![]() Here is a comparison of a B6 replicate analyzed by tSNE and UMAP in FlowJo. I anticipate that UMAP preforms much better in Python or R. I hypothesize that this failure may be due to FlowJo and not UMAP, as FlowJo is prone to crashing in my personal experience, even when analyzing simple flow cytometry data. However, when I concatenated all of the events together and attempted a UMAP run (~1.2 million events x 35 parameters) it analyzed the data for about 2 hours and ended up crashing. For comparison, a tSNE analysis of 10,000 events in FlowJo took about the same amount of time, however 266,644 events crashed the program, indicating that UMAP truly does preform better than tSNE with higher event numbers. I clocked the analysis times and 9,141 events x 35 parameters took less than 1 min to complete, and 266,644 events x 35 parameters took about 10 minutes. the CD4 T cells are are the 2 o clock position), but the abundance of events in each island, and the expression of various markers, will differ between samples. This is essential for comparison between samples as the geography of each tSNE plot will be identical (e.g. After clustering is finished you can visualize all of the input events on the tSNE plot, or select each individual sample. B6 #1-5 IL10KO #1-4), determine an input methods (equal events, all events, etc.) and then select markers for clustering. 9 FCS files total), select all of the files for clustering (e.g. For instance, for viSNE (used in Cytobank) or tSNE in Cytosplore you upload each FCS file (e.g. The main issue is that there is no expedient way to run an analysis on all of the samples at once. The UMAP plugin has very similar issues to the tSNE plugin that I discussed recently in the following post. While the developers work to resolve this we thought we’d try out the new UMAP FlowJo plugin. I, and other users, have reached out to the developers and to the community in an attempt to rectify this issue, but have had limited success. However, as mentioned in a previous post, the new program has a bug in the data transformation step which makes it impossible to use currently. ![]() Recently, Cytofkit announced they were including UMAP in the new version of the pipeline ( Cytofkit2), since we are avid users we explored this avenue first. tSNE and PhenoGraph have been sufficient for our analysis needs, so we’ve been waiting for a version of UMAP that would be more accessible to us and other bench scientists. Since most of the analysis algorithms we use are based in R, Java, or MATLAB, I’ve never tried Python and it would require some dedication to become comfortable with. Traditionally UMAP has been run through Python. However, finding a platform to use UMAP has been a challenge. UMAP is very similar to tSNE, however it allows the analysis of many more events in a shorter amount of time (for a detailed comparison of UMAP and tSNE, check out this publication: biorxiv/ Nature Biotechnology). UMAP is only about a year old, but it has become increasingly popular in the field. Eric and I have been very eager to upgrade to UMAP (as opposed to tSNE) as our go to dimensionality reduction tool for single-cell data. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |