DeepETPicker: Fast and accurate 3D particle picking for cryo-electron tomography using weakly supervised deep learning

Abstract

Picking particles of biological macromolecules from their cryo-electron tomograms is critical to solving their 3D structures in situ. To reach sub-nanometre resolution, large numbers of particles often need to be picked, a laborious and time-consuming task if performed manually. To date, however, the adoption of automated particle-picking methods remains limited because of the limitations in their accuracy, processing speed and, for those based on learning models, manual annotation cost. To overcome the limitations, we develop DeepETPicker, a deep learning model for fast and accurate picking of 3D particles from cryo-electron tomograms. The training of DeepETPicker requires only weak supervision with low numbers of simplified Gaussian-type labels, reducing the burden of manual annotation of tomograms under very low signal-to-noise ratios. The simplified labels combined with the customized and lightweight model architecture of DeepETPicker as well as GPU-accelerated pooling enable substantially improved accuracy and accelerated processing speed. When tested on simulated as well as real tomograms, DeepETPicker outperforms the competing state-of-the-art methods by achieving the highest overall accuracy and speed, which translate into better quality of picked particles and higher resolutions of final reconstruction maps. DeepETPicker is provided in open source with a user-friendly interface to support automated particle picking for high-resolution cryo-electron tomography in situ.