Data transformation/normalisation for heatmaps

[Jengel44]

Hi.
I'm wondering about how data is transformed/normalised for the heatmap plotting? I have asinh_transformed MFI values ranging from -8 to 10 and when plotted on the heatmap I get a scale from 0-1. Is the mean MFI first calculated in the data group and then normalised? And how is it normalised?
Thanks so much!

[tomashhurst]

Hi @Jengel44 ,

Great question! If I have understood your question correctly, you are talking about plotting aggregate expression data on a heatmap -- i.e. the median expression of each marker on each cluster/population.

From the last bit of this section: https://immunedynamics.io/spectre/simple-discovery/#5_Clustering_and_DR

The do.aggregate function does the 'aggregation', so the resulting table is rows = clusters, columns = markers, and the values are the median expression of each marker on each cluster.

exp <- do.aggregate(cell.dat, cellular.cols, by = "FlowSOM_metacluster")

Then the make.pheatmap function has an argument which 'normalises' the values in each column (marker) between 0 and 1. This is more accurately referred to as rescaling.

make.pheatmap(exp, "FlowSOM_metacluster", cellular.cols)

This means they all have the same colour range for what is effectively minimum to maximum for each marker. So if you had a range of -8 to 10, this would then be 'rescaled' to a range of 0 and 1.

normalise | DEFAULT = TRUE. Logical. Only applies to standard heatmaps (i.e. when is.fold = FALSE). TRUE to normalise each column between 0 and 1, FALSE to plot the raw values.

If you wanted to turn this feature OFF:

make.pheatmap(exp, "FlowSOM_metacluster", cellular.cols, normalise = FALSE)

[Jengel44]

Ok thanks so much for clarifying this for me. Great explanation for an R/spectre newbie!