Restoration and Reconstruction of 3D cryo-EM Images

Cryogenic electron microscopy (cryo-EM) has emerged as the preferred method for imaging macromolecular structures. Like most other high-powered imaging techniques it suffers from a trade-off between acquisition time and quality. Traditional denoising methods for cryo-EM include averaging and Fourier cropping. They are not only resource-intensive but also lose out on high-frequency details. Other denoising methods rely on constructing a model of the noisy process and rely on simplified assumptions. In this study, we train a neural network to model the noisy process and output denoised 3D cryo-EM density maps while minimizing the loss of high-frequency components; even in the absence of ‘clean’ target maps for training