Automatic cell type harmonization and integration across human lung disease data sets

Harmonizing cell types across multiple single-cell RNA-seq studies and assembling them into a common framework is central to building a standardized Human Cell Atlas. Here we present CellHint, a predictive clustering tree-based tool to resolve cell type differences in nomenclature, annotation resolution, and technical biases across datasets. CellHint accurately quantifies cell-cell transcriptomic similarities and places cell types into a relationship graph that hierarchically defines shared and unique cell subtypes. We aligned lung single-cell data sets from healthy individuals and from eight lung diseases: idiopathic pulmonary fibrosis, chronic obstructive pulmonary disease, nonspecific interstitial pneumonia, chronic hypersensitivity pneumonitis, interstitial lung disease, sarcoidosis, systemic sclerosis, and polymyositis. This allowed us to identify common, but differentially annotated cell types. For example, aberrant basaloid cells aligned with KRT5-/KRT17+ distal alveolar epithelial populations, as well as some AT1 cells in different studies of pulmonary fibrosis, highlighting the power of CellHint in enhancing the specificity and interpretability of cell types across diseases and studies. Harmonisation of cell types across diseases allowed us to define a fibrotic signature and understand the exact cell types most affected in the alveolar and airway regions, and to make detailed comparison of gene expression changes specific to the diseases studied here. Finally we apply CellHint to 12 tissues from 38 datasets, providing a deeply curated cross-tissue database with ~3.7 million cells and machine learning models for automatic cell type annotation across human tissues.