Research On Cell Type Annotation Methods For Single-cell RNA-sequencing Data

The emergence of single-cell RNA-sequencing technology has ushered in a new era of genomics and transcriptomic research,enabling a qualitative leap in our understanding of cellular heterogeneity.Among them,cell type annotation is a common bottleneck in the analysis of single-cell genomics experiments.Traditional cell type annotation methods generally cluster single-cell RNA sequencing data first,and then manually annotate its cell type through the specific marker gene of each cell cluster,which are difficult to adapt to the rapidly growing dataset scale and has large subjectivity.In recent years,with the deepening of research,it has become more and more common to train reference datasets with known annotation information through models such as neural networks to automatically predict unknown target datasets.Existing research methods often train the reference dataset and the target dataset separately,which cannot eliminate the inherent differences in data distribution between the two datasets.To achieve efficient annotation transfer,we propose two automated cell-type annotation methods for single-cell RNA-sequencing data.(1)A cell type annotation method based on dynamic distribution adaptive network is proposed.The method combines neural networks and dynamic distribution adaptive strategies to achieve annotation transfer from a reference dataset with known cell type information to an unknown target dataset.The experimental results on human pancreatic tissue datasets of different sequencing technologies and mouse tissue datasets of different ages show that the method can almost achieve cell type annotation performance superior to other advanced methods on the target dataset.In addition,we perform UMAP visualization on the latent space of the data after model training,which can effectively eliminate the data distribution difference between the source dataset and the target dataset,and is suitable for cell type annotation of different batches of single-cell RNA-sequencing datasets.(2)A cell type annotation method based on deep dynamic domain adversarial adaptive networks is proposed.This method combines deep networks and dynamic adversarial adaptation strategies to learn domain-invariant cellular gene expression information through end-to-end adversarial training,enabling efficient annotation transfer.We conduct experiments on human peripheral blood mononuclear cell datasets with disturbance conditions and rat tissue datasets of different ages,and the results show that the method can accurately annotate the cell types of unknown datasets,outperforming other advanced methods.In particular,compared to other methods,when there is a large difference in data distribution between the reference dataset and the target dataset,the annotation performance of this method is improved most significantly.