R network plot with node types fixed in columns

Question

I have two data frames that look like this:

y1 <- c(1, 0, 0) 
y2 <- c(0, 1, 0) 
y3 <- c(0, 0, 1) 
df1 <- data.frame(y1, y2, y3, row.names = c("x1", "x2", "x3"))

y1 <- c(1, 0, 0) 
y2 <- c(1, 0, 0) 
y3 <- c(1, 0, 0) 
df2 <- data.frame(y1, y2, y3, row.names = c("z1", "z2", "z3"))

I want to graph the relationships in these dataframes so that the x, y, and z values appear in columns with lines connecting them. Here's a rough example of what I'm looking to achieve:

I considered using a ggplot2 scatterplot with a categorical variable on the X axis to generate the columns, but I couldn't figure out how to generate the connecting lines between points from that. I also looked into ggnet2 for a network plot, but couldn't find any examples where nodes were fixed in columns.

EDIT:

My real use case has approximately 20 x points, 120 y points and 200 z points, so ideally the solution would scale fairly easily.

I attempted the following solution using a sankeynetwork plot from the networkD3 package

library(networkD3)
Nodes <- data.frame(name = c("x1", "x2", "x3", "y1", "y2", "y3", "z1", 
         "z2", "z3"), group = c("1", "1", "1", "2", "2", "2", "3", "3", 
         "3"))
Links <- data.frame(source = c(0, 1, 2, 3, 4, 5), target = c(3, 4, 5, 6, 
         6, 6), value = 1, 1, 1, 1, 1, 1)

sankeyNetwork(Links = Links, Nodes = Nodes, Source = "source",
          Target = "target", Value = "value", NodeGroup = "group", NodeID 
          = "name", sinksRight = FALSE)   

The result is sort of correct... however it is probably not ideal. Additionally there doesn't seem to be a clear way of forcing z2 and z3 to appear with z1 on the right side of the page without going into the underlying javascript, which I have no idea how to do (see d3 sankey charts - manually position node along x axis)

Is there a better solution, or a way to improve this one?

Thanks!

Answer 1

Here is one possible solution using geom_segment to draw the connecting edges. I don't how adaptable it will be to datasets larger or more complex than your example. I suspect there is a more graceful and scalable way to handle this with igraph or ggraph.

# Start with two data.frames: one for node positions,
# and one for edges you want to draw between nodes.

pos_dat = data.frame(node_id=paste(rep(c("x", "y", "z"), each=3), 
                                   rep(c("1", "2", "3"), times=3),
                                   sep=""),
                     type=rep(c("x", "y", "z"), each=3),
                     xpos=rep(c(1, 2, 3), each=3),
                     ypos=rep(c(1, 2, 3), times=3))

#   node_id type xpos ypos
# 1      x1    x    1    1
# 2      x2    x    1    2
# 3      x3    x    1    3
# 4      y1    y    2    1
# 5      y2    y    2    2
# 6      y3    y    2    3
# 7      z1    z    3    1
# 8      z2    z    3    2
# 9      z3    z    3    3


edge_dat = data.frame(start=c("x1", "x2", "x3", "y1", "y2", "y3"),
                        end=c("y1", "y2", "y3", "z1", "z1", "z1"))

#   start end
# 1    x1  y1
# 2    x2  y2
# 3    x3  y3
# 4    y1  z1
# 5    y2  z1
# 6    y3  z1

---

# Use two successive merges to join node x,y positions
# for each edge you want to draw.

tmp_dat = merge(edge_dat, pos_dat, by.x="start", by.y="node_id")
seg_dat = merge(tmp_dat, pos_dat, by.x="end", by.y="node_id")

# Remove unneeded columns and change column names for convenience.
seg_dat$type.x = NULL
seg_dat$type.y = NULL
names(seg_dat) = c("end", "start", "x", "y", "xend", "yend")

seg_dat
#   end start x y xend yend
# 1  y1    x1 1 1    2    1
# 2  y2    x2 1 2    2    2
# 3  y3    x3 1 3    2    3
# 4  z1    y1 2 1    3    1
# 5  z1    y2 2 2    3    1
# 6  z1    y3 2 3    3    1

---

# Finally, draw the plot.

library(ggplot2)

p = ggplot() +
    geom_segment(data=seg_dat, aes(x=x, y=y, xend=xend, yend=yend),
                 colour="grey50") +
    geom_point(data=pos_dat, aes(x=xpos, y=ypos, colour=type), size=10) +
    geom_text(data=pos_dat, aes(x=xpos, y=ypos, label=node_id)) +
    scale_colour_manual(values=c(x="steelblue", y="darkorange", z="olivedrab3"))

ggsave("plot.png", plot=p, height=4, width=6, dpi=150)