cell sizes in R plots

Question

[Assa Yeroslaviz]

Hi,

this is a follow-up from this Question.

I have created a plot using the ggplot2 package. But as the matrix is very large (almost 146000 rows), the single cells of the image are quite small.

I would like to know how to make the single cell sizes bigger, so I will get a better overview inside the image about the differences between the different cells.

I also would like to know if it is possible to make a bigger (longer) lengend of more than just five elements? I would like to create at least 20 different coloured groups of distinguish different colors

( a logical question - why does it plot it in a triangle?)

This is how I create the plot:

require(ggplot2)
pl1 <- ggplot(data, aes(y = partner1, x = partner2)) + geom_tile(aes(fill = Substract)) + scale_fill_continuous(low = "blue", high = "green") + scale_size(range = c(1, 20000))

Thanks
Assa

[Devon Ryan]

By "legend", do you instead mean tick labels? The actual legend is the color bar on the right.

BTW, the answer to your logic question is that your data is in a triangular shape (well, partner1~partner2 is). This is likely due to how you generated those values.

Edit: You can test what I mentioned above regarding shape with with(data, table(parner1>partner2)).

[Assa Yeroslaviz]

Actually I do mean the bar at the side of the image :-)

[Devon Ryan]

If you just want more colors then try scale_fill_gradientn().

[Irsan]

I wonder whether making such a heatmap of your data is the best way to vizualize the results in the first place. In the heatmap, genomic positions are converted to non-numeric values which makes it hard to see the relative distances along the genome. Furthermore, I think there is too much data per pixel and the distribution of "substraction-values" is very scewed towards the lower numbers which makes it hard to see the different colors (maybe you should log-transform the substraction-values).

But what do you really want to show to the viewer? The things I can think of are:

1. How much interactions are observed at what genomic position
2. Are there genomic hotspots where there are lots of interactions
3. What genomic locations co-relate (read: if genomic position A is interacting, often genomic position B is also interacting)

In any case, it is very hard to answer with the tile-plot you are trying to make.

In case of 1 and 2, make a karygram overview of the distribution (histogram) of interactions along the chromosomes. Then use geom_histogram() + facet_wrap(~chr,...)

In case of 3, compure the pairwise correlation coefficients between all positions and make a How Do I Draw A Heatmap In R With Both A Color Key And Multiple Color Side Bars?

[Assa Yeroslaviz]

It is not really continuous genomic positions, but bins of 1000 (or 5000) positions summarized into one value. And I do have only one chromosome. I thought a heatmap will be better, because of the better (coloured) overview. (Is there a way to do a histograms with different colours for specific values ranges?). I will try the histogram - BTW did you mean karyogram (Is it a GRanges Object?).

Answer 1

[Alex Reynolds]

When making high-density scatterplots, consider looking into hexbin() or sunflowerplot() as alternatives. Alternatively, you can do levelplot() with the lattice library.

[Assa Yeroslaviz]

thanks for the tips, but I don't think the sunflowerplot() and the hexbin() are the right plot option for me. Though I find the sunflowerplot a very nice thing - if it is possible to alter the plot to my needs it will be great.

I have three different columns I need to plot. one is for the x-axis, the second for the y-axis and the third one is the informations which come in the plot itself. The sunflowerplot compare two columns with each other.