1 ## dist.topo.R (2013-02-09)
3 ## Topological Distances, Tree Bipartitions,
4 ## Consensus Trees, and Bootstrapping Phylogenies
6 ## Copyright 2005-2013 Emmanuel Paradis
8 ## This file is part of the R-package `ape'.
9 ## See the file ../COPYING for licensing issues.
11 dist.topo <- function(x, y, method = "PH85")
13 if (method == "score" && (is.null(x$edge.length) || is.null(y$edge.length)))
14 stop("trees must have branch lengths for branch score distance.")
15 nx <- length(x$tip.label)
18 bp1 <- .Call("bipartition", x$edge, nx, x$Nnode, PACKAGE = "ape")
19 bp1 <- lapply(bp1, function(xx) sort(x$tip.label[xx]))
20 ny <- length(y$tip.label) # fix by Otto Cordero
21 ## fix by Tim Wallstrom:
22 bp2.tmp <- .Call("bipartition", y$edge, ny, y$Nnode, PACKAGE = "ape")
23 bp2 <- lapply(bp2.tmp, function(xx) sort(y$tip.label[xx]))
24 bp2.comp <- lapply(bp2.tmp, function(xx) setdiff(1:ny, xx))
25 bp2.comp <- lapply(bp2.comp, function(xx) sort(y$tip.label[xx]))
29 if (method == "PH85") {
33 if (identical(bp1[[i]], bp2[[j]]) | identical(bp1[[i]], bp2.comp[[j]])) {
39 dT <- q1 + q2 - 2 * p # same than:
40 ##dT <- if (q1 == q2) 2*(q1 - p) else 2*(min(q1, q2) - p) + abs(q1 - q2)
42 if (method == "score") {
49 if (identical(bp1[[i]], bp2[[j]]) | identical(bp1[[i]], bp2.comp[[j]])) {
50 dT <- dT + (x$edge.length[which(x$edge[, 2] == nx + i)] -
51 y$edge.length[which(y$edge[, 2] == ny + j)])^2
52 found1 <- found2[j] <- TRUE
56 if (found1) found1 <- FALSE
57 else dT <- dT + (x$edge.length[which(x$edge[, 2] == nx + i)])^2
60 dT <- dT + sum((y$edge.length[y$edge[, 2] %in% (ny + which(!found2))])^2)
66 .compressTipLabel <- function(x)
68 ## 'x' is a list of objects of class "phylo" possibly with no class
69 if (!is.null(attr(x, "TipLabel"))) return(x)
70 ref <- x[[1]]$tip.label
72 if (length(unique(ref)) != n)
73 stop("some tip labels are duplicated in tree no. 1")
75 ## serious improvement by Joseph W. Brown!
76 relabel <- function (y) {
78 if (!identical(label, ref)) {
79 if (length(label) != length(ref))
80 stop(paste("tree ", y, "has a different number of tips"))
81 ilab <- match(label, ref)
83 stop(paste("tree ", y, "has different tip labels"))
84 ie <- match(1:n, y$edge[, 2])
90 x <- unclass(x) # another killer improvement by Tucson's hackathon (1/2/2013)
91 x <- lapply(x, relabel)
92 attr(x, "TipLabel") <- ref
93 class(x) <- "multiPhylo"
97 prop.part <- function(..., check.labels = TRUE)
100 if (length(obj) == 1 && class(obj[[1]]) != "phylo")
103 ## class(obj) <- NULL # needed? apparently not, see below (2010-11-18)
106 if (ntree == 1) check.labels <- FALSE
107 if (check.labels) obj <- .compressTipLabel(obj) # fix by Klaus Schliep (2011-02-21)
108 for (i in 1:ntree) storage.mode(obj[[i]]$Nnode) <- "integer"
110 ## The 1st must have tip labels
111 ## Maybe simply pass the number of tips to the C code??
112 obj <- .uncompressTipLabel(obj) # fix a bug (2010-11-18)
114 clades <- .Call("prop_part", obj, ntree, TRUE, PACKAGE = "ape")
115 attr(clades, "number") <- attr(clades, "number")[1:length(clades)]
116 attr(clades, "labels") <- obj[[1]]$tip.label
117 class(clades) <- "prop.part"
121 print.prop.part <- function(x, ...)
123 if (is.null(attr(x, "labels"))) {
124 for (i in 1:length(x)) {
125 cat("==>", attr(x, "number")[i], "time(s):")
126 print(x[[i]], quote = FALSE)
129 for (i in 1:length(attr(x, "labels")))
130 cat(i, ": ", attr(x, "labels")[i], "\n", sep = "")
132 for (i in 1:length(x)) {
133 cat("==>", attr(x, "number")[i], "time(s):")
134 print(x[[i]], quote = FALSE)
139 summary.prop.part <- function(object, ...) attr(object, "number")
141 plot.prop.part <- function(x, barcol = "blue", leftmar = 4, ...)
143 if (is.null(attr(x, "labels")))
144 stop("cannot plot this partition object; see ?prop.part for details.")
146 n <- length(attr(x, "labels"))
147 layout(matrix(1:2, 2, 1), heights = c(1, 3))
148 par(mar = c(0.1, leftmar, 0.1, 0.1))
149 plot(1:L, attr(x, "number"), type = "h", col = barcol, xlim = c(1, L),
150 xlab = "", ylab = "Frequency", xaxt = "n", bty = "n")
151 plot(0, type = "n", xlim = c(1, L), ylim = c(1, n),
152 xlab = "", ylab = "", xaxt = "n", yaxt = "n")
153 for (i in 1:L) points(rep(i, length(x[[i]])), x[[i]], ...)
154 mtext(attr(x, "labels"), side = 2, at = 1:n, las = 1)
157 prop.clades <- function(phy, ..., part = NULL, rooted = FALSE)
161 ## Are we going to keep the '...' way of passing trees?
163 if (length(obj) == 1 && class(obj[[1]]) != "phylo")
164 obj <- unlist(obj, recursive = FALSE)
166 part <- prop.part(obj, check.labels = TRUE)
169 ## until ape 3.0-7 it was assumed implicitly that the labels in phy
170 ## are in the same order than in 'part' (bug report by Rupert Collins)
171 if (!identical(phy$tip.label, attr(part, "labels"))) {
172 i <- match(phy$tip.label, attr(part, "labels"))
173 j <- match(seq_len(Ntip(phy)), phy$edge[, 2])
175 phy$tip.label <- attr(part, "labels")
179 bp <- postprocess.prop.part(bp)
180 part <- postprocess.prop.part(part) # fix by Klaus Schliep
181 ## actually the above line in not needed if called from boot.phylo()
184 n <- numeric(phy$Nnode)
185 for (i in seq_along(bp)) {
186 for (j in seq_along(part)) {
187 ## we rely on the fact the values returned by prop.part are
188 ## sorted and without attributes, so identical can be used:
189 if (identical(bp[[i]], part[[j]])) {
190 n[i] <- attr(part, "number")[j]
199 boot.phylo <- function(phy, x, FUN, B = 100, block = 1,
200 trees = FALSE, quiet = FALSE, rooted = FALSE)
202 if (is.list(x) && !is.data.frame(x)) {
203 if (inherits(x, "DNAbin")) x <- as.matrix(x)
209 x <- matrix(x, n, nL/n, byrow = TRUE)
213 boot.tree <- vector("list", B)
214 if (!quiet) # suggestion by Alastair Potts
215 progbar <- utils::txtProgressBar(style = 3)
218 y <- seq(block, ncol(x), block)
219 boot.i <- sample(y, replace = TRUE)
220 boot.samp <- numeric(ncol(x))
221 boot.samp[y] <- boot.i
222 for (j in 1:(block - 1))
223 boot.samp[y - j] <- boot.i - j
224 } else boot.samp <- sample(ncol(x), replace = TRUE)
225 boot.tree[[i]] <- FUN(x[, boot.samp])
226 if (!quiet) utils::setTxtProgressBar(progbar, i/B)
228 if (!quiet) close(progbar)
229 for (i in 1:B) storage.mode(boot.tree[[i]]$Nnode) <- "integer"
230 storage.mode(phy$Nnode) <- "integer"
232 pp <- prop.part(boot.tree)
233 if (!rooted) pp <- postprocess.prop.part(pp)
234 ans <- prop.clades(phy, part = pp, rooted = rooted)
236 ##ans <- attr(.Call("prop_part", c(list(phy), boot.tree),
237 ## B + 1, FALSE, PACKAGE = "ape"), "number") - 1
239 class(boot.tree) <- "multiPhylo"
240 ans <- list(BP = ans, trees = boot.tree)
245 ### The next function transforms an object of class "prop.part" so
246 ### that the vectors which are identical in terms of split are aggregated.
247 ### For instance if n = 5 tips, 1:2 and 3:5 actually represent the same
248 ### split though they are different clades. The aggregation is done
249 ### arbitrarily. The call to ONEwise() insures that all splits include
251 postprocess.prop.part <- function(x)
255 w <- attr(x, "number")
259 for (i in 2:(N - 1)) {
262 for (j in (i + 1):N) {
265 if (length(A) + length(B) != n) next
276 labels <- attr(x, "labels")
279 attr(x, "number") <- w
280 attr(x, "labels") <- labels
281 class(x) <- "prop.part"
286 ### This function changes an object of class "prop.part" so that they
287 ### all include the first tip. For instance if n = 5 tips, 3:5 is
289 ONEwise <- function(x)
293 for (i in 2:length(x)) {
295 if (y[1] != 1) x[[i]] <- v[-y]
300 consensus <- function(..., p = 1, check.labels = TRUE)
302 foo <- function(ic, node) {
304 ## node: node number in the final tree
307 for (j in (ic + 1):m) {
308 wh <- match(pp[[j]], pool)
309 if (!any(is.na(wh))) {
310 edge[pos, 1] <<- node
312 edge[pos, 2] <<- nextnode <<- nextnode + 1L
320 ind <- pos:(pos + size - 1)
321 edge[ind, 1] <<- node
322 edge[ind, 2] <<- pool
327 if (length(obj) == 1) {
328 ## better than unlist(obj, recursive = FALSE)
329 ## because "[[" keeps the class of 'obj':
331 if (class(obj) == "phylo") return(obj)
333 if (!is.null(attr(obj, "TipLabel")))
334 labels <- attr(obj, "TipLabel")
336 labels <- obj[[1]]$tip.label
337 if (check.labels) obj <- .compressTipLabel(obj)
340 ## Get all observed partitions and their frequencies:
341 pp <- prop.part(obj, check.labels = FALSE)
342 ## Drop the partitions whose frequency is less than 'p':
343 if (p == 0.5) p <- 0.5000001 # avoid incompatible splits
344 pp <- pp[attr(pp, "number") >= p * ntree]
345 ## Get the order of the remaining partitions by decreasing size:
346 ind <- sort(unlist(lapply(pp, length)), decreasing = TRUE,
347 index.return = TRUE)$ix
351 edge <- matrix(0L, n + m - 1, 2)
360 structure(list(edge = edge, tip.label = labels,
361 Nnode = m), class = "phylo")