CyTOForum

viSNE is a recently published alternative to SPADE for high-dimensional data analysis [1]. Unlike SPADE, individual cells (not clusters of cells) are displayed on a 2-dimensional map which preserves the multi-dimensional separation of the cells. In this sense, it is somewhat similar to PCA analysis. Like SPADE, viSNE allows coloration of cells by marker intensity or fold-change. The original implementation of viSNE was limited to 10,000 events per display; the new algorithm is approximately 10-fold faster and thus can support up to 100,000 events. It is available from http://www.c2b2.columbia.edu/danapeerlab/html/cyt.html.

Reference

1. Amir, E.-A. D., Davis, K. L., Tadmor, M. D., Simonds, E. F., Levine, J. H., Bendall, S. C., et al. (2013). viSNE enables visualization of high dimensional single-cell data and reveals phenotypic heterogeneity of leukemia. Nature Biotechnology. doi:10.1038/nbt.2594

I was interested in knowing whether people here prefer using SPADE or viSNE for high-dimensional data analysis.
I started using mostly viSNE, but does SPADE have any major advantages that I should be aware of?
SPADE is perhaps better-known and already used by flow cytometry people, but what can you say about the advantages of each method?

The two approaches are different.

viSNE gives more reproducible "island" shapes from run to run. With SPADE, you get more variation in "tree" structure from run to run. As I understand, this is primarily due to the downsampling.

In my limited experience with viSNE, I think it *can* do a better job of finding low-frequency cell populations that might differ by only a couple markers out of 30+ clustering markers. For instance, I was better able to narrow down on pDCs vs mDCs in viSNE. In my (rather T-cell focused) panel, SPADE often lumps these together because they're low frequency and are highly-similar in 31 of the 33 markers I used in the tree (note: basophils are usually easily resolved from DCs, due to the differing expression of HLA-DR).


However, viSNE is mainly used for visualization; it's not clear to me from the paper or from playing with it how you can extract "numbers" (cell count/frequency, median marker intensities, fold-changes, etc) from it like you can using SPADE.