Positive-unlabeled convolutional neural networks for particle picking in cryo-electron micrographs.

Cryo-electron microscopy (cryoEM) is fast becoming the preferred method for protein structure determination. Particle picking is a significant bottleneck in the solving of protein structures from single particle cryoEM. Hand labeling sufficient numbers of particles can take months of effort and current computationally based approaches are often ineffective. Here, we frame particle picking as a positive-unlabeled classification problem in which we seek to learn a convolutional neural network (CNN) to classify micrograph regions as particle or background from a small number of labeled positive examples and many unlabeled examples. However, model fitting with very few labeled data points is a challenging machine learning problem. To address this, we develop a novel objective function, GE-binomial, for learning model parameters in this context. This objective uses a newly-formulated generalized expectation (GE) criteria to learn effectively from unlabeled data when using minibatched stochastic gradient descent optimizers. On a high-quality publicly available cyroEM data set, we show that CNNs trained with this objective classify particles accurately with very few positive training examples. Using 1000 randomly sampled particles (out of 100k total) as references, EMAN2’s byRef method achieves 33% precision at 90% recall. With the same 1000 labeled training particles, we improve this result by roughly 40% to 46% precision at 90% recall. Remarkably, we achieve 41% precision with 1/10th the number of labeled particles and still reach 34% precision with only 1/100th the number of labeled particles at 90% recall. At all numbers of labeled particles, we improve substantially over EMAN2’s area under the precision-recall curve (AUPR). Our relative performance increase is even greater on a difficult unpublished dataset supplied by the Shapiro lab. Furthermore, we show that incorporating an autoencoder improves generalization when very few labeled data points are available. We also compare our GE-binomial method with other positive-unlabeled learning methods never before applied to particle picking. We expect our particle picking tool, Topaz, based on CNNs trained with GE-binomial, to be an essential component of single particle cryoEM analysis and our GE-binomial objective function to be widely applicable to positive-unlabeled classification problems.