I have rna seq data , I trying to make heatmap but there are certain values which are quite low the lower range goes to some extent ,meanwhile the upper range is reasonable .So im trying to scale the data.Now my question if im scaling the data would it preserve the true biological meaning because when i plot scaled data vs the data thats not scaled i see a quite a difference
Any suggestion would be highly appreciated
•
written
by
1769mkc
1 answer
Presuming your genes are in rows and your columns are samples then scaling rows will preserve the biology within genes. If you do clustering then that will change, but that's typically less of an issue.
How you want to scale things is completely up to you.
Hi Devon! I read your comment and I am now a bit unsure of the type of scaling that I need to perform on my data. If I have genes as rows and samples in columns and the intention is to perform a clustering on samples, should scaling be done on column or on rows? I appreciate if you could explain this to me. Thanks.
Generally you want to scale things such that highly-expressed genes aren't driving the clustering, which would mean by rows. However, you can also do things like vst() in DESeq2 to put things on a more useful scale to begin with.
Thanks for the answer Devon! however, when I do gene-wise scaling on all genes will have sd=1. Thus, if I am not mistaken, significance of genes will be lost. So I am a little bit unclear what criteria for patients classification into potential subgroups of a cancer classification algorithms will use?
By the way, my gene expression data is derived from microarrays which are on log2 scale.
Then you're not clustering, you're classifying, which is completely different. You should post such things as a new question.
Log in to answer this question.
-
Scaling RNA-Seq data before clustering?
written by im •
-
Pheatmap and SD
written by Morris_Chair
-
How to do scaling in ComplexHeatmap
written by sophialovechan •
-
ngsplot heatmap scale discrepancy
written by Sadia •
-
gplots heatmap.2 scale function not generating Z-scores between -1 to +1
written by jaime alvarez •
-
heatmap.2 function from glpot issue
written by 1769mkc
-
extract dendrogram cluster from pheatmap
written by 1769mkc
-
data frame from deseq2
written by 1769mkc
-
PCR Array Heatmap - Scaling and Clustering
written by Biogear230 •
-
Differences Between Eisen Hierarchical Clustering In Bioconductor/R Versus Heatmap.2?
written by user
I am not sure if I understand your question right, although a colleague had once scaled the color range of the heatmap using a log scale. He plotted a histogram of the FPKM values to decide the color range.
P.S. I am not sure if the biological sense is retained, but I presume it should still make sense.
well i have values in order of -50 ,-60 i certainly dont want to put that in heatmap but at the same time i want to retain those differences .Im using pheatmap it doesn;t have the heatmap.2 kind thing where you can plot the histogram or density sort of thing .
RPKM or FPKM values are according to me relative in a linear scale. For example, if your values range from -50 to 500, you could make -50 relative to 0. In that case, your values on the heatmap would be 0-550, but I'm not sure if it would be considered data manipulation. I would wait for someone to reply on that.
thats true but it may be noise as well so im not sure..
Yes, that certainly is a possibility you can't rule out. I would however like to know how did you get readcounts in the negative scale, and what that means.
well those are all FPKM values so i have like 5 cell type data i think most likely they are getting highly down-regulated perhaps unless its noise