BPS2026 – PickET: An unsupervised method for localizing macromolecules in cryo-electron tomograms

Cryo-electron tomography (cryo-ET) data sets are rich sources of information capable of describing the localizations, structures, and interactions of macromolecules. Recent advances in hardware and software have massively improved cryo-ET data quality and throughput. Currently, one of the main bottlenecks in analyzing cryo-ET data is in localizing macromolecules in the reconstructed tomograms. Most current methods for localizing particles in cryo-electron tomograms are limited to macromolecules with known structures, require extensive manual annotations, and/or are computationally expensive. Only recently have unsupervised machine learning methods begun to be applied. Yet, their routine use in particle localization is still limited, as they rely on substantial manual input and require further improvements in both efficiency and accuracy. Here, we present PickET, a fully unsupervised method for localizing macromolecules in tomograms that does not rely on expert annotations and prior structures. Its performance is demonstrated on a diverse data set comprising a hundred and thirty-three tomograms from publicly available data sets, varying in sample types, sample preparation conditions, microscope hardware, and image processing workflows. We demonstrate that PickET can simultaneously localize macromolecules of various shapes, sizes, and abundance. Importantly, it can localize small macromolecules (<500 kDa) and less abundant macromolecules, which are challenging for supervised learning methods. PickET performs comparably to MiloPyP, a state-of-the-art method, and outperforms the latter on tomograms generated using more advanced plasma-FIB milling. The predicted particle localizations from PickET can be used for 3D classification and de novo structural characterization. Our fully unsupervised approach is efficient and scalable and enables high-throughput analysis of cryo-ET data.