no features detected

[LJK1991]

Hello,

I am trying out flowfi after a guest speaker presented it.
While trying out the Refine tab i try to load my fcs files to rank the features.
Upon doing so i get an error:

When i press 'OK'

With a bit of Google and AI i find that it might be due to FlowKit not finding the $PnN header. When i load my .fcs file in notepad++ i can see it is FCS3.2 and indeed does not have a $PnN header.
Possibly it is because the file is in FCS3.2 format and not lower (or higher) is the best solution to just export it as a FCS3.1 using the S8 software?
or should i wait for an update of FlowFi?

Here is a snapshot of the notepad++ open FCS file

Kind regards,
Lucas

[jameswilsenach]

Hi Lucas!

Thanks so much for trying out the software, it is still in an early pre-release stage for some of the newer features so your testing of the software is invaluable. In my experience, FlowKit is a great project but it has not been kept up to date with new FCS versions.

My advice would be that if you can export your parameters in a CSV (tabular or excel compatible format), I find this is generally less liable to versioning issues. This should be possible within the BD ecosystem if you have FlowJo.

Thanks so much for inspecting the file and showing your results. I will follow this up, I suspect there may be a way to default the software to an alternative field if it is unable to find either of the $PnN fields, but this may take some time.

In the meantime, if you can export as a CSV I would recommend doing this.

[LJK1991]

exporting to csv works. Just a usage question, There is the option to turn of particular parts of the file for ranking. I see that it loads in every bandpass filter for every lazer channel as well as the unmixed signal and the imaging paramters. As i am largely looking for information on the imaging parameters is it fair/ok to not use the other information. For the spectral information or better said the data after unmixxing i kind of know what i want to be looking at. But maybe I should be using the other information. What are your suggestions?

Additionally, do i need to perform this on the entire fcs file prior to debree gating for example or should i do this on subset of datapoints e.g. specific gated populations? to me the latter sounds more logical.

[jameswilsenach]

Hi Lukas!

This is actually a subtle point and somewhat depends.

The current importance metric relies on something called Laplace Scoring (more measures are to follow in the next release), this is a measure that constructs a nearest neighbour graph based on the supplied parameters and then calculates how important each parameter is to the graph's overall structure. If the goal is to see which parameters are most influential within the imaging set then it's perfectly valid to calculate importance scores based only on these but you will be looking only at the image parameter's internal; structure. This will certainly provide a speed up, it's important to note though that the scores will be different if you had instead used the entire dataset (including spectral parameters), so it depends on your goals. In my experience it is good to narrow the set down to those you are potentially interested in using.

On the debris question, in general I would say remove these prior to running the refine algorithm. All machine learning methods benefit from a higher signal to noise ratio and by removing the debris you are looking more at the structure of data and less at the structure of the noise which may form clusters (e.g. a cluster of platelet events with similar DNA size ~ 0 and autofluorescence size ~ small) you are disinterested in.

MK cells are highly morphologically heterogeneous which can make reliable purification really tricky! If you want to see which parameters shifted most in importance after debris gating, it is possible to do this by loading an output CSV in FlowFI. This is an experimental feature and is likely to return the parameters you already gated on, but may be useful for hypothesis generation, to see which other parameters are more important when debris is present (the raw ungated sample) vs when it isn't (the debris gated sample). You could try gating on some of these if the purity level (e.g. when inspecting in CellView) is still too low.

I hope this helps!