Original title: HyGAnno: Hybrid graph neural network-based cell type annotation for single-cell ATAC sequencing data
Authors: Weihang Zhang,Yang Cui,Martin Loza,Sung-Joon Park,Kenta Nakai
In this article, the authors address the need for reliable cell type annotations in single-cell omics data. While computational approaches have been developed for annotating single-cell RNA sequencing (scRNA-seq) data, the same high-quality cell labels are lacking in single-cell ATAC sequencing (scATAC-seq) data. This is due to the extreme sparsity and inconsistent chromatin accessibility between datasets. To overcome this limitation, the authors propose a novel automated cell annotation method called HyGAnno. This method transfers cell type information from a well-labeled scRNA-seq reference to an unlabeled scATAC-seq target using a parallel graph neural network. Unlike existing methods, HyGAnno utilizes genomewide accessibility peak features to improve the training process. The authors also introduce a reference-target cell graph to detect cells with low prediction reliability based on graph connectivity patterns. The results of testing HyGAnno on large datasets demonstrate its advantages in accurate cell annotation, interpretable cell embedding, robustness to noisy reference data, and adaptability to tumor tissues. The authors declare no competing interests.
Original article: https://www.biorxiv.org/content/10.1101/2023.11.29.569114v2