3
$\begingroup$

I returned a FeaturePlot from Seurat to ggplot. My plot has a weird range of colours as below

enter image description here

I produced this plot by this code

> head(mat[1:4,1:4])
             s1.1        s1.2 s1.3       s1.4
DDB_G0267178    0 0.009263254    0 0.01286397
DDB_G0267180    0 0.000000000    0 0.00000000
DDB_G0267182    0 0.000000000    0 0.03810585
DDB_G0267184    0 0.000000000    0 0.00000000
> 

I have converted expression matrix to a binary matrix by 2 as a threshold

mat[mat < 2] <- 0
mat[mat > 2] <- 1

> head(exp[1:4,1:4])
             s1.1 s1.2 s1.3 s1.4
DDB_G0267382    0    0    0    1
DDB_G0267438    0    0    0    1
DDB_G0267466    0    0    0    0
DDB_G0267476    0    0    1    0
> 
> exp=colSums(exp)
> exp=as.matrix(exp)
> colnames(exp)="value"
> exp=as.data.frame(exp)
> cc <- AddMetaData(object = seurat, metadata = exp)
> cc <- SetAllIdent(object = cc, id = "value")
> TSNEPlot(object = cc, do.return= TRUE)

How I can convert this range to a gradient of colours for example in 8-18, 18-28, 28-38, 38-48 range in blue to yellow please? Something like below

enter image description here

Thank you for any help

Then by ggplot now I scaled my colours but I don't like my clusters as so and I don't know how to retain my clusters as a featureplot by this new color gradient

> head(cc@meta.data)
     nGene    nUMI    orig.ident res.0.7 CELL STAGE GENO dataset stage.nice celltype value
s1.1  4331  373762 SeuratProject       0 s1.1   H16   WT       1        H16        0    34
s1.2  5603 1074639 SeuratProject       0 s1.2   H16   WT       1        H16        0    26
s1.3  2064   49544 SeuratProject       0 s1.3   H16   WT       1        H16        0    27
s1.4  4680  772399 SeuratProject       1 s1.4   H16   WT       1        H16        1    29
s1.5  3876  272356 SeuratProject       1 s1.5   H16   WT       1        H16        1    21
s1.6  2557  122314 SeuratProject       0 s1.6   H16   WT       1        H16        0    31

> ggplot(as.data.frame(cc@meta.data), aes(x = cc@meta.data$CELL, y = cc@meta.data$res.0.7, colour =cc@meta.data$value)) + 
+     geom_point(size = 5) +
+     scale_colour_gradient(low = "yellow", high = "blue")

enter image description here

By below code I obtained a tsne in link

> cols <-  scales::seq_gradient_pal(low="beige", high="red", space="Lab")(seq(from=0, to=1,length.out=48))
> 
> TSNEPlot(cc, colors.use=cols)

enter link description here

Now I want to know how I could convert this range to a 8-18, 18-28, 28-38, 38-48 colour range as a gradient of blue to yellow?

$\endgroup$
  • 1
    $\begingroup$ Can you share the code used to generate the first plot? Essentially what you put in to the call to FeaturePlot(). $\endgroup$ – Kohl Kinning Oct 29 '18 at 19:41
  • 1
    $\begingroup$ Looks like your value column is not treated as an integer but as a factor. $\endgroup$ – Peter Menzel Oct 29 '18 at 21:09
  • 1
    $\begingroup$ Coloring aside, what are trying to accomplish with >colnames(exp)="value"? Unless I'm missing something you are trying to assign one string to a vector of values. This shouldn't work. You should be getting an error like length of 'dimnames' [2] not equal to array extent. Do you ultimately want a plot of the values in one column of the matrix named exp? $\endgroup$ – Kohl Kinning Oct 29 '18 at 21:44
  • 1
    $\begingroup$ The exp and the mat matrices do not match. Please show a reproducible, minimal example. Also by converting the exp to a data.frame it might convert your numeric values to factors, which is why you see them with each factor into a different color. $\endgroup$ – llrs Oct 30 '18 at 10:36
  • 1
    $\begingroup$ @Llopis, the data tables ought not match. exp is a binarized version of mat as Feresh Teh states above. Feresh Teh: I would think you would want rowSums(), that way all of the columns for a given row will be reduced to the sum and you'll have a value (expression?) for each cell. When you add metadata to a Seurat object, it will be in this format--a value for each cell. $\endgroup$ – Kohl Kinning Oct 30 '18 at 14:22
3
$\begingroup$

If you would like to color discrete intervals on a gradient as opposed to having a continuous gradient (like your second plot), use this approach.

It is similar to the approach in the answer I posted with the continuous scale, but we simply break up the continuous scale in to intervals and color them by these intervals.

#generate values for testing purposes, one value for each cell, add to object
value <- sample(seq(from=8, to=46, by=1), size = length(rownames(exp@meta.data)), replace=TRUE)
exp <- AddMetaData(object=exp, metadata=value, col.name="value")

#encode the continuous values as factors, determined by the interval they fall in to
value_breaks <- cut(exp@meta.data$value, breaks = c(8,18,28,38,46), include.lowest=TRUE, right=FALSE)

#name the breaks by cell so they can be added by AddMetaData(), add them
names(value_breaks) <- rownames(exp@meta.data)
exp <- AddMetaData(object=exp, metadata=value_breaks, col.name="value_breaks")

exp <-  SetIdent(exp, ident.use=exp@meta.data$value_breaks)

#create a gradient between specified intervals
#length.out is the number of different colors, number of factor levels
cols <- scales::seq_gradient_pal(low="blue", high="yellow", space="Lab")(seq(from=0, to=1,length.out=4))

TSNEPlot(exp, colors.use=cols)

intervals colored

The label in the legend are classic mathematical notation for intervals. You can add your own labels if you wish in the cut() function.

$\endgroup$
  • $\begingroup$ Thanks a lot, you solved my long lasting question, I produced this final picture image.ibb.co/dsZzsf/Rplot22.png here I am plotting 300 cell types markers hopefully to gene 2 clusters yellow and another one blue, but as you are seeing the threshold I have selected for converting mat (expression values) of these genes to a binary matrix has not been good because my tsne does not look brilliant. If you where me, what threshold you would selected to converting your expression values to binary? I have used the average of expression of these genes in clusters as threshold. for example 2 $\endgroup$ – Angel Nov 1 '18 at 17:46
  • 1
    $\begingroup$ That deserves a post of its own! That exact question was asked on here a few weeks back, but I think it was removed. There were at least four different "answers." It is a difficult questions with all answers (that I've seen) with their own limitations. $\endgroup$ – Kohl Kinning Nov 4 '18 at 16:05
  • $\begingroup$ Thank you, in CellRouter R package, developer uses a centring on data before plotting but I am not able to understand what he is doing. But his centring is very good github.com/edroaldo/cellrouter/blob/master/CellRouter_Class.R in plotDRExpression function $\endgroup$ – Angel Nov 4 '18 at 16:35
3
$\begingroup$

TSNEPlot()

TSNEPlot() will always treat your variables as discrete. My approach is to manually generate a gradient with unique colors for each factor level and pass it to the cols.use argument in TSNEPlot().

#generate values for testing purposes, one value for each cell
value <- sample(seq(from=8, to=48, by=1), size = length(rownames(unfiltered_cca@meta.data)), replace=TRUE)
names(value) <- rownames(exp@meta.data)
exp <- AddMetaData(object=exp, metadata=value, col.name="value")
exp <-  SetIdent(exp, ident.use=exp@meta.data$value)
#create a gradient between specified colors, multiply by sequence to get appropriate length
#length.out is the number of different colors, number of factor levels
cols <-  scales::seq_gradient_pal(low="beige", high="red", space="Lab")(seq(from=0, to=1,length.out=48))


TSNEPlot(exp, colors.use=cols)

enter image description here

I would present this with no legend:

TSNEPlot(exp, colors.use=cols, no.legend = TRUE)

enter image description here

FeaturePlot()

You can also simply use FeaturePlot() instead of TSNEPlot() to visualize the gradient. Using the same data as above:

FeaturePlot(object = exp, features.plot = "value", reduction.use = "tsne", no.legend = FALSE, cols.use = c("beige", "red"))

enter image description here

You ask for a continuous scale, but this is not what is shown in your second plot. You have a different color for each discrete range of values. I assume here that you want a continuous gradient. To achieve something like the second plot you provided requires a different approach.

$\endgroup$
  • $\begingroup$ Sorry, your solution returned colourless featureplot $\endgroup$ – Angel Nov 1 '18 at 9:37
  • $\begingroup$ TSNEPlot() returns a colourless plot. I have used cc@meta.data$value as my value and in col I used cols <- scales::seq_gradient_pal(low="beige", high="red", space="Lab")(seq(from=10, to=48, length.out=34)), however returned a colourless lot $\endgroup$ – Angel Nov 1 '18 at 14:13
  • 1
    $\begingroup$ seq(from=0, to=1, length.out=34) is essential. You need a sequence of decimals for the color gradient. The length.out is correct. $\endgroup$ – Kohl Kinning Nov 1 '18 at 14:14
  • $\begingroup$ Actually like my second plot I would like to have a colour range for example 8-18, 18-28, 28-38, 38-48 in gradient of colours but I don't know how to do that $\endgroup$ – Angel Nov 1 '18 at 14:15
  • $\begingroup$ Thanks a lot, now I have a colourful tsne plot, sorry do you know how to convert this wide range of colours to 4 ranges 8-18, 18-28, 28-38, 38-48 as a yellow to blue gradient like my second plot? $\endgroup$ – Angel Nov 1 '18 at 14:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.