Scaling data analyses in cellular cryoET using comprehensive segmentation

bioRxiv (Cold Spring Harbor Laboratory)

Abstract

Abstract Automation and improved hardware have greatly accelerated the rate of data generation in cryoET. As the field moves towards quantitative cryoET , the scale of the resulting datasets presents a significant challenge for analysis and interpretation. To explore ways of handling datasets comprising thousands of tomograms, we investigated a comprehensive segmentation strategy – assigning an ontology-based identity to every voxel in a dataset – that is based on the sequential application of multiple convolutional neural networks. Using an openly available dataset of over 1800 Chlamydomonas reinhardtii tomograms as a test case, we demonstrate the segmentation of 25 different subcellular features across the full dataset, while requiring only a few seconds of processing time per tomogram. We show how the approach enables the representation of large datasets as searchable databases and propose the usage of ontology-based segmentations for improving two common processing tasks in cryoET. First, we explore context-aware particle picking as a method to retain biological context when selecting particles for subtomogram averaging and other downstream analyses. Secondly, we demonstrate area-selective template matching , where we use segmentation-based masks to avoid redundant computations in template matching and enable >500-fold faster processing in specific cases. To illustrate the utility of the approach, all segmentation results have also been made available online via cryopom.streamlit.app .

Authors 3

1. Mart G.F. Last corresponding Aachen Department of Cell and Chemical Biology

   Leiden University Medical Center · MRC Laboratory of Molecular Biology

   Affiliation as printed

   Department of Cell and Chemical Biology, Leiden University Medical Center, Leiden, the Netherlands

   Structural Studies Division, MRC Laboratory of Molecular Biology, Cambridge, United Kingdom

   MRC Laboratory of Molecular Biology

2. Lenard M. Voortman Aachen Department of Cell and Chemical Biology

   Leiden University Medical Center

   Affiliation as printed

   Department of Cell and Chemical Biology, Leiden University Medical Center, Leiden, the Netherlands

   Leiden University Medical Center

3. Thomas Harry Sharp

   University of Bristol

   Affiliation as printed

   School of Biochemistry, University of Bristol, Bristol, United Kingdom

   University of Bristol