Extended Data Fig. 5: Random approximation of the global variance score. | Nature Computational Science

Extended Data Fig. 5: Random approximation of the global variance score.

From: Dynamic visualization of high-dimensional data

Extended Data Fig. 5

Random approximation of the global variance score for different choices of the number of random neighbors used, k, and for different combinations of DR visualization methods and data. (A) UMAP visualizations with synthetic data drawn from a mixture of Gaussians (n=1000, p=50). (B) t-SNE visualizations for the same synthetic data. (C) UMAP visualizations for the mouse subventricular zone (SVZ) single-cell transcriptomics data. (D) t-SNE visualizations for the same SVZ data. Generally, the random method for computing variance scores (see Methods for details) is a good approximation of the global variance score and the quality of this approximation increases with k. Shaded region corresponds to 95% confidence interval in all panels.

Back to article page