2024/05/08 – Aaditya Rangan

Aaditya Rangan
Associate Professor, Courant Institute, New York University and Flatiron Institute.

Date and time: May 8th, 2024 Wednesday, at 8am PDT / 11am EDT / 3pm BST / 5pm CEST / 11pm China

Ab-initio Reconstruction from small datasets: using entropy-maximization and principal modes (EMPM).

In this talk I’ll describe a simple algorithm for obtaining a low-resolution model from a small number (~1000) picked-particle images. The algorithm (termed EMPM) is similar to classical alternating-minimization, but with a slight twist: the entropy of the viewing-angle distribution is maximized during the search. This entropy-maximization greatly improves robustness and accuracy, allowing EMPM to outperform many existing ab-initio reconstruction algorithms. Because of its low computational overhead, this algorithm could facilitate statistical strategies such as bootstrapping and cross-validation, along with the rapid assessment of particle quality.

2024/04/10 Yuntao Liu

Yuntao Liu

Postdoctoral scholar, California NanoSystems Institute, UCLA

Date and time: Apr 10th, 2024
Wednesday, at 8am PST / 11am EDT / 3pm GMT / 5pm CET / 11pm China

For the zoom and gather.town links, please join the One World Cryo-EM mailing list.

Single-particle IsoNet: an AI-based method to overcome the “preferred” orientation problem

While advances in single-particle cryoEM have enabled the structural determination of macromolecular complexes at atomic resolution, particle orientation bias (the so-called “preferred” orientation problem) remains a complication for most specimens. Existing solutions have relied on biochemical and physical strategies applied to the specimen and are often complex and challenging. Here, we develop spIsoNet, an end-to-end self-supervised deep-learning software to address the preferred orientation problem. Using preferred-orientation views to recover molecular information in under-sampled views, spIsoNet improves both angular isotropy and particle alignment accuracy during 3D reconstruction. We demonstrate spIsoNet’s capability of generating near-isotropic reconstructions from representative biological systems with limited views, including ribosomes, β-galactosidases, and a previously intractable hemagglutinin trimer dataset. spIsoNet can also be generalized to improve map isotropy and particle alignment of preferentially oriented molecules in subtomogram averaging. Therefore, without additional specimen-preparation procedures, spIsoNet provides a general computational solution to the preferred orientation problem. 

2024/03/06 Quanquan Gu

Quanquan Gu

Head of AI for Drug Design, ByteDance Research

Date and time: Mar 6th, 2024
Wednesday, at 8am PST / 11am EST / 4pm GMT / 5pm CET / 12am China

For the zoom and gather.town links, please join the One World Cryo-EM mailing list.

CryoSTAR: Leveraging Structural Prior and Constraints for Cryo-EM Heterogeneous Reconstruction

Resolving conformational heterogeneity in cryo-electron microscopy (cryo-EM) datasets remains a significant challenge in structural biology. Previous methods have often been restricted to working exclusively on volumetric densities, neglecting the potential of incorporating any pre-existing structural knowledge as prior or constraints. In this talk, I will present a novel methodology, cryoSTAR, that harnesses atomic model information as structural regularization to elucidate such heterogeneity. Our method uniquely outputs both coarse-grained models and density maps, showcasing the molecular conformational changes at different levels. Validated against four diverse experimental datasets, spanning large complexes, a membrane protein, and a small single-chain protein, our results consistently demonstrate an efficient and effective solution to conformational heterogeneity with minimal human bias. By integrating atomic model insights with cryo-EM data, cryoSTAR represents a meaningful step forward, paving the way for a deeper understanding of dynamic biological processes. This is joint work with Yilai Li, Yi Zhou, Jing Yuan and Fei Ye.

More graphics are available at: https://bytedance.github.io/cryostar/.

2023/12/06 Marc A. Gilles

Marc A. Gilles

Postdoctoral Research Associate, Program in Applied & Computational Mathematics, Princeton University

Date and time: Dec 6th, 2023
Wednesday, at 8am PST / 11am EST / 4pm GMT / 5pm CET / 12am China

For the zoom and gather.town links, please join the One World Cryo-EM mailing list.

A Bayesian Framework for Cryo-EM Heterogeneity Analysis using Regularized Covariance Estimation

Proteins and the complexes they form are central to nearly all cellular processes. Their flexibility, expressed through a continuum of states, provides a window into their biological functions. Cryogenic-electron microscopy (cryo-EM) is an ideal tool to study these dynamic states as it captures specimens in non-crystalline conditions and enables high-resolution reconstructions. However, analyzing the heterogeneous distribution of conformations from cryo-EM data is challenging. Current methods face issues such as a lack of explainability, overfitting caused by lack of regularization, and a large number of parameters to tune; problems exacerbated by the lack of proper metrics to evaluate or compare heterogeneous reconstructions. To address these challenges, we present RECOVAR, a white-box method based on principal component analysis (PCA) computed via regularized covariance estimation that can resolve intricate heterogeneity with similar expressive power to neural networks with significantly lower computational demands.

2023/11/01 Bronwyn Lucas

Bronwyn Lucas

Assistant Professor of Biochemistry, Biophysics and Structural Biology, UC Berkeley

Date and time: Nov 1st, 2023
Wednesday, at 8am PDT / 11am EDT / 3pm GMT / 5pm CET / 11pm China

For the zoom and gather.town links, please join the One World Cryo-EM mailing list.

Baited reconstruction for high-resolution in situ structure determination without template bias

Cryogenic electron microscopy (cryo-EM) has revolutionized structural biology, rapidly increasing the number of available molecular structures. Because of this, as well as advances in structure prediction, the focus of structural biology has begun to shift to studying macromolecular structures in their native cellular environment. A dominant feature of cryo-EM images is shot noise, making the identification of small particles of interest difficult. This is further compounded by structural noise if these particles are imaged against a background of other molecules, such as inside a cell. High-resolution template matching is currently the most effective method for visual proteomics in 2D and 3D. We have shown that 2D template matching (2DTM) can be used to localize complexes with high precision, even in the presence of cellular background. Once localized, these particles may be averaged together in 3D reconstructions; however, in both 2D and 3D, regions included in the template may suffer from template bias, leading to inflated resolution estimates and making the interpretation of high-resolution features unreliable.

I will describe our new work where we evaluated conditions that minimize template bias and show that molecular features not present in the template can be reconstructed at high resolution from targets found by 2DTM, extending prior work at low-resolution. Moreover, I will describe a quantitative metric for template bias to aid the interpretation of 3D reconstructions calculated with particles localized using high-resolution templates and fine angular sampling. Baited reconstruction allows us to retain the benefits of precise particle localization in situ with high-resolution template matching while accounting for and minimizing overfitting.

2023/05/03 Amit Moscovich

Amit Moscovich

Assistant Professor, Department of Statistics and Operations Research, Tel Aviv University

Date and time: May 3rd, 2023
Wednesday, at 8am PDT / 11am EDT / 4pm BST / 5pm CEST / 11pm China

For the zoom and gather.town links, please join the One World Cryo-EM mailing list.

Tools for heterogeneity in cryo-EM: manifold learning, disentanglement and optimal transport

Despite much recent progress, the reconstruction and analysis of macromolecules with continuous conformational heterogeneity remains a key challenge.  In this talk we present two promising tools that can aid in various stages of the reconstruction and analysis pipelines.

In the first part of the talk, the notion of manifold factorization and disentanglement is presented.  This is an approach to dimensionality reduction where different aspects of the data are assigned separate coordinates.  It has potential applications both for the reconstruction of single-particle cryo-EM samples with continuous heterogeneity and for the analysis of the reconstructed volumetric datasets.

In the second part of the talk, we will discuss the potential applications of optimal transport for cryo-EM, in particular for the analysis of heterogeneous samples, as well as for class-averaging and particle picking.  Optimal transport metrics are closely related to physical motion, making them a natural choice for many of the core problems in cryo-EM.  Historically, computational bottlenecks have limited the applicability of optimal transport.  However, recent advances in computational optimal transport have yielded fast approximation schemes that can be readily used for the analysis of high-resolution images and volumetric arrays.

2023/01/18 Carlos Oscar Sorzano

Carlos Oscar Sorzano

Staff Researcher at the National Center of Biotechnology (CSIC)

Date and time: January 18th, 2023
Wednesday, at 8am PST / 11am EST / 4pm GMT / 5pm CET / midnight China

For the zoom and gather.town links, please join the One World Cryo-EM mailing list.

Bias, variance and map validation in Single Particle Analysis by CryoEM

Single Particle Analysis by CryoEM has established as a mature technique to elucidate the three-dimensional structure of biological macromolecules. Once the experimental data is acquired, reconstructing the macromolecule’s map involves the estimation of millions of parameters. Due to the low Signal-to-Noise Ratio, this estimation is prone to mistakes. Depending on the error size, these mistakes may result into small perturbations in the reconstructed map (variance) or large artifacts (bias). In this talk we will discuss different kinds of mistakes that can be committed and how to identify and tackle them.

2023/03/01 Petar Petrov

Petar Petrov

Postdoctoral Scholar at University of California, Berkeley

Date and time: March 1st, 2023
Wednesday, at 8am PST / 11am EST / 4pm GMT / 5pm CET / midnight China

For the zoom and gather.town links, please join the One World Cryo-EM mailing list.

Laser phase-contrast Cryo-EM and associated computational opportunities

A phase plate can provide optimum image contrast for weak-phase objects in transmission  electron microscopy, but approaches toward realizing a phase plate have suffered from  instabilities. We have developed a phase plate that is based on coherently phase-shifting the  electron wave function by a laser beam, which is built up to a record-high intensity of ~400  GW/cm^2 by resonance in a Fabry-Perot cavity. We have demonstrated contrast enhancement  with the laser phase plate (LPP) and shown the long-term stability of the device, as well as  generated a high-resolution map of 20S proteasome particles using a standard single-particle  cryo-electron microscopy (Cryo-EM) workflow. 

This talk will focus on our recent work to move beyond proof-of-concept, as well as the  computational opportunities that lie ahead. To demonstrate the benefits of the LPP to Cryo-EM  as well as cryo-electron tomography, we will soon begin working with a state-of-the-art  microscope equipped with a spherical aberration corrector, gun monochromator, and post column energy filter. We will explore improvements to the phase plate design and pursue new  strategies for image acquisition and processing, such as high-resolution two-dimensional  template matching.

2023/02/01 Tamir Bendory

Tamir Bendory

Electrical Engineering Assistant Professor at Tel Aviv University

Date and time: February 1st, 2023
Wednesday, at 8am PST / 11am EST / 4pm GMT / 5pm CET / midnight China

For the zoom and gather.town links, please join the One World Cryo-EM mailing list.

Recovering small molecular structures using Cryo-EM

Any current Cryo-EM algorithmic pipeline entails recovering the 3-D structure after particle picking. However, the signal-to-noise ratio of the data, and thus the reliability of particle picking, drops with the molecular mass of the specimens. Accordingly, it is commonly believed that Cryo-EM cannot be used to map molecules with a molecular mass below a certain threshold. Challenging this misconception, I will argue that finding the particle picking is not a prerequisite for structure determination and thus small molecules are, at least in principle, within reach of Cryo-EM. Then, I will introduce two computational frameworks to bypass particle picking and show numerical results.

2022/11/09 Willem Diepeveen

Willem Diepeveen

Doctoral Candidate at Cambridge University

Date and time: November 9nd, 2022
Wednesday, at 8am PST / 11am EST / 4pm GMT / 5pm CET / midnight China

For the zoom and gather.town links, please join the One World Cryo-EM mailing list.

Revisiting Orientation Estimation in Cryo-EM Volume Refinement

In Cryo-EM 3D map refinement, popular software packages jointly reconstruct the 3D map while estimating orientations. The orientation estimation can roughly be categorised in two type of approaches: marginalisation and optimisation. While the former tends to be more robust to noise, the latter has better consistency with respect to the data. So far it appears difficult to obtain both data-consistency and noise-robustness in a single method.

In this talk we will revisit the orientation estimation process. In particular, we develop an alternative that can be interpreted as being “in between marginalisation and optimisation” and argue that this new method is both robust to noise and data-consistent. Additionally, the framework of lifting-based global optimisation on manifolds allows analysis of the proposed methods that lead to several practical theoretical guarantees. Both theoretical results and performance on simulated data will be tested in numerical experiments.