The cryo-electron microscopy (cryo-EM) single-particle analysis requires tens of thousands of particle projections to reveal structural information of macromolecular complexes. However, due to the low signal-to-noise ratio and the presence of high contrast artifacts and contaminants in the micrographs, the semiautomatic and fully automatic particle picking algorithms tend to suffer from high false-positive rates, which degrades the confidence of structure determination. In this study, we introduce PickerOptimizer (PO), a transfer learning-based classification neural network for particle pruning in cryo-EM, as an additional strategy to complement the current automated particle picking algorithms. To achieve high classification performance with minimal human intervention, we adopted two key strategies: (1) utilizing the transfer learning techniques to train the convolutional neural network, where the knowledge gained from public classification datasets is applied to the field of cryo-EM. (2) Designing a multiloss strategy, a combination of multiple loss functions, to guide the optimization of the network parameters. To reduce the domain shift between cryo-EM images and natural images for pretraining, we build the first image classification dataset for cryo-EM, which contains positive and negative samples collected from EMPIAR entries. The PO is tested on 14 public experimental datasets, achieving accuracy and F1 scores above 95% in most cases. Furthermore, three case studies are provided to verify the model performance by applying PO on problematic particle selections, showing that our algorithm achieved better or comparable performance compared with other particle pruning strategies.

Single-particle cryo-electron microscopy (cryo-EM) has recently become a mainstream technique for the structural determination of macromolecules. Typical cryo-EM workflows collect hundreds of thousands of single-particle projections from thousands of micrographs using particle-picking algorithms. However, the number of false positives selected by these algorithms is large, so that a number of different \'cleaning steps\' are necessary to decrease the false-positive ratio. Most commonly employed techniques for the pruning of false-positive particles are time-consuming and require user intervention. In order to overcome these limitations, a deep learning-based algorithm named Deep Consensus is presented in this work. Deep Consensus works by computing a smart consensus over the output of different particle-picking algorithms, resulting in a set of particles with a lower false-positive ratio than the initial set obtained by the pickers. Deep Consensus is based on a deep convolutional neural network that is trained on a semi-automatically generated data set. The performance of Deep Consensus has been assessed on two well known experimental data sets, virtually eliminating user intervention for pruning, and enhances the reproducibility and objectivity of the whole process while achieving precision and recall figures above 90%.