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

Distance metrics play important roles in the clustering and visualization of high-dimensional data. In single-cell genomics, PCA and t-SNE are widely used as tools for dimension reduction, clustering and/or visualization. They are based on similarity measures between gene expression vectors. For complicated single-cell studies, there could be multifaceted underlying relations among the cells according to different angles of study. Fixed metrics cannot provide the flexibility for exploring the data from different angles. We developed scMetric, an R package that apply a metric learning algorithm to scRNA-seq data. It allows users to give example samples to tell expected angle they would use to analyze the data, and the package learns the metric from the examples and apply the metric for downstream clustering and visualization. The package also outputs the genes that are weighted as more important in learned metric.