 3D classifications to disentangle and interpret the combined conformational, orientational, and translational heterogeneity. Gradual conformational transitions give raise to many intermediate conformational states. Continuous conformational heterogeneity in cryo-EM data (a mixture of many intermediate conformational states), due to such gradual conformational transitions, is both an obstacle for high-resolution 3D reconstruction of different states and an opportunity to obtain the information about multiple coexisting states at once. Objective: We aim at developing new methods for analyzing continuous conformational heterogeneity in cryo-EM data, which will be fast, user-friendly, and allow obtaining the full conformational landscape. Methods: HEMNMA method [1], that we specifically developed for analyzing continuous conformational heterogeneity in cryo-EM data, determines the conformation, orientation, and position of the complex in each single particle image by analyzing images using normal modes (motion directions simulated for a given atomic structure or EM map), which in turn allows determining the full conformational space of the complex but at the price of high computational cost. To speed up HEMNMA, we recently combined it with a deep learning approach. This deep learning extension of HEMNMA is referred to as DeepHEMNMA [2]. Results: With a synthetic dataset, we have shown that DeepHEMNMA is more than 40 times faster than HEMNMA [2]. With an experimental dataset, we have shown that DeepHEMNMA reveals the conformational heterogeneity that is out of reach of standard methods [2]. Conclusion: DeepHEMNMA is a fast and user-friendly software that allows obtaining the full conformational landscape of biomolecules. It is available in ContinuousFlex, open-source software package that we are developing. ContinuousFlex provides user-friendly graphical interface to several methods for analyzing continuous conformational heterogeneity in vitro [1-3] and in situ [4-5], and it is available as a plugin for Scipion [6-7]. In this talk, I will present DeepHEMNMA and its performance using synthetic and experimental cryo-EM images. Also, I will briefly introduce ContinuousFlex. [1] https://doi.org/10.1016/j.str.2014.01.004 [2] https://doi.org/10.3389/fmolb.2022.965645 [3] https://doi.org/10.1016/j.jmb.2022.167483 [4] https://doi.org/10.3389/fmolb.2021.663121 [5] https://doi.org/10.1016/j.jmb.2021.167381 [6] https://doi.org/10.1016/j.jsb.2022.107906 [7] https://pypi.org/project/scipion-em-continuousflex

LS 4: Image analysis of large data sets

Invited talk

Microscopy image restoration and downstream analysis – recent improvements and hopes for the future

Abstract text (incl. figure legends and references) The necessity to analyze scientific images is as old as the ability to acquire such data. While this analysis did initially happen by observation only, modern microscopy techniques now enable us to image at unprecedented spatial and temporal resolutions, through the 'eyes' of many and very diverse imaging modalities. The unfathomable amounts of data acquired in the context of biomedical research cannot any longer be analyzed by manual observation alone. Instead, algorithmic solutions are helping researchers to study and quantify large image data. In the past years, our abilities to use artificial neural networks (ANNs) for the automated analysis of scientific image data gained significant traction, and many important analysis problems have now much improved solutions based on ANNs. At the same time, we start being aware of limitations that come with this new set of machine learning approaches. In my talk I would like to update you on some of the latest algorithmic developments from our and other labs. More specifically, I will talk about improved but easy to use image restoration and segmentation methods and the efforts of our community to store, share, and run ANN based methods via the BioImage Model Zoo -- a Horizon Europe funded infrastructure we are currently establishing. Finally, I will carefully attempt to look into the future and share my predictions about how artificial intelligence will help us make valuable scientific discoveries at elevated pace.

LS 4
Image analysis of large data sets

Appointment

Session chairs

Description

Programme

LS 4Image analysis of large data sets

Appointment

Session chairs

Description

Programme

LS 4
Image analysis of large data sets