Dimensionality reduction is often used to interpret complex biological datasets. For example, gene expression profiles obtained from large numbers of single cells are reduced to two- or three-dimensional representations that reveal underlying biological structures of the data. However, current dimensionality reduction methods, such as t-distributed stochastic neighbor embedding (t-SNE), may not preserve global or local data structure, and they can be sensitive to noise or computationally expensive. In this issue, Moon et al. address all these limitations with PHATE (Potential of Heat-diffusion for Affinity-based Transition Embedding), an improved nonlinear method for dimensionality reduction.
Vallejos (2019), "Exploring a world of a thousand dimensions", Nature Biotechnology