Webinar: High Dimensional Flow Cytometric Characterization of Complex Tissues With Infinity Flow

This webinar will provide basic training in the implementation of Infinity Flow. Infinity Flow combines sophisticated machine learning approaches with parallelized flow cytometric staining to profile the coexpression patterns of hundreds of markers in tandem at single-cell resolution in complex tissue samples across both mouse and human specimens. Dr. Mark Headley of the Fred Hutchinson Cancer Research Center will provide background on the basic principles of Infinity Flow as well as its application to understanding complex biology in human and mouse specimens.

What you will learn:

1. Basic principles of high dimensional flow cytometry analysis via Infinity Flow
2. Proper design of a backbone panel to define cell populations in a complex tissue sample
3. Application and downstream analysis of Infinity Flow Data.

Additional Q&A With Dr. Mark Headley

Due to great interest in this talk, many submitted questions weren’t answered during the webinar. Dr. Headley graciously answered these questions, below.

1. Does Infinity Flow implement the preprocessing (debris, doublets, dead cells, using transformation, compensation, etc.)? Or does this need to be done manually before putting the data into Infinity Flow?
   
   This needs to be done manually - since there is still a good amount of subjectiveness in singlet/live/dead gating, it’s important for the researcher to make those calls. We then export the final gate (live singlets) out of FlowJo, and those become the files to use in the Infinity Flow pipeline.
    
2. Does Infinity Flow work for lower numbers of backbone markers? Is FSC and SSC data useful?
   
   Definitely works with lower numbers of backbone markers. Basically, the way to think about it is if your populations are well defined, the imputation framework will work well on them. So lower numbers of backbone markers would be better for a more focused experiment (like just looking at T cells, for example.) Absolutely, FSC and SSC provide lots of useful data to the algorithm that can help delineate distinct populations.
    
3. So, in your experience: How many cells do I need per population to get meaningful subsetting? How many steady state lungs did you run for your immunoprofiling?
   
   15 steady-state lungs were used in that experiment. But in retrospect, this is overkill based on what we now know. We stained 1 x 10⁶ cells per well, but you can easily get away with 5 x 10⁵ per well or fewer, depending on what you're trying to accomplish.
    
4. What would help getting the needed populations effectively, since the number of cells is really low in Infinity Flow?
   
   Not sure what you're asking here? The number of cells in Infinity Flow need not be really low - though depending on how you run your experiment, it can be. Basically you just need to ensure that there are 50-100 of the population you care about in the final analysis, and the algorithm will work very well. Thus, depending on what you are trying to accomplish, you may need to either collect more events, perform some sort of pre-enrichment, etc... if you're really trying to assess something very rare.
    
5. Does this method work with intracellular markers such as transcription factors? Staining for these can vary depending on temperature, etc. Thank you.
   
   Not as yet.
    
6. Have you considered staining enough cells to have some be sorted in the wells you’re interested in? And if you have seen that some wells have PE markers on unique subpopulations, can you consider pulling them and sorting?
   
   This would be very tough to do as part of the initial experiment, as the Infinity Flow pipeline is not run dynamically, and the time it takes to acquire all the samples, process, and find something interesting would make going back to the same plate a day or two later to sort stuff out not especially practical.
    
7. How does compensation work with Infinity Flow?
   
   All compensation is done beforehand with standard methods - and then you output the pre-compensated channels into a new FCS file, which is read into the Infinity Flow pipeline.
    
8. Is color compensation between backbone and Infinity markers a concern when comparing between the wells?
   
   If you've done a good job with your panel design, staining, and comp, this should be relatively consistent feature across wells, since all of the backbones should be the same intensity, and the Infinity markers are all in the same channel. That said, since brightness does vary across Infinity markers, it can lead to compensation artifacts if your Infinity channel ends up saturating.
    
9. Is background fluor better with Infinity Flow on an Aurora?
   
   If by background fluor - you mean autofluorescence - then generally yes, the ability of spectral instruments to subtract out autofluorescence is a big boon to the protocol, as you actually not only clean up your other channels, but you end up with a new autofluorescence channel which can be used to define population heterogeneity.
    
10. Is it possible to use different experiments (i.e., different day of sample run) if you have the same backbone in them?
    
    Only if you've properly calibrated your cytometer to make sure the output is equivalent.
     
11. Are the Infinity Flow antibodies all PE of different intensities?
    
    Yes - we use the BioLegend LEGENDScreen™ Kit. The “different intensities” is because every antibody is different and every labeling reaction varies.
     
12. Great talk, thank you. Can you specify what the scales of the heatmaps generated by Infinity Flow refer to?
    
    Currently the heatmaps (coming out of the pipeline in PDF form) are a standard Brightest Signal to Dimmest Signal heatmap. Thus, if you look at the scale bar, it will vary on each heatmap. You can, of course, generate heatmaps in FlowJo, R, using plugins like Cluster Explorer with scaling as you like.
     
13. How robust are the imputation algorithms used? Should the data be validated experimentally?
    
    Very robust - but yes, one should always validate. We do this by validating things of interest that come out of the analysis. That said, you can also internally validate all empirical measurements, as the empirical data measured in each well is preserved as part of the output. So you can directly compare imputed versus measured on the cells from a given well to ensure accuracy.
     
14. Can you comment on the computational power you need for the analysis pipeline, please? Similar to scRNA, or does a normal work-station PC suffice?
    
    Depends on a few things. We've run many of these analyses on a relatively standard Macbook Pro laptop (though it will obviously take quite a bit of time to do so). If you want to get your data faster, you can take advantage of High Performance Computing resources, etc... But yes, I often run these analyses on a basic off-the-shelf Mac.
     
15. Any future direction for a GUI version of the code?
    
    Not currently.
     
16. Is there a move towards standardization of antibody clones and expected results to use in this type of analysis? Is there a body that determines what "panel"? Thinking about how to compare results between labs and experiments.
    
    This has not been approached as yet. The clones used in the Infinity panel themselves are pretty standardized if you use the same LEGENDScreen™ Kits as we do - but the backbone will always be lab specific.