Data rescaling for a more informative heatmap?

Question

[j_weld]

I have a 1001 x 42 matrix of values. Most of the values are 0, and for those non-zeros, most of them are big than 50, but the maximum value is 15474.

Answer 1

[Asaf]

I would remove uninformative columns if there are any and, most importantly, transform the numbers to log scale, it will be much easier to track changes this way.

[j_weld]

A lot better but still not very

[ATpoint]

Yes, read the link in my answer, that you know why. Heatmaps on just log2 cluster by expression level, not by difference. Try to remove uninformative genes as well.

[colindaven]

It is a lot better in my view. Try also removing severe outliers. Most middle columns are almost completely uninformative though, as far as I can see, yet dominate the heatmap.

Answer 2

[ATpoint]

Standardize the log2 expression values using the Z-score. Here is some hands-on code with data examples:

Scaling RNA-Seq data before clustering?