plotting interactions in R with two data sets

Question

[frymor]

Hi all,

I have a data set of two postions on the genome with a third value for number of interactions. I would like to plot this data set so I can see how many interactions are on each position.

the data set looks like that (this is only a subset of the complete, very long list):

partner1    partner2    Interactions
1    10001    11
1    15001    1
1    20001    1
1    25001    4
1    30001    8
5001    20001    1
5001    40001    3
5001    45001    15
5001    50001    1
10001    15001    3
10001    20001    3
10001    25001    6
10001    30001    12
15001    70001    2
15001    90001    6
15001    95001    5
15001    100001    1
20001    4195001    30
20001    4200001    62
20001    4205001    81
20001    4210001    3
25001    30001    5
25001    40001    22
25001    45001    13
4200001    4210001    318
4200001    4215001    2
4205001    4210001    308
4205001    4215001    2
4210001    4215001    1

i would like to have the column 'partner1' on the x-axis, the column 'partner2' on the y-axis and the number of interactions (3rd column) in the plot with the option to have there either a point, the number itself of a colored gradient like in the heatmaps.

Does anyone know of an R package for creating such plots, or for that matter, any other way of doing it?

thanks

Assa

[t.candelli]

I think the best way to represent this sort of data would be with a heatmap. is there a directionality between partner one and partner 2? e.g. the values 1 5000 8 are different from 5000 1 8 in your table

[frymor]

yes there is a difference. The information on the two partner columns are genomic positions. So it make a difference whether the first or the second partner is on a specific position. Doesn't it?

How would you put the data into a heatmap?

Answer 1

[Irsan]

There are many possibilities, one of them is using ggplot2 (R-library)

library(ggplot2)
ggplot(data) + geom_tile(aes(x=factor(partner1),y=factor(partner2),fill=Interactions))

[frymor]

I have tried with ggplot.

require(ggplot2)
pl1 <- ggplot(subset, aes(y = factor(partner1), x = factor(partner2))) + geom_tile(aes(fill = Interactions)) + scale_fill_continuous(low = "blue", high = "green") + scale_size(range = c(1, 200))

With the small subset I get a similar plot to the one you posted. But with the complete data set I get a different picture:

Is there a simple explanation for that? Does the order of the columns of the two partner columns make a difference?

[Irsan]

first prepare your data frame

data$partner1 <- factor(data$partner1, levels=sort(unique(data$partner1)))

(and also for partner2) then plot without the factor() part

[frymor]

That still didn't change anything. I still get the plot on only half of the window. I can't figure why, as I have for both columns the same amount of factors (842 vs. 843).

[Assa Yeroslaviz]

is it possible to make the legend a bit more comprehensive? I won't to have more than just 5 different categories. I need a much bigger separation - something like 20 or 25 different color points.

Answer 2

[dariober]

What about this...

## Dummy data

dat<- data.frame(partner1= 1:100, partner1= 1:100, Interactions= 1:100)

ncols<- length(unique(dat$Interactions))<br />
cols<- data.frame(<br />
    colour= colorRampPalette(c("blue", "red"))(ncols),
    Interactions= sort(unique(dat$Interactions)), stringsAsFactors= FALSE)

dat<- merge(dat, cols)

## Unocmment to Make colour transparent, it might look better
#trasp<- '80'
#dat$colour<- paste(dat$colour, trasp, sep= '')

## Plot symbol
plot(x= dat$partner1, y= dat$partner2, pch= 19, col= dat$colour, cex= 2)

## As text
plot(x= dat$partner1, y= dat$partner2, type= 'n')
text(x= dat$partner1, y= dat$partner2, labels= dat$Interactions, col= dat$colour, cex= 0.5)

[frymor]

Thanks I will give it a try...

Answer 3

[t.candelli]

I'm going to use the "pheatmap" package to draw a heatmap of your data. with the code below I generate a matrix from your dataframe so that it can be used as an argument for pheatmap.

library(pheatmap)

names<-unique(c(data[,1], data[,2]))
mat<-matrix(data=0, nrow=length(names), ncol=length(names))
rownames(mat)<-sort(names)
colnames(mat)<-sort(names)

for (i in 1:nrow(data))
{
  partner1 <- as.character(data[i,1])
  partner2 <- as.character(data[i,2])
  interactions <- data[i,3]

  mat[partner1, partner2] <- interactions
}

pheatmap(mat, cluster_cols=F,  cluster_rows=F)

Answer 4

[Manu Prestat]

A good solution to such a problem is to draw a network representation where:

- partners are nodes

- column 3 is the thickness of the link

THE SOFT for that is Cytoscape

Answer 5

[theobroma22]

I would use a circle plot and have the ribbon thickness represent the strength of the interaction.