Cluster data in heat map in R ggplot

Question

Please see my plot below: enter image description here

my code:

 > head(data)
              X0      X1      X2       X3       X4       X5       X6        X7        X8        X9
 NM_001001144 6.52334 9.75243 5.62914 6.833650 6.789850 7.421440 8.675330 12.117600 11.551500  7.676900
 NM_001001327 1.89826 3.74708 1.48213 0.590923 2.915120 4.052600 0.758997  3.653680  1.931400  2.487570
 NM_001002267 1.70346 2.72858 2.10879 1.898050 3.063480 4.435810 7.499640  5.038870 11.128700 22.016500
 NM_001003717 6.02279 7.46547 7.39593 7.344080 4.568470 3.347250 2.230450  3.598560  2.470390  4.184450
 NM_001003920 1.06842 1.11961 1.38981 1.054000 0.833823 0.866511 0.795384  0.980946  0.731532  0.949049
 NM_001003953 7.50832 7.13316 4.10741 5.327390 2.311230 1.023050 2.573220  1.883740  3.215150  2.483410

pd <- as.data.frame(scale(t(data)))
pd$Time <- sub("_.*", "", rownames(pd))
pd.m <- melt(pd)
pd.m$variable <- as.numeric(factor(pd.m$variable, levels =     rev(as.character(unique(pd.m$variable))), ordered=F))
p <- ggplot(pd.m, aes(Time, variable))
p  + geom_tile(aes(fill = value)) + scale_fill_gradient2(low=muted("blue"), high=muted("red")) +
  scale_x_discrete(labels=c("0h", "0.25h", "0.5h","1h","2h","3h","6h","12h","24h","48h")) + 
   theme_bw(base_size=20) + theme(axis.text.x=element_text(angle=0, vjust=0.5, hjust=0, size=12),
   axis.text.y=element_text(size=12), strip.text.y=element_text(angle=0, vjust=0.5, hjust=0.5, size=12),
   strip.text.x=element_text(size=12)) + labs(y="Genes", x="Time (h)", fill="")

Is there a way to cluster the plot so that the plot displays the dynamics in the time course. I would like to use the clustering that comes out of:

 hc.cols <- hclust(dist(t(data)))

enter image description here

Just fyi, I think it's frowned upon on SO to add "solved" to the title of your question. I removed it. — Rich Scriven, Aug 27 '14 at 15:29

Beasterfield · Accepted Answer · 2014-08-27T14:03:06.963

15

You can achieve this by defining the order of Timepoints in a dendrogram after you have applied hclust to your data:

data <- scale(t(data))
ord <- hclust( dist(data, method = "euclidean"), method = "ward.D" )$order
ord
[1]  2  3  1  4  8  5  6 10  7  9

The only thing you have to do then is transforming your Time-column to a factor where the factor levels are ordered by ord:

pd <- as.data.frame( data )
pd$Time <- sub("_.*", "", rownames(pd))
pd.m <- melt( pd, id.vars = "Time", variable.name = "Gene" )

pd.m$Gene <- factor( pd.m$Gene, levels = colnames(data), labels = seq_along( colnames(data) ) )
pd.m$Time <- factor( pd.m$Time, levels = rownames(data)[ord],  labels = c("0h", "0.25h", "0.5h","1h","2h","3h","6h","12h","24h","48h") )

The rest is done by ggplot automatically:

ggplot( pd.m, aes(Time, Gene) ) +
  geom_tile(aes(fill = value)) +
  scale_fill_gradient2(low=muted("blue"), high=muted("red"))

enter image description here

edited Aug 27 '14 at 14:03

answered Aug 27 '14 at 13:46

Beasterfield

7,023
2
38
47

awesome, exactly what I wonted. Looked great on the whole dataset. – user3741035 Aug 27 '14 at 14:03
Sorry, in the first version the data where clustered by genes. This is fixed now. But notice, that the time axis is now out of order of course. So maybe you want to cluster by gene instead of clustering by time. Unless you are not expecting recurring effects over time, this would also make more sense to me. – Beasterfield Aug 27 '14 at 14:05
I actually want the version where it is clustered on time. – user3741035 Aug 27 '14 at 14:11
Sorry for bothering again. How would you go about to add row names in the plot? – user3741035 Aug 29 '14 at 19:59
Since you are assigning numbers to the gene names they appear like 1,2,3 and so on. – user3741035 Aug 29 '14 at 21:08
@user3741035 Oh, just leave this step `pd.m$Gene <- factor( pd.m$Gene, levels = colnames(data), labels = seq_along( colnames(data) ) )` out. – Beasterfield Sep 01 '14 at 07:46
Hmm. If I leave it out it does not cluster. – user3741035 Sep 01 '14 at 17:20
Maybe because "genes" is not numberic? – user3741035 Sep 01 '14 at 17:30
1

@user3741035 are you clustering now by timepoints or genes? Leave the line, I told you to remove, but remove there the label assignment: `pd.m$Gene <- factor( pd.m$Gene, levels = colnames(data)[ord])`. – Beasterfield Sep 01 '14 at 17:40
Is there a way to cluster based on time, but at the same time keep the order of the time points? Do I need to split the dataset then? – user3741035 Sep 02 '14 at 06:15
How? Clustering == changing order by similarity. But feel free to make a proper example and ask this as a separate question. – Beasterfield Sep 02 '14 at 06:41
I have asked a separate question. Maybe my data is the problem. – user3741035 Sep 02 '14 at 07:42

BrodieG · Answer 2 · 2014-08-27T13:39:18.387

3

I don't think ggplot supports this out of the box, but you can use heatmap:

 heatmap(
   as.matrix(dat), Rowv=NA,
   Colv=as.dendrogram(hclust(dist(t(as.matrix(dat)))))
 )

enter image description here

Note this won't look like yours because I'm just using the head of your data, not the whole thing.

Here we specify the clustering manually with a dendogram derived from your hclust with the Colv argument. You can specify the clustering manually too through the Colv argument if the one used by default doesn't line up with what you want.

edited Aug 27 '14 at 13:39

answered Aug 27 '14 at 13:29

BrodieG

51,669
9
93
146

How do I set Rowv like the dendrogram above. – user3741035 Aug 27 '14 at 13:32
Try: `heatmap(as.matrix(dat), Rowv=NA, Colv=as.dendrogram(hclust(dist(t(as.matrix(dat))))))` – BrodieG Aug 27 '14 at 13:39
@user3741035, also, note I meant `Colv`, not `Rowv` – BrodieG Aug 27 '14 at 13:39

score 1 · Answer 3 · answered Jan 03 '19 at 22:37

Thought I'd add you don't need to transform the columns in the data.frame to factors, you can use ggplot's scale_*_discrete function to set the plotting order of axes. Simply set the plotting order using the limits argument and the labels using the labels argument as shown below.

data<-read.table(text="X0      X1      X2       X3       X4       X5       X6        X7        X8        X9
 NM_001001144 6.52334 9.75243 5.62914 6.833650 6.789850 7.421440 8.675330 12.117600 11.551500  7.676900
 NM_001001327 1.89826 3.74708 1.48213 0.590923 2.915120 4.052600 0.758997  3.653680  1.931400  2.487570
 NM_001002267 1.70346 2.72858 2.10879 1.898050 3.063480 4.435810 7.499640  5.038870 11.128700 22.016500
 NM_001003717 6.02279 7.46547 7.39593 7.344080 4.568470 3.347250 2.230450  3.598560  2.470390  4.184450
 NM_001003920 1.06842 1.11961 1.38981 1.054000 0.833823 0.866511 0.795384  0.980946  0.731532  0.949049
 NM_001003953 7.50832 7.13316 4.10741 5.327390 2.311230 1.023050 2.573220  1.883740  3.215150  2.483410", header = TRUE, stringsAsFactors = FALSE)
data <- scale(t(data))
ord <- hclust( dist(data, method = "euclidean"), method = "ward.D" )$order
pd <- as.data.frame( data )
pd$Time <- sub("_.*", "", rownames(pd))
pd.m <- melt( pd, id.vars = "Time", variable.name = "Gene" )
ggplot( pd.m, aes(Time, Gene) ) +
  geom_tile(aes(fill = value)) +
  scale_x_discrete(limits=pd.m$Time[ord], labels = c("0h", "0.25h", "0.5h","1h","2h","3h","6h","12h","24h","48h"))+
  scale_y_discrete(limits=colnames(data), labels = seq_along(colnames(data)))+
  scale_fill_gradient2(low=muted("blue"), high=muted("red"))

Cluster data in heat map in R ggplot

3 Answers3

Linked