Dataset Information

Miffi: Improving the accuracy of CNN-based cryo-EM micrograph filtering with fine-tuning and Fourier space information.

ABSTRACT: Efficient and high-accuracy filtering of cryo-electron microscopy (cryo-EM) micrographs is an emerging challenge with the growing speed of data collection and sizes of datasets. Convolutional neural networks (CNNs) are machine learning models that have been proven successful in many computer vision tasks, and have been previously applied to cryo-EM micrograph filtering. In this work, we demonstrate that two strategies, fine-tuning models from pretrained weights and including the power spectrum of micrographs as input, can greatly improve the attainable prediction accuracy of CNN models. The resulting software package, Miffi, is open-source and freely available for public use (https://github.com/ando-lab/miffi).

SUBMITTER: Xu D

PROVIDER: S-EPMC10888874 | biostudies-literature | 2024 Feb

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Miffi: Improving the accuracy of CNN-based cryo-EM micrograph filtering with fine-tuning and Fourier space information.

Xu Da D Ando Nozomi N

bioRxiv : the preprint server for biology 20240227

Efficient and high-accuracy filtering of cryo-electron microscopy (cryo-EM) micrographs is an emerging challenge with the growing speed of data collection and sizes of datasets. Convolutional neural networks (CNNs) are machine learning models that have been proven successful in many computer vision tasks, and have been previously applied to cryo-EM micrograph filtering. In this work, we demonstrate that two strategies, fine-tuning models from pretrained weights and including the power spectrum o ...[more]

PMID: 38405773

Similar Datasets

Project description:Preparation of cryoelectron microscopy (cryo-EM) grids for imaging of amyloid fibrils is notoriously challenging. The human islet amyloid polypeptide (hIAPP) serves as a notable example, as the majority of reported structures have relied on the use of nonphysiological pH buffers, N-terminal tags, and seeding. This highlights the need for more efficient, reproducible methodologies that can elucidate amyloid fibril structures formed under diverse conditions. In this work, we demonstrate that the distribution of fibrils on cryo-EM grids is predominantly determined by the solution composition, which is critical for the stability of thin vitreous ice films. We discover that, among physiological pH buffers, HEPES uniquely enhances the distribution of fibrils on cryo-EM grids and improves the stability of ice layers. This improvement is attributed to direct interactions between HEPES molecules and hIAPP, effectively minimizing the tendency of hIAPP to form dense clusters in solutions and preventing ice nucleation. Furthermore, we provide additional support for the idea that denatured protein monolayers forming at the interface are also capable of eliciting a surfactant-like effect, leading to improved particle coverage. This phenomenon is illustrated by the addition of nonamyloidogenic rat IAPP (rIAPP) to a solution of preaggregated hIAPP just before the freezing process. The resultant grids, supplemented with this "spectator protein", exhibit notably enhanced coverage and improved ice quality. Unlike conventional surfactants, rIAPP is additionally capable of disentangling the dense clusters formed by hIAPP. By applying the proposed strategies, we have resolved the structure of the dominant hIAPP polymorph, formed in vitro at pH 7.4, to a final resolution of 4 Å. The advances in grid preparation presented in this work hold significant promise for enabling structural determination of amyloid proteins which are particularly resistant to conventional grid preparation techniques.

Dataset Information

Miffi: Improving the accuracy of CNN-based cryo-EM micrograph filtering with fine-tuning and Fourier space information.

Publications

Miffi: Improving the accuracy of CNN-based cryo-EM micrograph filtering with fine-tuning and Fourier space information.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets