Normalization and batch correction

[Cicevk]

Hi,

Thank you for the package.
A reviewer asked a question saying that 'Background correction' is mentioned without a clear description of normalisation or batch correction.
Could you comment on whether normalization or batch correction is incorporated in the code? If not, why is this not necessary? I understand that people draw parallels between single-cell RNA-seq analysis and this high-throughput protein analysis.

Many thanks for your help.

Best regards,

[Cicevk]

P.S. From what I have read, I would guess that there should be no batch correction in the code since the entire analysis is a single batch. If more batches were to be combined, the code would probably have to be expanded to meet these requirements. I only used the basic InfinityFlow function for one run of the 253-PE antibody analysis.

[ebecht]

Hi Cicevk,

The "background correction" only covers autofluorescence / unspecific antibody substraction, it will not perform batch correction or any kind of normalization. Actually it would likely make different batches less comparable so if you're analyzing multiple samples acquired in distinct experiments I would suggest using the predictions of the raw intensities instead.

As for the necessity of batch correction, I think that if the backbone panel measurements of the distinct samples somewhat overlap on a UMAP, your data shows minimal batch effects and should be comparable across samples.

Another way to investigate this would be to look at the histograms for each marker and make sure that the modes of the distributions are not shifted across batches.

Hopefully that helps, good luck with the reviewing process
Best,
Etienne

Edit: Sorry I missed the second post, with a single experiment it is fine to use background corrected data (which agains targets aufluorescence / non-specific antibody binding). In this setting there's usually no need for batch correction since all of the wells are acquired in one session. We actually do perform a small normalization to correct for potential inter-well / plates variations (per channel Z-scoring of each backbone marker).

[Cicevk]

Hi ebecht,

Many thanks for the explanation. I will provide flow graphs, as a validation of my analysis in addition.

Additionally, it would also help me to know, if on clustering except 253-PE antibodies, also backbone composition has an influence. My intuition would say that backbone would also influence clustering. This is something I was wondering for a while, but had only one data set to test that. Does it mean that if I would enrich backbone with markers that can in combination distinguish 4 or 5 populations, would that increase the number of clusters I see in the InfinityFlow analysis?

Many thanks and have a great Friday.
Best,
Vladislava

[ebecht]

Hi,
Sorry I had missed that question. Yes your intuition is correct.
Best
Etienne