As single-cell technologies evolved over years, diverse single-cell atlas datasets have been rapidly accumulated. Integrative analyses harmonizing such datasets provide opportunities for gaining deep biological insights. In this talk, we present a computational approach developed for fast and accurate integration of large-scale single-cell atlases. Our method incorporates generative adversarial networks and auto-encoder structures into a unified framework. Through integration of numerous datasets, we show that our method outperforms other state-of-the-art methods in terms of scalability and accuracy.