세미나

The recent advent of single-cell RNA sequencing technology has enabled the detailed characterization of human cells in various organs from diverse disease states. As single-cell data continues to pour out, it has become crucial to integrate them effectively into a comprehensive atlas. However, the discrepancy in metadata terminology and bioinformatic analysis pipeline across publicly deposited datasets often hinders the integration at the cell count matrix level. In this study, using an automatic public data search process, we unbiasedly collected over 20 million single-cell transcriptomic profiles from more than 500 independent studies, which contain more than 2,000 single-cell transcriptome datasets from diverse human organs and disease states. We invented a single-cell data remapping pipeline for the efficient re-analysis of the whole dataset from the raw sequence files at its highest genome coverage while excluding the biases from computational data processing steps. Metadata information has been curated and classified to provide harmonized terminology for the entire dataset. The integration of remapped single-cell transcriptome dataset minimizes the batch effect, allowing for the robust identification of cell types and the organ-specific, disease-specific, and sex-specific gene signatures for each cell type. As our remapping pipeline utilizes a genomic binning approach, the splicing patterns and intergenic transcripts were also retrieved, maximizing the interpretability of the single-cell transcriptome. Using this reference atlas of human cell types, we provide a universal reference for the deconvolution and interpretation of multi-organ spatial transcriptomics data collection. In conclusion, we represent a fully curated, annotated, and harmonized cell network that could provide a fundamental axis for future data integration.