Control Pheatmap cluster_col

Question

[lgspeight]

I have two columns or variables. When I use cluster_col = T, my pheatmap is difficult to interpret trying to distinguish salinity and population. When I use cluster_col = F, pheatmap automatically clusters population first, then salinity. I want it to first cluster salinity and within each salinity cluster each population. How can I do this? I have already tried rearranging the order of population and salinity.

df <- data.frame(colData(dds)[,c("salinity","population")])
rownames(df) <- colnames(dds)
colnames(df) <- c("salinity","population")

pheatmap(
  assay(vsd)[lfcorder[1:10],], 
  cluster_rows=F, 
  show_rownames=T,
  cluster_cols=T,
  annotation_col=df
)

pheatmap(
  assay(vsd)[lfcorder[1:10],], 
  cluster_rows=F, 
  show_rownames=T,
  cluster_cols=F,
  annotation_col=df
)

Answer 1

[Ram]

Have you tried rearranging the columns of assay(vsd) so they're in the desired order? You'll need to sort df, pick its rownames and pass it to the assay(vsd) like so:

df <- df[order(....)]
pheatmap(
  assay(vsd)[lfcorder[1:10],rownames(df)], 
  cluster_rows=F, 
  show_rownames=T,
  cluster_cols=F,
  annotation_col=df
)

EDIT: You can also sort using dplyr's arrange instead of base R's order (as you've mentioned in your comment below)

[lgspeight]

Thanks Ram!

I ended up using

df <- df %>% arrange(desc(salinity))

I think this is the same idea you were talking about.

[Ram]

Yep - dplyr vs base R for sorting is the difference between our approaches. Glad things worked out!

Given it was the same concept, can you please accept my answer to mark the post resolved? I'll edit it and add a statement about how dplyr can be used as well.