Increasingly large scRNA-seq datasets demand better and more scalable analysis tools. Here, the authors introduce a scalable unsupervised deep embedding algorithm that clusters scRNA-seq data by iteratively optimizing a clustering objective function and enables removal of batch effects.