Calculating Bead Median question

[laderast]

Hi,

I am trying to port the Matlab bead Normalizer code to R.

The one thing I’m struggling with is the setNormalizationBaseline() function in normalize_folder.m. I understand that bead_means is a matrix of the median values for a the beadsbead across samples versus time.

What I’m unsure of (and this might just be my Matlab inexpertise) is how the median value is calculated across samples, since the number of timepoints is different for each sample. For example, is the median value for the first timepoint for Bead1(La139)Di in the example data calculated by just taking the first value for each of the four samples and finding the median?

In figure 2C in the paper, this appears to be the case, but I found the wording confusing because timepoint can refer to within-sample (as in timepoint at which a bead value was collected), or across samples (which correspond to different days).

[rfinck]

Great to hear that you're working on an R version!

The median that is calculated is the global median per isotope over all identified beads, so the result is one value per isotope (i.e. a 1 x n vector where n = 5 for the commercially available beads). This is calculated from a matrix that has concatenated the beads from each individual file. That does weight samples that have more timepoints more heavily, which was an intentional choice so that a small sample with high/low values does not over-contribute. However, this particular choice of baseline is indeed a heuristic and could just as easily be mean, or mean of per-file medians, or independently calibrated values such as are used by the CyTOF2 normalization software and which you can emulate by using an FCS file of gated beads to calculate your baseline.

Your question caused me to look at the compute_bead_slopes function in norm_cytof, and I believe I found a bug that was harmless as long as you were using normalize_folder to run the normalizer. If a baseline wasn't specified, it was computing a default baseline poorly. That functionality has now been removed.

Please let me know if any of this is not clear or if you have any more questions.

[laderast]

Thanks for the detailed reply. I found it really helpful.

For the R-package, I will probably implement it as a mean of per-file medians, since it will be easier to calculate, rather than concatenating all of the bead files to calculate the medians.

I'll keep you updated on the status of the package.

Thanks,
Ted