Cell-type annotation with accurate unseen cell-type identification using multiple references

Abstract

The recent advances in single-cell RNA sequencing (scRNA-seq) techniques have stimulated efforts to identify and characterize the cellular composition of complex tissues. With the advent of various sequencing techniques, automated cell-type annotation using a well-annotated scRNA-seq reference becomes popular but relies on the diversity of cell types in the reference. There are generally unseen cell types in the query data of interest because most data atlases are obtained for different purposes and techniques. When annotating new query data, identifying unseen cell types is fundamental not only for improving annotation accuracy but also for novel biological discoveries. Here, we propose mtANN (multiple-reference-based scRNA-seq data annotation), a new method to automatically annotate query data while accurately identifying unseen cell types with the aid of multiple references. Key innovations of mtANN include the integration of deep learning and ensemble learning to improve prediction accuracy, and the introduction of a new metric defined from three complementary aspects to distinguish between unseen cell types and shared cell types. In addition, a data-driven method is provided to adaptively select threshold for unseen cell-type identification. We demonstrate the advantages of mtANN over state-of-the-art methods for unseen cell-type identification and cell-type annotation on two benchmark dataset collections, as well as its predictive power on a collection of COVID-19 datasets. The source code and tutorial are available at https://github.com/Zhangxf-ccnu/mtANN. Author summarySingle-cell transcriptomics is rapidly advancing our understanding of the cellular composition of complex tissues and organisms. With the advent of various sequencing techniques, automatic cell-type annotation using well-annotated single-cell RNA sequencing (scRNA-seq) references has become popular. Compared with unsupervised cell-type annotation methods, it can be more easily applied to different data, saving labor and time costs. However, it relies on the diversity of cell types in the reference so there are generally unseen cell types in the query data. These unseen cell types need to be identified when annotating new sequencing data not only for improving annotation accuracy but also for novel biological discoveries. To address these issues, we propose mtANN, a new method to automatically annotate query data while accurately identify unseen cell types with the help of multiple references. We demonstrate the annotation performance of mtANN in PBMC and Pancreas collections when different proportions of unseen cell types are present in the query dataset. We also verify the practical application of mtANN in a collection of COVID-19 datasets for patients with different symptoms. When there are unseen cell types in the query dataset, mtANN is able to identify the unseen cell types and accurately annotate the shared cell types, especially the two cell types that are biologically similar.