CryoGAN: A New Reconstruction Paradigm for Single-Particle Cryo-EM via Deep Adversarial Learning

Cryo-electron microscopy (Cryo-EM) has revolutionized structural biology over the last decade by delivering 3D structures of biomolecules at near-atomic resolution. It produces thousands of noisy projections from separate instances of the same but randomly oriented biomolecule. These noisy projections are then used to reconstruct the 3D structure of the biomolecule. However, the current reconstruction pipelines are computationally expensive since they rely on maximum likelihood techniques which need to estimate pose or conformation for each projection.

CryoGAN is a new paradigm for single-particle cryo-EM reconstruction based on unsupervised deep adversarial learning. CryoGAN sidesteps the pose-estimation problem by using a generative adversarial network (GAN) to learn the 3D structure whose simulated projections most closely match the acquired projections in a distributional sense. The architecture of CryoGAN resembles that of standard GAN, with the twist that the generator network is replaced by a model of the cryo-EM image acquisition process. CryoGAN is an unsupervised algorithm that only demands projection images. No initial volume estimate or prior training is needed. CryoGAN requires minimal user interaction and can provide reconstructions in a matter of hours on a high-end GPU. In addition, it is backed by mathematical guarantees on the recovery of the correct structure. Moreover, its extension, called MultiCryoGAN can reconstruct continuous conformations of dynamic biomolecules, thus helping solve the most important open problem in the field without pose or conformation estimation.