FastIntegration: a versatile R package for accessing and integrating large-scale single-cell RNA-seq data

Constructing atlas-scale comprehensive cell maps from publicly available data enables extensive data mining to achieve novel biological insights. Data integration to construct such maps require both harmonized datasets and an atlas-scale capable data integration tool. The first requirement is met by DISCO, a comprehensive repository of harmonized publicly available single cell data with standardized annotation. To meet the second requirement, the tool must have the capacity to integrate hundreds if not thousands of samples within an acceptable time frame. Moreover, it should output batch-corrected gene expression values to facilitate downstream analyses. Here, we present FastIntegration, a package which allows users to access and integrate public data in a convenient way. FastIntegration provides a fast and high-capacity version of Seurat Integration which can integrate more than 4 million cells within 2 days. It outputs batch corrected values for all genes that we can use for downstream analyses. For the first time, we demonstrated that using batch corrected values can improve the performance of downstream analyses. In particular, we found more accurately identified differentially expressed genes for cell types that are not shared between batches. Moreover, we also showed that FastIntegration outperforms existing methods for both homogeneous and heterogeneous data integration. FastIntegration also provides an API for programmatic access to data hosted on DISCO, a single-cell RNA-seq database that contains more than 5200 single-cell datasets. Users can filter for data at the sample level by tissue, disease and platform etc., and at cell level by specifying the expressed and unexpressed genes.