pytom-match-pick: A tophat-transform constraint for automated classification in template matching

Abstract

• A new software for template matching (TM) in cryo-electron tomography (cryo-ET), pytom-match-pick, is introduced and its implementation of point spread function weighting and background normalization are validated. • TM in cryo-ET is often prone to false positives when using it for automated annotation, hence a novel application of a tophat transform on score maps is tested and proven effective at filtering steep local maxima. • The tophat transform is integrated into a dual-constraint thresholding strategy and improves automated classification of macromolecules in a simulated benchmark and, in experimental data, leading to improved subtomogram averages. Template matching (TM) in cryo-electron tomography (cryo-ET) enables in situ detection and localization of known macromolecules. However, TM faces challenges of weak signal of the macromolecules and interfering features with a high signal-to-noise ratio, which are often addressed by time-consuming, subjective manual curation of results. To improve the detection performance we introduce pytom-match-pick, a GPU-accelerated, open-source command line interface for enhanced TM in cryo-ET. Using pytom-match-pick, we first quantify the effects of point spread function (PSF) weighting and show that a tilt-weighted PSF outperforms a binary wedge with a single defocus estimate. We also assess previously introduced background normalization methods for classification performance. This indicates that phase randomization is more effective than spectrum whitening in reducing false positives. Furthermore, a novel application of the tophat transform on score maps, combined with a dual-constraint thresholding strategy, reduces false positives and improves precision. We benchmarked pytom-match-pick on public datasets, demonstrating improved classification and localization of macromolecules like ribosomal subunits and proteasomes that led to fewer artifacts in subtomogram averages. This tool promises to advance visual proteomics by improving the efficiency and accuracy of macromolecule detection in cellular contexts.