Dear team of the fantastic mixOmics-package,
I am using the cimDiablo feature on a block.splsda object related to single cell molecular biology.
CimDiablo returns a color key range from -3 to 3 and I can inspect the cimDiablo matrix using cor.mat or cor.mat$mat.
However, I do not understand what this correlation matrix or the color key show exactly / mathematically.
It can’t be a Pearson correlation coefficient, as the range would only be -1 to 1. It doesn’t seem to be the weight or loadings from the block.splsda object either, as these are component-specific numbers, e.g. more than one number per measurement.
So how can I accurately describe the nature of these correlation values in cimDiablo and the color-key in our upcoming paper?
Best, Bernd
The values shown in cimDiablo() output (and therefore described by the colour key) refer to the number of standard deviations away from the mean of a feature a given sample is. By default, these are “trimmed” to within 3 standard deviations so the data’s distribution can be seen. I’m not sure what the “cor mat” you are referring to is. cimDiablo() is not depicting correlations. It is depicting the values of each sample compared across all the inputted blocks. That is why the standard deviation of each column is trimmed, to prevent outliers causing the colour key to be skewed.
Thank you Max, this is very helpdul.
Best , Bernd