scMetric: An R package of metric learning and visualization for single-cell RNA-seq data

Abstract

Single cell RNA-seq data provide high-dimensional transcriptome features of each cell and enables the systematic study of cell types and cell states based on molecular characteristics, which are usually regarded as clusters in the high-dimensional space. Usually not all genes are equally informative for the studied biological question and the clusters are embedded in some lower dimensional subspaces.\n\nThere are several popular dimensionality reduction and visualization methods for scRNA-seq data such as PCA and t-SNE1. Different methods may utilize different kinds of distance metrics. Some methods take distance metric as a hyper-parameter for users to choose. Different metrics can result in different data distribution in the low dimensional space2 and therefore result in different final results. But there is no way to know which metric is \"the correct one\" for the particular biological study at hand. A typical practice is to ...