t-SNE transformation: a normalization method for local features of single-cell RNA-seq data

Abstract

Single-cell RNA sequencing has been widely used by biology researchers. There are many analysis tools developed accordingly. However, almost all of them use log transformation in the process of normalization, which may result in system bias on global features of datasets. It is considered that they may not be suitable for researchers who expect local and detailed features of datasets, such as rare cell population and independent expressed genes. In this study, we developed a method called t-SNE transformation to replace log transformation. We found that it can well respond to some specific bio-markers in real datasets. When the cluster number was changed, t-SNE transformation was steadier than log transformation. Further study showed that clustering after t-SNE transformation detected the residual cells more accurately after majority cells of one type were removed manually. It was also sensitive to a highly-variated independent gene added artificially. In conclusion, t-SNE transformation is an alternative normalization for detecting local features, especially interests arouse in cell types with rare populations or highly-variated but independently expressed genes. 2 potent RNA-sequencing techniques to describe transcriptomic information stored in every cell of samples. It is 3 often used to analyze global features including general cell types, their relationships within [1] and structural or 4 functional alterations during certain biological processes [4, 25]. It can be operated by various analyzing tools 5 such as Seurat [18], SC3 [13] or Monocle [22].6