scSHEFT enables multiomics label transfer from scRNA-seq to scATAC-seq through dual alignment

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell multiomics data is rapidly emerging, driving a trend in label transfer from well-annotated single-cell RNA sequencing (scRNA-seq) to less-annotated data like single-cell Assay for Transposase-Accessible Chromatin sequencing (scATAC-seq).
• This cross-omics label transfer aims to annotate cell types, including novel ones, by leveraging gene expression profiles.
• Heterogeneity between scRNA-seq and scATAC-seq presents significant challenges for accurate cell type identification and hinders the discovery of novel cell types.

Purpose of the Study:

• To introduce scSHEFT, a novel computational tool designed to overcome the challenges of cross-omics label transfer.
• To enable simultaneous consideration of gene expression count data, peak count data, and Gene Activity Scores to bridge heterogeneous features.
• To improve the accuracy and scope of cell type annotation in single-cell multiomics data.

Main Methods:

• scSHEFT transforms scATAC-seq data into Gene Activity Scores using prior knowledge to harmonize features.
• Raw ATAC-seq embeddings are incorporated to mitigate information loss during feature transformation.
• A dual alignment strategy is employed, combining an anchor-based approach for interomics alignment and a contrastive-based strategy for preserving intraomics heterogeneity.

Main Results:

• scSHEFT demonstrates superior performance compared to 11 state-of-the-art methods.
• The tool was benchmarked across seven diverse datasets, showcasing its effectiveness.
• scSHEFT excels in handling datasets of varying scales and technical noise levels.

Conclusions:

• scSHEFT provides a robust solution for label transfer across heterogeneous single-cell multiomics data.
• The method enhances cell type identification and facilitates the discovery of novel cell types.
• scSHEFT represents a significant advancement in the analysis of single-cell multiomics datasets.