knitr::opts_chunk$set(echo = TRUE)
One of the most challenging issues when converting biological pathways to graphs is also a major problem for
graphs in general: what are the best ways to display the result? In the previous vignettes (Getting Started,
and Metrics) we included default plots of the graphs, which leave something to be desired. Our initial
constructions and plots use rectangles or circles (the only shapes available by default in igraph
) of a
fixed size, so the labels (such as gene names) often extend outside their containers. In addition, many edges
cross and many nodes overlap, making it harder to visually identify the underlying topological structure of
the graph that we learned from the pathway.
In this vignette, we want to look at methods that may help achieve better plots.
First, we load the required libraries:
library(WayFindR) suppressMessages( library(igraph) )
Now we are ready to load our GPML file and convert it into an igraph object:
xmlfile <- system.file("pathways/WP3850.gpml", package = "WayFindR") G <- GPMLtoIgraph(xmlfile) class(G)
First, we repeat the original plot of this pathway that we included in the "Getting Started" vignette (Figure 1).
set.seed(13579) L <- igraph::layout_with_graphopt plot(G, layout=L) title("WP3850") nodeLegend("topleft", G) edgeLegend("bottomright", G)
Next, we change circles to ellipses, and try to find sensible sizes so the nodes can contain the labels (Figure 2). This helps with (mostly) fitting nodes to labels, but it doesn't do anything about layout issues.
wc <- which(V(G)$shape == "circle") G <- set_vertex_attr(G, "shape", index = wc, value = "ellipse") plot(0,0, type = "n") opar <- par(mai = c(0.05, 0.05, 1, 0.05)) sz <- (strwidth(V(G)$label) + strwidth("oo")) * 92 G <- set_vertex_attr(G, "size", value = sz) G <- set_vertex_attr(G, "size2", value = strheight("I") * 2 * 92) set.seed(13579) L <- layout_with_graphopt(G) plot(G, layout = L) title("WP3850") edgeLegend("bottomleft", G) nodeLegend("bottomright", G) par(opar)
Now we use a different method to try to improve the layout (Figure 3). First, we compute coordinates
from the layout_nicely
algorithm, which mostly avoids edge-crossings when possibly, but doesn't alleviate
the issue with overlapping nodes. Then we pass those coordinates to one of the "force-directed" layout
algorithms (in this case, Kamada-Kawai, but several other options are available). The second step
reduces the overlap problem.
set.seed(12345) L <- layout_nicely(G) L2 <- layout_with_kk(G, coords=L) plot(G, layout = L2) title("WP3850") edgeLegend("bottomleft", G) nodeLegend("bottomright", G) par(opar)
Finally, we switch R packages entirely. While igraph
clearly has a much wider and more useful implementation
of graph-theoretic algorithms from the computer science world, Rgraphviz
has a better selection of node shapes
and a different (though smaller) set of layout algorithms. We have wrapped umuch of the conversion from one
format to another in our own "as.graphNEL
" function. The result (Figure 4) clearly does a better job of
fitting nodes to labels and avoiding overlaps, but it does have several edge crossings that are undesirable
GN <- as.graphNEL(G) suppressMessages( library(Rgraphviz) ) plot(GN)
We can also explore different layout algorithms in Rgraphviz.
plot(GN, "twopi")
Any scripts or data that you put into this service are public.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.