A Flow Artist for High-Dimensional Cellular Data

TL;DR

FlowArtist is a neural network designed to embed high-dimensional point cloud data by jointly learning the underlying vector field, improving visualization of velocity-informed structures in dynamic datasets like single-cell RNA data.

Contribution

We introduce FlowArtist, a neural network that simultaneously embeds points and learns their velocity fields, enhancing visualization of dynamic high-dimensional data.

Findings

Better separation of velocity-informed structures in embeddings

Improved visualization of single-cell RNA velocity data

Effective on toy and real datasets

Abstract

We consider the problem of embedding point cloud data sampled from an underlying manifold with an associated flow or velocity. Such data arises in many contexts where static snapshots of dynamic entities are measured, including in high-throughput biology such as single-cell transcriptomics. Existing embedding techniques either do not utilize velocity information or embed the coordinates and velocities independently, i.e., they either impose velocities on top of an existing point embedding or embed points within a prescribed vector field. Here we present FlowArtist, a neural network that embeds points while jointly learning a vector field around the points. The combination allows FlowArtist to better separate and visualize velocity-informed structures. Our results, on toy datasets and single-cell RNA velocity data, illustrate the value of utilizing coordinate and velocity information in tandem for embedding and visualizing high-dimensional data.