Single-cell RNA-seq analysis through a hybrid deep learning and clustering approach provides robust cell type profiling
The recent emergence of single-cell RNA sequencing (scRNA-Seq) technology has allowed the characterization of heterogeneous cell populations by accurately measuring transcript expression levels in individual cells. Such studies routinely use pattern recognition methods to discover novel information in single cell sequencing data, which might correspond to new cell types or cell developmental stages.
Current dimension reduction and clustering methods are not well suited to deal with the considerable amount of noise created by the experiments themselves and the biological noise due to variation in cell states. In this work, we implement a novel deep learning and clustering approach to learn a compressed, informative and robust representation of the scRNA-Seq data, in an unbiased and unsupervised manner. We show that our method performs better than classical and state-of-the-art techniques to identify cell types in heterogenous populations.