Technological advances have made it possible to obtain quantitative genome-wide measurements of the transcriptome from individual cells. Reduced sequencing costs have made it possible to profile thousands of cells in a single experiment, and with international efforts such as the Human Cell Atlas (HCA), large collections of scRNA-seq data will be available in the near future. However, computational methods for analyzing such large datasets are currently not available. Here, I will present some of the methods developed in our group for comparing scRNA-seq data collected from different experiments. I will demonstrate how unsupervised feature selection based on dropouts can help overcome batch effects and how it can be used for scmap, a fast and accurate method for projecting cells between datasets.