Statistical Method scDEED Detects Dubious t-SNE and UMAP Embeddings and Optimizes Hyperparameters

scDEED assigns a reliability score to each 2D embedding to indicate how much the data point’s mid-range neighbors change in the 2D space

The article was co-authored with Christy Lee, Ph.D. student of Statistics at UCLA.

t-SNE (t-Distributed Stochastic Neighbor Embedding) and UMAP (Uniform Manifold Approximation and Projection) are non-linear dimensionality reduction techniques for visualizing high-dimensional data, particularly in the context of single-cell analysis for visualizing cell clusters. However, it is important to note …

embedding statistics thoughts-and-theory umap visualization