geom_tree2 won't work with aes(color=CategoricalVAR) about ggtree

[hadley]

Are you sure you should be using compute_group() and not compute_panel()?

[GuangchuangYu]

@hadley it works by using compute_panel(). Thank you!

[GuangchuangYu]

@hadley Sorry to bother you. I have another issue.

require(ape)
require(ggtree)
x=rmtree(3, 10)
df=fortify(x)

> head(df)
          node parent branch.length         x y label isTip    branch angle
Tree #1.1    1     13    0.03709821 0.6850552 2    t8  TRUE 0.6665061    18
Tree #1.2    2     13    0.65754318 1.3055002 3    t1  TRUE 0.9767286   -18
Tree #1.3    3     12    0.88514687 1.1904847 1   t10  TRUE 0.7479112    54
Tree #1.4    4     16    0.27829065 1.3919800 5    t3  TRUE 1.2528347   -90
Tree #1.5    5     16    0.37527841 1.4889678 6    t5  TRUE 1.3013286  -126
Tree #1.6    6     15    0.01499930 0.9624178 4    t6  TRUE 0.9549182   -36
              .id
Tree #1.1 Tree #1
Tree #1.2 Tree #1
Tree #1.3 Tree #1
Tree #1.4 Tree #1
Tree #1.5 Tree #1
Tree #1.6 Tree #1

geom_tree2() now works fine with phylo object, but not multiPhylo (a list of phylo). If the object is multiPhylo, after fortify() the df will have an additional column, .id.

What I expected is:

ggplot(df) + geom_tree() + facet_wrap(~.id)

But geom_tree2 only output the first panel.

ggplot(df) + geom_tree2() + facet_wrap(~.id)

The data has no problem, as I can scale the color based on .id.

ggplot(df) + geom_tree2() + aes(color=.id)

It seems that I should write a compute_layer() function. I read some source code of ggplot2, but it is too complicated for me. Can you give me some hints?

[hadley]

compute_panel should take care of this for you - have you checked to verify that it's called three times?

[GuangchuangYu]

geom_tree <- function(layout="rectangular", ...) {
    x <- y <- parent <- NULL
    lineend  = "round"
    if (layout == "rectangular" || layout == "fan" || layout == "circular") {
        list(
            geom_segment(aes(x    = x[parent],
                             xend = x,
                             y    = y,
                             yend = y),
                         lineend  = lineend, ...),

            geom_segment(aes(x    = x[parent],
                             xend = x[parent],
                             y    = y[parent],
                             yend = y),
                         lineend  = lineend, ...)
            )
    } else if (layout == "slanted" || layout == "radial" || layout == "unrooted") {
        geom_segment(aes(x    = x[parent],
                         xend = x,
                         y    = y[parent],
                         yend = y),
                     lineend  = lineend, ...)
    }
}

To allow subsetting (i.e. x[parent]) properly, I change the parent node number in fortify.multiPhylo().

fortify.multiPhylo <-  function(model, data, layout="rectangular", 
                                ladderize=TRUE, right=FALSE, mrsd=NULL, ...) {

    df.list <- lapply(model, function(x) fortify(x, layout=layout, ladderize=ladderize, right=right, mrsd=mrsd, ...))
    if (is.null(names(model))) {
        names(df.list) <- paste0("Tree ", "#", seq_along(model))
    } else {
        names(df.list) <- names(model)
    }
    df <- do.call("rbind", df.list)
    df$.id <- rep(names(df.list), times=sapply(df.list, nrow))
    df$.id <- factor(df$.id, levels=names(df.list))

    nNode <- sapply(df.list, nrow)
    nNode2 <- cumsum(c(0, nNode[-length(nNode)])) 
    df$parent <- df$parent + rep(nNode2, times=nNode)
    return(df)
}

It works fine. So ggplot use the whole data frame in each panel when using facet_wrap(), otherwise subsetting will be failed.

But for geom_tree2, https://github.com/GuangchuangYu/ggtree/blob/master/R/geom_tree.R#L1-L144, only the first panel was plotted correctly and I am sure the compute_panel was called three times.

I test it without modify the parent node number, and it works with facet_wrap()

fortify.multiPhylo2 <-  function(model, data, layout="rectangular", 
                                ladderize=TRUE, right=FALSE, mrsd=NULL, ...) {

    df.list <- lapply(model, function(x) fortify(x, layout=layout, ladderize=ladderize, right=right, mrsd=mrsd, ...))
    if (is.null(names(model))) {
        names(df.list) <- paste0("Tree ", "#", seq_along(model))
    } else {
        names(df.list) <- names(model)
    }
    df <- do.call("rbind", df.list)
    df$.id <- rep(names(df.list), times=sapply(df.list, nrow))
    df$.id <- factor(df$.id, levels=names(df.list))

    ## nNode <- sapply(df.list, nrow)
    ## nNode2 <- cumsum(c(0, nNode[-length(nNode)])) 
    ## df$parent <- df$parent + rep(nNode2, times=nNode)
    return(df)
}

With geom_tree2(), facet_wrap() use the subset of data in each panel. This is the reason why it fail with fortify.multiPhylo but not fortify.multiPhylo2.

So fortify.multiPhylo + geom_tree + facet_wrap works,
and fortify.multiPhylo2 + geom_tree2 + facet_wrap works too.

Q1

Why facet_wrap() behave different with geom_tree (use whole data in each panel) and geom_tree2 (use data subset in each panel)?

Q2

Is there any way to group the data, so that ggplot()+geom_tree2() can plot the trees correctly without changing the parent node number?

[GuangchuangYu]

Q2 was solved by keeping the original parent node number un-changed with updated Stat.

Q1 is still an open question. Why the calculation procedure of facet_wrap() is different in geom_tree and geom_tree2?