How to Use t-SNE Effectively

Abstract

We present a new technique called "t-SNE" that visualizes high-dimensional data by giving each datapoint a location in a two or three-dimensional map. The technique is a variation of Stochastic Neighbor Embedding (Hinton and Roweis, 2002) that is much easier to optimize, and produces significantly better visualizations by reducing the tendency to crowd points together in the center of the map. t-SNE is better than existing techniques at creating a single map that reveals structure at many different scales. This is particularly important for high-dimensional data that lie on several different, but related, low-dimensional manifolds, such as images of objects from multiple classes seen from multiple viewpoints. For visualizing the structure of very large data sets, we show how t-SNE can use random walks on neighborhood graphs to allow the implicit structure of all of the data to influence the way in which a subset of the data is displayed. We illustrate the performance of t-SNE on a wide variety of data sets and compare it with many other non-parametric visualization techniques, including Sammon mapping, Isomap, and Locally Linear Embedding. The visualizations produced by t-SNE are significantly better than those produced by the other techniques on almost all of the data sets.

Current best practices in single‐cell RNA‐seq analysis: a tutorial

“…Thus, these visualizations may exaggerate differences between cell populations and overlook potential connections between these populations. A further difficulty is the choice of its perplexity parameter, as t‐SNE graphs may show strongly different numbers of clusters depending on its value (Wattenberg et al , ). Common alternatives to t‐SNE are the Uniform Approximation and Projection method (UMAP; preprint: McInnes & Healy, ) or graph‐based tools such as SPRING (Weinreb et al , ).…”

Visualization of Biomedical Data

“…(c) t-SNE (126) provides more insight, revealing local similarities as well as overall variation in the data set. However, t-SNE can be more difficult to apply, as it requires setting a manually adjustable parameter (perplexity) (127). (d ) Diffusion maps (128,129) model relationships between points in the data set as a diffusion process that is then reduced to a lower-dimensional map.…”

Machine learning applications in cell image analysis

“…Results from such methods are directly interpreted by researchers in the context of biological questions being targeted. Although detailed technical knowledge of a method used is usually not required, it is important to understand the consequence of assumptions and parameters used 21 , 22 …”

“…This includes choosing a particular method, as many methods often can be used to address the same problem. Moreover, not all parameters can be automatically ‘learned’ from data, and the choice of parameters can dramatically affect results 21 , 22 …”