Cell-cell Similarity Learning And Cell Type Discovery By Integrating Alternative Polyadenylation

Single-cell RNA sequencing(scRNA-seq)can study cell heterogeneity with high resolution,which makes it possible to analyze single-cell gene expression from the whole transcriptome.However,due to the low mRNA capture rate and sequencing depth of scRNA-seq,single-cell gene expression data are often accompanied by high variability and high sparsity,which hinders the reliable quantification of cell similarity and the determination of cell type.However,alternative polyadenylation(APA)is used to describe transcriptional information from scRNA-seq data,which can dynamically vary among cell types,indicating the possibility of identifying cell types by APA expression profile.In this study,we proposed a comprehensive framework called scLAPA,which integrates single-cell gene expression and APA information from the same scRNA-seq data by the strategy of similarity network fusion for cell-cell similarity learning and cell type clustering.Using seven public scRNA-seq datasets from animals and plants,two application scenarios,cell-cell similarity learning and single-cell clustering,were proposed for the comparison of scLAPA with other seven similarity measurements and five single-cell clustering algorithms.Moreover,the influence of different similarity measurements,clustering algorithms and model parameters on the performance of scLAPA framework were also investigated.Results showed that scLAPA has more stable and efficient performance in learning cell-cell similarity and cell clustering compared with other methods.Additionally,two hidden subsets of peripheral blood mononuclear cells that can not be detected by only the gene expression data were discovered by using scLAPA.scLAPA is a flexible and efficient single-cell transcriptome data analysis framework and toolkit,which provides an effective and stable fusion strategy.It can integrate both layers of gene and APA information for learning cell-cell similarity and improving cell type clustering.New cell types can also be detected with scLAPA by incorporating APA information to enhance gene expression profile.Additional tools and methods can be embedded in scLAPA and it can also be integrated into most existing scRNA-seq pipelines or tools.scLAPA is available at https://github.com/BMILA B/scLAPA.