Multipanel ggplot with subfigure labels (and drawing igraph structures with ggplot2)

I want to make some multipanelled plots in R where, for a given figure:

- each of the subfigures is a ggplot (or other grid::grob object)

- each of the subfigures is given a label in the top-left hand corner

I've been using ggplot to draw graphs (the graph theory kind). It isn't quite set up to do this and indeed, igraph has perfectly good graph-drawing functions. They just don't return grobs (neither does sna::gplot), so they are a bit limited to my mind. Hence, I've set up a couple of simple functions to make ggplot figures out of igraph structures using the igraph::layout functions

Here's an example figure from my current dissertation. We first set up a theme to remove the weird default background for ggplot figures, import some libraries and define a function that returns a label in a rectangle.

library(igraph)
library(grid)
library(gridExtra)
library(ggplot2)

theme_nothing <- function(
  base_size = 12, base_family = "Helvetica"
  ){
  # blank plotting background for ggplot2
  # - for removing all axes etc when plotting igraphs in ggplot2
  # - copied from ggplot2 docs
  theme_bw(base_size = base_size, base_family = base_family
    ) %+replace% theme(
    rect = element_blank(),
    line = element_blank(),
    text = element_blank(),
    legend.position = 'none'
    )
  }

labelGrob <- function(label){

  # function to generate subfigure labels
  require(grid)
  require(gridExtra)
  textGrob(label, y = unit(0.9, 'npc'))
  }

Then we define a function to convert an igraph object into a dataframe of x/y coords so that 
it can be plotted in ggplot.




igraph_to_ggplot_df <- function(
  X,
  coord.func = layout.fruchterman.reingold
  ){
  # Makes a dataframe of vertex / edge positions from an igraph
  #   object
  stopifnot(is(X, "igraph"))

  # Give the vertices some names, if they don't already have them
  # then use the coord.func to define X/Y positions for each vertex
  v.names  <- names(V(X))
  if(is.null(v.names)){v.names <- paste0('v', 1:length(V(X)))}
  v.coords <- scale(coord.func(X))
  dimnames(v.coords) <- list(v.names, c('x', 'y'))
  # Obtain the edges for the graph, give them names (if they don't
  #   already have them) and then associate the start and end point
  #   of each edge with the X/Y coords of the relevant vertex
  e.list   <- get.edgelist(X)
  stopifnot(nrow(e.list) > 0)
  e.names <- names(E(X))
  if(is.null(e.names)){
    # assume the graph isn't null
    e.names <- paste0('e', 1:nrow(e.list))
    }
  dimnames(e.list) <- list(e.names, c('head', 'tail'))
  stopifnot(length(intersect(v.names, e.names)) == 0)
  # convert the edge / vertex coords into a single dataframe for use in ggplot
  edge_df <- data.frame(
    element.name = rep(e.names, 2),
    element.label = '',
    element.type = 'edge',
    x      = v.coords[as.vector(e.list[, c('head', 'tail')]), 'x'],
    y      = v.coords[as.vector(e.list[, c('head', 'tail')]), 'y']
    )
  vertex_df <- data.frame(
    element.name = rownames(v.coords),
    element.label = rownames(v.coords),
    element.type = 'vertex',
    x = v.coords[, 'x'],
    y = v.coords[, 'y']
    )
  graph_df <- rbind(edge_df, vertex_df)
  graph_df
  }

Then we define a function to turn igraph objects into ggplot graphical objects, so that 
these can be arranged and plotted out using grid.



igraph_ggplot <- function(
  X,
  coord.func = layout.fruchterman.reingold,
  offset = 0.5
  ){
  # Converts a graph into x/y coords, 
  #   then plots the graph as straight-line joined coords
  d.frame <- igraph_to_ggplot_df(X, coord.func)
  # Note that if I don't change the offset myself,
  #   arrangeGrob cuts off distal bits of my graph
  x.range <- range(d.frame$x)
  y.range <- range(d.frame$y)
  g <- ggplot(data = d.frame,
    aes(x = x, y = y, group = element.name, label = element.label)
    ) + geom_line(
    ) + geom_point(colour = 'darkgoldenrod1', size = 10
    ) + geom_point(colour = 'white',          size = 7
    ) + geom_text(hjust = 0.5,
                  vjust = 0.5,
                  aes(colour = element.type)
    ) + theme_nothing(
    ) + xlim(x.range[1] - offset, x.range[2] + offset
    ) + ylim(y.range[1] - offset, y.range[2] + offset)
  g
  }




This is just an example of a figure generated with the above code. We define the incidence
matrix of a graph, and plot out the incidence, adjacency and Laplacian matrices of that graph
(as tables) and a picture of the graph itself. The labels are added using arrangeGrob and
 labelGrob: I define an arrangement of 8 panels (2rows x 4 columns). The labels are 
placed into the first and third columns (which are 1/4 the width of the second and fourth 
columns.)
fig1_2_1 <- function(){
  # Define a graph using it's incidence matrix, then plot
  # Signed incidence matrix
  D <- matrix(c(1, 1, 1, 0,
                -1,0, 0, 1,
                0,-1, 0,-1,
                0, 0,-1, 0,
                0, 0, 0, 0
                ),
              byrow = TRUE, nrow = 5,
              dimnames = list(paste0('v', 1:5),
                              paste0('e', 1:4)))
  B <- abs(D)        # Unsigned incidence matrix
  Q <- D %*% t(D)    # Laplacian
  deg <- diag(Q)     # Degree vector
  A <- diag(deg) - Q # Adjacency matrix
  
  X <- graph_from_adjacency_matrix(
    A, mode = 'undirected'
    )
  g1 <- tableGrob(A)
  g2 <- tableGrob(D)
  g3 <- igraph_ggplot(X, offset = 0.25)
  g4 <- tableGrob(Q)
  ag <- arrangeGrob(labelGrob('a'), g1,
                    labelGrob('b'), g2,
                    labelGrob('c'), g3,
                    labelGrob('d'), g4,
                    nrow = 2,
                    widths = c(1,4,1,4))

  grid.draw(ag)
  NULL
  }


------
Voila!





---------
# TODO - show how to do the same thing with grid.draw(a); grid.draw(b)
 where 'a' contains figures and 'b' contains panel labels

# TODO - eugh! work out why blogger is making my code/paragraphs so ugly?