Watch our introduction to viSNE video of viSNE being run and analyzed on a public example dataset.
For more reading, visit the table of contents for articles on viSNE.
What is viSNE?
viSNE is a tool for reducing high-parameter biological data down to two dimensions, making it easy for scientists to not only visually identify interesting and rare biological subsets, but also gate single cell events across different samples. viSNE uses the Barnes-Hut implementation of the t-SNE algorithm (van der Maaten, 2008). Read more about viSNE in this publication from the Pe’er lab at Columbia University:
viSNE enables visualization of high dimensional single-cell data and reveals phenotypic heterogeneity of leukemia
A viSNE analysis can be run on a data set by logging into Premium or Enterprise Cytobank and using Experiment Navigation bar to open the viSNE menu and create a new analysis.
How do I analyze my viSNE results or export data?
Since viSNE results are two added channels within an FCS file, all the same tools to analyze FCS data can be used to analyze viSNE results.
One way to analyze viSNE results is with dot plots colored by channel. This visualization type colors each event in a biaxial dot plot by its intensity on a third channel.
To find out more about why certain populations are segregated in a viSNE map, the populations can be gated and then an Illustration can be made with those gates and all channels of interest. Histograms or heatmaps work well for this analysis. What emerges is a "fingerprint" of the expression profile of each viSNE population.
Where is my viSNE analysis?
The viSNE analysis can be found on the Experiment Summary page of the original Cytobank Experiment alongside SPADE analyses under the “Advanced Analyses” section. A viSNE analysis can be cloned and it will show up in the Inbox as a cloned experiment.
Does viSNE downsample or upsample data?
Unlike SPADE, viSNE does not do any downsampling or upsampling of the data selected for analysis. The data selected for viSNE analysis during setup are used in their entirety throughout the analysis. The only downsampling that may occur is a result of having more events in the source files than are selected for viSNE. In this case, the original files are randomly downsampled from the population(s) selected in order to meet the user-entered event count for the viSNE run.
I am not allowed to run viSNE, why?
viSNE is a premium functionality and only available on Premium and Enterprise Cytobank.
Creation of a new viSNE analysis is only accessible by users with Illustration level project access or higher to an Experiment. You may not have sufficient access to the Cytobank Experiment in question.
What is event sampling and why is it important for viSNE?
Currently, viSNE in Cytobank allows a maximum of 800,000 events per analysis. Events are sampled from gated populations in the experiment. Unlike SPADE, which uses density-dependent downsampling, viSNE uses random sampling and analysis is performed only on sampled events.
What is the difference between proportional and equal sampling?
Proportional Sampling: the algorithm samples from gated populations while preserving their relative abundance.
Equal Sampling: the algorithm samples equally from gated populations regardless of their relative abundance. The number of events sampled per population will be limited by the population with the fewest events.
What does the distance between two points on a viSNE plot mean?
The viSNE plot is driven by local probabilities of the data points instead of a universal linear distance metric. viSNE iteratively minimizes the Kullback-Leiber divergence between the joint probabilities of the low-dimensional viSNE plot and the original data. More simply -- points that are likely (probable) to be close to each other in high dimensional space are squished together on the viSNE plot. The closer events are on the viSNE plot, the closer they were in n-dimensional space prior to dimensionality reduction. Even more simply -- if events are grouped together on the viSNE map, they are similar to each other based on all the channels included in the run.
Which Population should I select while setting up a viSNE analysis?
Selecting populations for viSNE analysis is a matter of the questions being asked of the data. For open-ended analysis of all population lineages in a sample, choose a lightly gated top level population (e.g. CD45+). For viSNE analysis of a more specific lineage, choose that lineage or its direct parent (e.g. CD3+).
Which Channels should I select for my viSNE analysis?
The channels selected should contribute to the separation of the desired populations. Channels with good staining and event separation will lead to more interpretable viSNE results. Channels with little range or variety among events won’t substantially contribute to separation of events in the viSNE map. For example, if all events are negative for a marker, that channel will contribute poorly to the separation of populations in viSNE. Also, channels of different scale types (linear, asinh, etc) should generally not be mixed in the same viSNE analysis.
Do the Scale settings affect how viSNE runs on my data?
Yes, the scale settings in Cytobank influence how viSNE interprets the distribution of data, but the only setting of importance is scale argument in the case of arcsinh scales. It is good to assess the quality of data scaling before running a viSNE analysis, as results can be skewed by poor scaling. For discussion and examples of data scaling, see this useful article.
Can I run viSNE on fluorescent data?
viSNE can be run on FCS data regardless of instrument origin. With fluorescent data sets, however, proper compensation is important, and more attention may need to be paid to proper scale settings.
Which FCS files should I select for my viSNE analysis and what files are generated?
The FCS files to select for viSNE analysis depend on what is being examined in the experiment. viSNE combines all data selected for analysis and creates a new FCS file for each population-file combination. Each new file has new tSNE channels which can then be visualized in Cytobank as channels.
Why do my viSNE plots look slightly different when run on the same data set?
Although viSNE plots tend to have similar groupings of cell subsets, viSNE plots may look slightly different. t-SNE uses a non-convex objective function.
Why is there a (v) in the channel names in viSNE experiment?
A (v) appears in the channel name of viSNE files to indicate the channels that were used in the creation of the viSNE map. The channel name can be edited on the channel annotation page by clicking “setup” on the channels figure dimension in the working illustration.
Can I run viSNE or SPADE on my viSNE-generated FCS files?
Yes. To do this, click to start a new viSNE or SPADE run from the viSNE experiment. Cytobank will clone the viSNE analysis and create a cloned experiment accessible from the Inbox. SPADE or viSNE can then be executed from within this cloned experiment. It is not advised to run viSNE on the tSNE channels of an existing viSNE analysis because this yields poor results, even though one might try to do this to clean up an existing viSNE map.
Why is the symbol for viSNE a cherry?
If you Google "viSNE" and look at the image results, it should be clear. A deeper explanation of why this is the case will be left as an exercise to the reader.