Problem with Helios normalizer
Posted: Tue Aug 02, 2016 5:17 am
Hi everyone,
We recently discovered a weird issue with the Helios normalizer that I thought everyone should be aware of: when normalizing and concatenating multiple files together, the Pd102 channel does not get normalized. This is illustrated in the example below, in which a large barcoded sample was collected as 3 separate acquisitions (on a CyTOF2). The signal distribution of each of the barcode channels is shown in the first three rows. The next 3 rows show the signal distribution when the files were normalized individually, resulting in a characteristic signal boost that we see with our CyTOF data post normalization. All this looks fine. The the 5th row shows the result of concatenation of the files without normalization, and the 6th row shows simultaneous concatenation and normalization (i.e., both boxes checked in the software).
The 6th scenarios is where the issue manifests (highlighted with a red arrow); the post normalization distribution of Pd102 is identical to the pre-normalization, whereas all the other channels are normalized to the level seen when the files are normalized individually.
This causes issues when attempting to debarcode the files. As you can see, when samples are barcoded well, the barcode separation threshold is fairly even for all the files. However, after normalization/concatenation you end up with this bimodal distribution, where samples 1-10 (which all use the Pd102 label) have a much lower separation than 11-20. We've found that to effectively debarcode the normalized data to maximize recovery and minimize doublet contamination you have to run the software twice with two different separation thresholds and select one set of files from one and another set of files from the other (which is a pain).
Turns out that doing a 2 step process where you normalizing the files first and then concatenating with the fluidigm software doesn't work (doing this in fact removes the normalization from the data, which I knew was an issue with version 1 of the CyTOF2 software and I had assumed would have been fixed by now but apparently not). Concatenating first and then normalizing results in the same issue of Pd102 not being normalized.
The best solution I've found so far is to normalize the individual files and then run the Cytobank concatenator tool, which preserves the signal like it's supposed to (doing the reciprocal doesn't work and brings up an error message when you try to normalize).
Unfortunately, this largely eliminates one of the main reasons I use the Fluidigm normalizer in favor of Rachel Fink's one, which is that it's a bit faster and more convenient to be able to simultaneously normalize and concatenate a big batch of files.
Adeeb
We recently discovered a weird issue with the Helios normalizer that I thought everyone should be aware of: when normalizing and concatenating multiple files together, the Pd102 channel does not get normalized. This is illustrated in the example below, in which a large barcoded sample was collected as 3 separate acquisitions (on a CyTOF2). The signal distribution of each of the barcode channels is shown in the first three rows. The next 3 rows show the signal distribution when the files were normalized individually, resulting in a characteristic signal boost that we see with our CyTOF data post normalization. All this looks fine. The the 5th row shows the result of concatenation of the files without normalization, and the 6th row shows simultaneous concatenation and normalization (i.e., both boxes checked in the software).
The 6th scenarios is where the issue manifests (highlighted with a red arrow); the post normalization distribution of Pd102 is identical to the pre-normalization, whereas all the other channels are normalized to the level seen when the files are normalized individually.
This causes issues when attempting to debarcode the files. As you can see, when samples are barcoded well, the barcode separation threshold is fairly even for all the files. However, after normalization/concatenation you end up with this bimodal distribution, where samples 1-10 (which all use the Pd102 label) have a much lower separation than 11-20. We've found that to effectively debarcode the normalized data to maximize recovery and minimize doublet contamination you have to run the software twice with two different separation thresholds and select one set of files from one and another set of files from the other (which is a pain).
Turns out that doing a 2 step process where you normalizing the files first and then concatenating with the fluidigm software doesn't work (doing this in fact removes the normalization from the data, which I knew was an issue with version 1 of the CyTOF2 software and I had assumed would have been fixed by now but apparently not). Concatenating first and then normalizing results in the same issue of Pd102 not being normalized.
The best solution I've found so far is to normalize the individual files and then run the Cytobank concatenator tool, which preserves the signal like it's supposed to (doing the reciprocal doesn't work and brings up an error message when you try to normalize).
Unfortunately, this largely eliminates one of the main reasons I use the Fluidigm normalizer in favor of Rachel Fink's one, which is that it's a bit faster and more convenient to be able to simultaneously normalize and concatenate a big batch of files.
Adeeb