Visualizing Structure and Transitions for Biological Data Exploration

Abstract

With the advent of high-throughput technologies measuring high-dimensional biological data, there is a pressing need for visualization tools that reveal data structure and emergent patterns in an intuitive form. We present PHATE, an unsupervised visualization method that captures both local and global non-linear structure in data by preserving informational distance between data points. PHATE reveals a range of patterns in data including continual progressions, branches, and clusters without imposing any prior assumptions on the latent data structure. We apply PHATE on a wide variety of datasets, including CyTOF, single-cell RNA-sequencing (scRNA-seq), Hi-C, and gut microbiome data and show its effectiveness in generating new insights into the underlying systems. We apply and experimentally validate PHATE using a newly generated scRNA-seq dataset of human germ layer differentiation in which PHATE reveals a dynamic picture of the main developmental branches in unparalleled detail.