Application of machine learning to associative scRNA-seq data gene expression and alternative polyadenylation sites clustering

Abstract

Cell type identification is a vital step in the analysis of scRNA-seq data. Transcriptome subtype pivotal information such as alternative polyadenylation (APA) obtained from standard scRNA-seq data can also provide valid clues for cell type identification with no alteration of experimental techniques or increased experimental costs. Furthermore, using multimodal analysis techniques and their methods, more confident cell type identification results can be obtained. For that purpose, we constructed a workflow framework: On five different scRNA-seq datasets, 18 methods based on machine learning that have not yet been applied to identify cell types by association APA and single-cell gene expression fusion were compared with three single-cell clustering methods, and compared these method against the advanced method scLAPA based on similarity network fusion (SNF). In our experiments, we used the adjusted Rand index (ARI) as a metric. We found that unsupervised methods like WMSC and supervised methods like MOGONET have more robust and excellent results in associating APA with single-cell gene expression clustering than methods based only on single-cell gene expression clustering and advanced scLAPA methods.