Cryogenic Electron Microscopy (cryo-EM) is a powerful imaging technique for determining the high-resolution structures of biological macromolecules. However, it still faces several challenges, including an extremely low signal-to-noise ratio, heterogeneity, and preferred orientation. In this talk, we will present our computational approaches to addressing these challenges, covering the entire cryo-EM data processing pipeline. Specifically, we will discuss methods for data evaluation, processing, post-processing, and model building, combining mathematical models and unsupervised deep learning approaches. We will also present experimental results from biological laboratories to demonstrate the effectiveness of the proposed algorithms.