Single particle cryo-electron microscopy (cryoEM) is an imaging technology that allows direct observation of proteins in native and near-native states in atomic detail. However, determining the structure of a protein from projections captured by the EM requires collecting millions of examples.

By leveraging positive-unlabeled learning, we can use a small number of example protein projections, provided by the user, to train a neural network to detect proteins of any size or shape. Topaz, the implementation of this framework, can detect significantly more proteins than all other software tested while also minimizing manual labeling required by the user. Furthermore, the positive-unlabeled approach to object detection is broadly applicable to other microscopy domains.

• cryoEM image collection and processing procedures;
• particle picking in cryoEM images with Topaz using positive-unlabeled learning; and
• prospects in the automation of cryoEM.
  

Content

content goes here