Biology subjects

Kinman, L. F.

Publications and source records attributed to Kinman, L. F..

4 recordsLinked to original sources

bioRxiv2026

Heterogeneous reconstruction algorithms for cryoEM achieve limited particle classification accuracy on real benchmark datasets

The emergence of single-particle cryoEM as a powerful method for structure determination has in large part been fueled by its ability to resolve both single static structures and complex conformational landscapes. Indeed, modern approaches to the heterogeneous reconstruction task can resolve 100s-1,000s of different maps from a single cryoEM dataset. How accurate these algorithms are, however, has proven difficult to rigorously assess, due to a lack of suitable benchmark datasets containing both realistic noise features and ground-truth labels. To address this obstacle, we recently developed a series of benchmark datasets that leverage the targeting power of Cas9 and the programmable heterogeneity of DNA to newly offer access to ground-truth per-particle structural labels in real data. Here, we challenged two popular heterogeneous reconstruction algorithms with mixed particle stacks resampled in silico from these datasets, finding that existing approaches resolve the encoded heterogeneity with limited accuracy. In particular, in realistic particle stacks with complex, multi-scale, and multi-axis heterogeneity, we observed that reconstruction of encoded heterogeneity depended strongly on the application of prior information about where heterogeneity was expected, and that individual particle assignments were made with significant error even when the correct structural states were reconstructed. Both molecular breathing motions and data collection features, such as defocus and projection angle, contributed to the observed particle assignment error. These results highlight important shortcomings of existing heterogeneous reconstruction methods and suggest new avenues for method development in both data collection strategies and in heterogeneous classification and reconstruction algorithms.

biochemistry↗

bioRxiv2026

Engineered Cas9 complexes establish an experimentally grounded benchmark for heterogeneous cryoEM reconstruction methods

Single-particle cryoEM is increasingly used to resolve conformational and compositional ensembles, yet objective evaluation of heterogeneous reconstruction methods remains limited by the scarcity of experimental benchmarks with per-particle ground-truth labels. Indeed, many widely used experimental"benchmark" datasets necessarily validate observed states retrospectively while purely synthetic datasets provide ground-truth labels but typically fail to capture experimentally realistic complexities including confounding structural heterogeneity, imaging noise, contaminants, and orientation biases, which dominate real-world analyses. Here we develop an experimentally grounded benchmark dataset for heterogeneous reconstruction using catalytically inactive Streptococcus pyogenes Cas9 bound to a constant sgRNA and to target DNA duplexes engineered to carry extensions of defined length. We assembled, purified, vitrified, and imaged thirteen complexes independently, such that the dataset-of-origin provides an unambiguous label for each particles encoded state while preserving the full experimental complexity of cryoEM data. Independent refinements of the pure datasets recover the engineered DNA-extension signal and define a simple quantitative readout, DNA-extension occupancy, that increases monotonically with designed extension length. The same reconstructions also reveal substantial non-encoded conformational variability within the Cas9 core, showing that this benchmark embeds a known structural signal within broader structural heterogeneity that methods must confront in practice. To separate these axes of variation, we used systematic deep classification to generate curated particle subsets depleted of selected domain motions while retaining the encoded labels. We further provide pooled particle stacks with standardized per-particle poses in a common reference frame and a lightweight framework for in silico particle pooling to generate challenge datasets with user-defined ground-truth distributions of encoded and non-encoded structural heterogeneity. Together, this resource supports robust benchmarking of heterogeneous reconstruction algorithms and provides a biochemically tractable model system for evaluating entire cryoEM pipelines, including alternative data-collection and preprocessing approaches, under experimentally realistic conditions.

biochemistry↗

bioRxiv2024

Automated model-free analysis of cryo-EM volume ensembles with SIREn

Cryogenic electron microscopy (cryo-EM) has the potential to capture snapshots of proteins in motion and generate hypotheses linking conformational states to biological function. This potential has been increasingly realized by the advent of machine learning models that allow 100s-1,000s of 3D density maps to be generated from a single dataset. How to identify distinct structural states within these volume ensembles and quantify their relative occupancies remain open questions. Here, we present an approach to inferring variable regions directly from a volume ensemble based on the statistical co-occupancy of voxels, as well as a 3D-convolutional neural network that predicts binarization thresholds for volumes in an unbiased and automated manner. We show that these tools recapitulate known heterogeneity in a variety of simulated and real cryo-EM datasets, and highlight how integrating these tools with existing data processing pipelines enables improved particle curation and the construction of quantitative conformational landscapes.

biochemistry↗

bioRxiv2022

Uncovering structural ensembles from single particle cryo-EM data using cryoDRGN

CryoDRGN is a machine learning system for heterogeneous cryo-EM reconstruction of proteins and protein complexes from single particle cryo-EM data. Central to this approach is a deep generative model for heterogeneous cryo-EM density maps, which we empirically find effectively models both discrete and continuous forms of structural variability. Once trained, cryoDRGN is capable of generating an arbitrary number of 3D density maps, and thus interpreting the resulting ensemble is a challenge. Here, we showcase interactive and automated processing approaches for analyzing cryoDRGN results. Specifically, we detail a step-by-step protocol for analysis of the assembling 50S ribosome dataset (Davis et al., EMPIAR-10076), including preparation of inputs, network training, and visualization of the resulting ensemble of density maps. Additionally, we describe and implement methods to comprehensively analyze and interpret the distribution of volumes with the assistance of an associated atomic model. This protocol is appropriate for structural biologists familiar with processing single particle cryo-EM datasets and with moderate experience programming in Python and navigating Jupyter notebooks. It requires 3-4 days to complete.

biochemistry↗