Next-Generation Sequencing Data Analysis, 2nd Edition

Next-Generation Sequencing Data Analysis170

to the top reduced dimensions, or principal components if PCA was used

for dimension reduction. The actual rendering of the visualization from the

dimension reduced data is mostly achieved by non- linear methods t- SNE

and UMAP, which have better performance than other non- linear methods.

Although PCA as the primary dimensionality reduction method may also be

used to visualize global data structure, and as a linear approach the distances

between cells or clusters in a PCA plot provide consistent depiction of their

relationships, its disadvantage as a visualization tool is its relative lack of

power to provide sufcient separation between distinct cell types.

As a non- linear approach, t- SNE typically ...