Uniform Manifold Approximation and Projection (UMAP) is a dimensionality reduction technique to construct graphs showing the complexity of specimens and how the cell clusters diverse between each other according to a selected group of features (i.e. genes). Principal component analysis (PCA) prior UMAP plotting speeds up the computational process. It is closely related with t-distributed Stochastic Neighbor Embedding (t-SNE) plot, where the main difference is that the UMAP plot provides the ability to interpret the relation between the clusters based on their distance in the plot, thus how far the cells differ from each other. Both UMAP and t-SNE plot can reveal cellular populations with unique features, ameliorating the identification of new cell types, gene markers etc.
Author: Filippos Charitidis (ESR9)
