/archive/source.archive/bm.and.ou/trait.rjmcmc.10022010.R

http://github.com/eastman/auteur · R · 679 lines · 583 code · 69 blank · 27 comment · 105 complexity · 079f4e060120ba67d63151052ee5e759 MD5 · raw file 

    

  1. ## Written by LJ HARMON & AH: 2008

    2. 

  2. ## Modified by JM EASTMAN: 09.2010

    3. 

  3. 

    4. 

  4. rjMCMC.trait<-function (phy, data, ngen=1000, sampleFreq=100, model=c("OU","BM"), probAlpha=0.05, probMergeSplit=0.05, probRoot=0.1, theta.rates.ratio=1, heat=1, parallel=FALSE, fileBase="result") 

    5. 

  5. {

    6. 

  6. 	if(length(model)>1)stop(paste("Please specify either ", sQuote("OU"), " or ", sQuote("BM")))

    7. 

  7. ## MORE ERROR CHECKING NEEDED HERE ## 

    8. 

  8. 	if(parallel) require(multicore)

    9. 

  9. 	require(ouch)

    10. 

  10. 	require(geiger)

    11. 

  11. 	

    12. 

  12. ### prepare objects for rjMCMC

    13. 

  13. 	cur.model		<- model								

    14. 

  14. 

    15. 

  15. 	dataList		<- prepare.ouchdata(phy, data)			

    16. 

  16. 	ape.tre			<- cur.tre <- dataList$ape.tre			

    17. 

  17. 	orig.dat		<- dataList$orig.dat						

    18. 

  18. 	ouch.tre		<- dataList$ouch.tre					

    19. 

  19. 	ouch.dat		<- dataList$ouch.dat						

    20. 

  20. 	nn				<- length(ape.tre$edge.length)	

    21. 

  21. 	node.des		<- sapply(unique(c(ape.tre$edge[1,1],ape.tre$edge[,2])), function(x) get.descendants.of.node(x, ape.tre))

    22. 

  22. 	names(node.des) <- c(ape.tre$edge[1,1], unique(ape.tre$edge[,2]))

    23. 

  23. 	ll <- list()										

    24. 

  24. 

    25. 

  25. # initialize parameters

    26. 

  26. 	init.rate		<- generate.starting.point(data,ape.tre,node.des,theta=FALSE)

    27. 

  27.     cur.rates		<- init.rate$values			

    28. 

  28. 	cur.delta.rates	<- init.rate$delta

    29. 

  29. 	cur.root		<- adjust.value(mean(orig.dat))

    30. 

  30. 	cur.alpha		<- adjust.rate(0.001)

    31. 

  31. 	init.theta		<- generate.starting.point(data,ape.tre,node.des,theta=TRUE)

    32. 

  32. 	cur.theta		<- init.theta$values							

    33. 

  33. 	cur.delta.theta	<- init.theta$delta	

    34. 

  34. 	cur.root.theta	<- assign.root.theta(cur.theta,ape.tre)

    35. 

  35. 	cur.regimes		<- as.factor(c(cur.root.theta,cur.theta)) 

    36. 

  36. 	cur.tre			<- apply.BMM.to.apetree(ape.tre, cur.rates)

    37. 

  37. 	if(model=="OU")	{

    38. 

  38. 		mod.cur = new.ou.lik.fn(cur.rates, cur.alpha, cur.regimes, cur.theta, cur.tre, ouch.dat)

    39. 

  39. 	} else {

    40. 

  40. 		mod.cur = new.bm.lik.fn(cur.rates, cur.root, cur.tre, orig.dat)

    41. 

  41. 	}

    42. 

  42. 

    43. 

  43. 

    44. 

  44. # proposal counts

    45. 

  45. 	nRateProp =		0	

    46. 

  46. 	nRateSwapProp = 0									

    47. 

  47. 	nRcatProp =		0										

    48. 

  48. 	nRootProp =		0

    49. 

  49. 	nAlphaProp =	0										

    50. 

  50. 	nTcatProp =		0										

    51. 

  51. 	nThetaProp =	0

    52. 

  52. 	nThetaSwapProp= 0

    53. 

  53. 	nRateOK =		0											

    54. 

  54. 	nRateSwapOK =	0

    55. 

  55. 	nRcatOK =		0											

    56. 

  56. 	nRootOK =		0

    57. 

  57. 	nAlphaOK =		0										

    58. 

  58. 	nTcatOK =		0

    59. 

  59. 	nThetaOK =		0

    60. 

  60. 	nThetaSwapOK =	0

    61. 

  61. 

    62. 

  62. 	cOK=c(												

    63. 

  63. 		nRateOK,

    64. 

  64. 		nRateSwapOK,

    65. 

  65. 		nRcatOK, 

    66. 

  66. 		nRootOK, 

    67. 

  67. 		nAlphaOK,

    68. 

  68. 		nThetaOK,

    69. 

  69. 		nThetaSwapOK,

    70. 

  70. 		nTcatOK 

    71. 

  71. 	)		

    72. 

  72. 	

    73. 

  73. 	tickerFreq=ceiling(ngen/30)

    74. 

  74. 	

    75. 

  75. # file handling

    76. 

  76. 	errorLog=file(paste(getwd(),paste(cur.model, fileBase, "rjTraitErrors.log",sep="."),sep="/"),open='w+')

    77. 

  77. 	generate.error.message(i=NULL, mod.cur=NULL, mod.new=NULL, lnR=NULL, errorLog=errorLog, init=TRUE)

    78. 

  78. 	runLog=file(paste(getwd(),paste(cur.model, fileBase, "rjTrait.log",sep="."),sep="/"),open='w+')

    79. 

  79. 	generate.log(bundled.parms=NULL, cur.model, file=runLog, init=TRUE)

    80. 

  80. 

    81. 

  81. ### Begin rjMCMC

    82. 

  82.     for (i in 1:ngen) {

    83. 

  83.         lnLikelihoodRatio <- lnHastingsRatio <- lnPriorRatio <- 0

    84. 

  84. 		startparms = c(nRateProp, nRateSwapProp, nRcatProp, nRootProp, nAlphaProp, nThetaProp, nThetaSwapProp, nTcatProp)

    85. 

  85. 	

    86. 

  86. ## OU IMPLEMENTATION ## 

    87. 

  87. 		if(cur.model=="OU") {

    88. 

  88. 			if (runif(1) < (2 * probMergeSplit)) {

    89. 

  89. 				if(runif(1)<(theta.rates.ratio/(theta.rates.ratio+1))) {	# adjust theta categories

    90. 

  90. 					nr=split.or.merge(cur.delta.theta, cur.theta, ape.tre, node.des, theta=TRUE)

    91. 

  91. 					new.alpha=cur.alpha

    92. 

  92. 					new.theta=nr$new.values ##

    93. 

  93. 					new.delta.theta=nr$new.delta ##

    94. 

  94. 					new.root.theta<-nrt<-assign.root.theta(new.theta, ape.tre) ## 

    95. 

  95. 					new.regimes=as.factor(c(nrt,new.theta)) ##

    96. 

  96. 					new.rates=cur.rates 

    97. 

  97. 					new.delta.rates=cur.delta.rates 

    98. 

  98. 					nTcatProp=nTcatProp+1

    99. 

  99. 				} else {							# adjust rate categories

    100. 

  100. 					nr=split.or.merge(cur.delta.rates, cur.rates, ape.tre, node.des, theta=FALSE)

    101. 

  101. 					new.alpha=cur.alpha

    102. 

  102. 					new.root.theta=cur.root.theta

    103. 

  103. 					new.theta=cur.theta

    104. 

  104. 					new.delta.theta=cur.delta.theta

    105. 

  105. 					new.regimes=cur.regimes

    106. 

  106. 					new.rates=nr$new.values ##

    107. 

  107. 					new.delta.rates=nr$new.delta ##

    108. 

  108. 					nRcatProp=nRcatProp+1

    109. 

  109. 				}

    110. 

  110. 				lnHastingsRatio=nr$lnHastingsRatio

    111. 

  111. 				lnPriorRatio=nr$lnPriorRatio

    112. 

  112. 			} else { # neither split nor merge

    113. 

  113. 				if(runif(1)<probAlpha) {			# adjust alpha

    114. 

  114. 					new.alpha=adjust.rate(cur.alpha) ##

    115. 

  115. 					new.root.theta=cur.root.theta

    116. 

  116. 					new.theta=cur.theta

    117. 

  117. 					new.delta.theta=cur.delta.theta

    118. 

  118. 					new.regimes=cur.regimes 

    119. 

  119. 					new.rates=cur.rates

    120. 

  120. 					new.delta.rates=cur.delta.rates

    121. 

  121. 					nAlphaProp=nAlphaProp+1

    122. 

  122. 				} else { 

    123. 

  123. 					if(runif(1)>(theta.rates.ratio/(theta.rates.ratio+1))) {

    124. 

  124. 						if(runif(1)<0.2 | sum(cur.delta.rates)==0) {

    125. 

  125. 							new.alpha=cur.alpha		# adjust rates

    126. 

  126. 							new.root.theta=cur.root.theta

    127. 

  127. 							new.theta=cur.theta

    128. 

  128. 							new.delta.theta=cur.delta.theta

    129. 

  129. 							new.regimes=cur.regimes 

    130. 

  130. 							new.rates=adjust.rate(cur.rates) ##

    131. 

  131. 							new.delta.rates=cur.delta.rates

    132. 

  132. 							nRateProp=nRateProp+1

    133. 

  133. 						} else {					# swap rates

    134. 

  134. 							nr=switcheroo(cur.delta.rates, cur.rates, ape.tre, node.des, theta=FALSE)

    135. 

  135. 							new.alpha=cur.alpha		

    136. 

  136. 							new.root.theta=cur.root.theta

    137. 

  137. 							new.theta=cur.theta

    138. 

  138. 							new.delta.theta=cur.delta.theta

    139. 

  139. 							new.regimes=cur.regimes 

    140. 

  140. 							new.rates=nr$new.values ##

    141. 

  141. 							new.delta.rates=nr$new.delta ##

    142. 

  142. 							nRateSwapProp=nRateSwapProp+1

    143. 

  143. 						}

    144. 

  144. 					} else {	

    145. 

  145. 						if(runif(1)<0.2 | sum(cur.delta.theta)==0) {			# adjust theta

    146. 

  146. 							new.alpha=cur.alpha 

    147. 

  147. 							new.theta=adjust.value(cur.theta) ##

    148. 

  148. 							new.delta.theta=cur.delta.theta

    149. 

  149. 							new.root.theta<-nrt<-assign.root.theta(new.theta, ape.tre) ## 

    150. 

  150. 							new.regimes=as.factor(c(nrt,new.theta)) ##

    151. 

  151. 							new.rates=cur.rates

    152. 

  152. 							new.delta.rates=cur.delta.rates

    153. 

  153. 							nThetaProp=nThetaProp+1	

    154. 

  154. 						} else {					# swap theta

    155. 

  155. 							nr=switcheroo(cur.delta.theta, cur.theta, ape.tre, node.des, theta=TRUE)

    156. 

  156. 							new.alpha=cur.alpha 

    157. 

  157. 							new.theta=nr$new.values ##

    158. 

  158. 							new.delta.theta=nr$new.delta ##

    159. 

  159. 							new.root.theta<-nrt<-assign.root.theta(new.theta, ape.tre) ## 

    160. 

  160. 							new.regimes=as.factor(c(nrt,new.theta)) ##

    161. 

  161. 							new.rates=cur.rates

    162. 

  162. 							new.delta.rates=cur.delta.rates

    163. 

  163. 							nThetaSwapProp=nThetaSwapProp+1	

    164. 

  164. 						}

    165. 

  165. 					}

    166. 

  166. 				}

    167. 

  167. 			}

    168. 

  168. 			

    169. 

  169. #			cur.tre=apply.BMM.to.apetree(ape.tre, cur.rates)

    170. 

  170. 			new.tre=apply.BMM.to.apetree(ape.tre, new.rates)

    171. 

  171. 			if(!parallel) {

    172. 

  172. #				mod.cur=new.ou.lik.fn(cur.rates, cur.alpha, cur.regimes, cur.theta, cur.tre, ouch.dat)

    173. 

  173. 				mod.new=new.ou.lik.fn(new.rates, new.alpha, new.regimes, new.theta, new.tre, ouch.dat)

    174. 

  174. 			} else {

    175. 

  175. 				mod.cur=list(cur.rates, cur.alpha, cur.regimes, cur.theta, cur.tre, ouch.dat)

    176. 

  176. 				mod.new=list(new.rates, new.alpha, new.regimes, new.theta, new.tre, ouch.dat)

    177. 

  177. 				res=mclapply(list(mod.cur, mod.new), function(x) new.ou.lik.fn(x[[1]], x[[2]], x[[3]], x[[4]], x[[5]], x[[6]]))

    178. 

  178. 				res[[1]]->mod.cur

    179. 

  179. 				res[[2]]->mod.new

    180. 

  180. 			}

    181. 

  181. 		} else if(cur.model=="BM"){

    182. 

  182. ## BM IMPLEMENTATION ##

    183. 

  183. 			if (runif(1) < (2 * probMergeSplit)) {							# adjust rate categories

    184. 

  184. 				nr=split.or.merge(cur.delta.rates, cur.rates, ape.tre, node.des, theta=FALSE)

    185. 

  185. 				new.rates=nr$new.values

    186. 

  186. 				new.delta.rates=nr$new.delta

    187. 

  187. 				new.root=cur.root

    188. 

  188. 				nRcatProp=nRcatProp+1

    189. 

  189. 				lnHastingsRatio=nr$lnHastingsRatio

    190. 

  190. 				lnPriorRatio=nr$lnPriorRatio

    191. 

  191. 			} else { 

    192. 

  192. 				if(runif(1)<probRoot) {										# adjust root

    193. 

  193. 					new.root=adjust.value(cur.root)

    194. 

  194. 					new.rates=cur.rates

    195. 

  195. 					new.delta.rates=cur.delta.rates

    196. 

  196. 					nRootProp=nRootProp+1

    197. 

  197. 				} else {													# adjust rates

    198. 

  198. 					if(runif(1)<0.05 | sum(cur.delta.rates)==0) {

    199. 

  199. 						new.rates=adjust.rate(cur.rates)

    200. 

  200. 						new.delta.rates=cur.delta.rates

    201. 

  201. 						new.root=cur.root

    202. 

  202. 						nRateProp=nRateProp+1

    203. 

  203. 					} else {												# swap rate classes

    204. 

  204. 						nr=switcheroo(cur.delta.rates, cur.rates, ape.tre, node.des, theta=FALSE)

    205. 

  205. 						new.rates=nr$new.values ##

    206. 

  206. 						new.delta.rates=nr$new.delta ##

    207. 

  207. 						new.root=cur.root

    208. 

  208. 						nRateSwapProp=nRateSwapProp+1

    209. 

  209. 					}

    210. 

  210. 				}

    211. 

  211. 			}

    212. 

  212. 			

    213. 

  213. #			cur.tre=apply.BMM.to.apetree(ape.tre, cur.rates)

    214. 

  214. 			new.tre=apply.BMM.to.apetree(ape.tre, new.rates)

    215. 

  215. 			if(!parallel) {

    216. 

  216. #				mod.cur=new.bm.lik.fn(cur.rates, cur.root, cur.tre, orig.dat)

    217. 

  217. 				mod.new=new.bm.lik.fn(new.rates, new.root, new.tre, orig.dat)

    218. 

  218. 			} else {

    219. 

  219. 				mod.cur=list(cur.rates, cur.root, cur.tre, orig.dat)

    220. 

  220. 				mod.new=list(new.rates, new.root, new.tre, orig.dat)

    221. 

  221. 				res=mclapply(list(mod.cur, mod.new), function(x) new.bm.lik.fn(x[[1]], x[[2]], x[[3]], x[[4]]))

    222. 

  222. 				res[[1]]->mod.cur

    223. 

  223. 				res[[2]]->mod.new

    224. 

  224. 			}

    225. 

  225. 			if(is.infinite(mod.new$lnL)) {

    226. 

  226. 				failures="./failed.model.parameters/"

    227. 

  227. 				failed.trees=paste(failures, "trees", sep="/")

    228. 

  228. 				lapply(list(failures, failed.trees), function(x) if(!file.exists(x)) dir.create(x))

    229. 

  229. 				

    230. 

  230. 				pdf(file=paste(failed.trees, paste(i,"failed.model.pdf",sep="."),sep="/"))

    231. 

  231. 					plot(new.tre)

    232. 

  232. 				dev.off()

    233. 

  233. 				write.table(i,file=paste(failures, "failed.rates.txt",sep="/"),append=TRUE,col=FALSE,row=FALSE,quote=FALSE)

    234. 

  234. 				write.table(matrix(new.rates,nrow=1),file=paste(failures,"failed.rates.txt",sep="/"),append=TRUE,col=FALSE,row=FALSE,quote=FALSE)

    235. 

  235. 				write.table(matrix(c(i, new.root),nrow=1),file=paste(failures,"failed.root.txt",sep="/"),append=TRUE,col=FALSE,row=FALSE,quote=FALSE)

    236. 

  236. 			}

    237. 

  237. 		}

    238. 

  238. 		

    239. 

  239. 		endparms = c(nRateProp, nRateSwapProp, nRcatProp, nRootProp, nAlphaProp, nThetaProp, nThetaSwapProp, nTcatProp)

    240. 

  240. 		

    241. 

  241. 		if(!inherits(mod.new, "try-error")) {

    242. 

  242. 			lnLikelihoodRatio = mod.new$lnL - mod.cur$lnL

    243. 

  243. 		} else {

    244. 

  244. 			lnLikelihoodRatio = -Inf

    245. 

  245. 			mod.new$lnL=NaN

    246. 

  246. 		}

    247. 

  247. 		r=assess.lnR((heat * lnLikelihoodRatio + heat * lnPriorRatio + lnHastingsRatio)->lnR)

    248. 

  248. 

    249. 

  249. 		if (runif(1) <= r$r) {			# adopt proposal

    250. 

  250. 			if(cur.model=="OU") {

    251. 

  251. 				decision="adopt"

    252. 

  252. 				cur.alpha <- new.alpha

    253. 

  253. 				cur.root.theta <- new.root.theta

    254. 

  254. 				cur.regimes <- new.regimes

    255. 

  255. 				cur.theta <- new.theta

    256. 

  256. 				cur.delta.theta <- new.delta.theta

    257. 

  257. 			} else {

    258. 

  258. 				decision="adopt"

    259. 

  259. 				cur.root <- new.root

    260. 

  260. 			}

    261. 

  261. 			cur.rates <- new.rates

    262. 

  262. 			cur.delta.rates <- new.delta.rates

    263. 

  263. 			curr.lnL <- mod.new$lnL

    264. 

  264. 			cur.tre <- new.tre

    265. 

  265. 			cOK=determine.accepted.proposal(startparms, endparms, cOK)

    266. 

  266. 		} else {						# deny proposal	

    267. 

  267. 			decision="reject"

    268. 

  268. 			curr.lnL <- mod.cur$lnL 

    269. 

  269. 		}

    270. 

  270. 		

    271. 

  271. 		if(is.infinite(curr.lnL)) stop("starting point has exceptionally poor likelihood")

    272. 

  272. 

    273. 

  273. 		if(r$error) generate.error.message(i, mod.cur, mod.new, lnR, errorLog)

    274. 

  274. 		

    275. 

  275. 		if(i%%tickerFreq==0) {

    276. 

  276. 			if(i==tickerFreq) cat("|",rep(" ",9),toupper("generations complete"),rep(" ",9),"|","\n")

    277. 

  277. 			cat(". ")

    278. 

  278. 		}

    279. 

  279. 		

    280. 

  280. 		if(i%%sampleFreq==0) {

    281. 

  281. 			bundled.parms=list(gen=i, mrate=exp(mean(log(cur.rates))), cats=sum(cur.delta.rates)+1, root=cur.root, alpha=cur.alpha, reg=(sum(cur.delta.theta)+1), theta=mean(cur.theta), lnL=curr.lnL)

    282. 

  282. 			generate.log(bundled.parms, cur.model, file=runLog)

    283. 

  283. 			ll$lnL=rbind(ll$lnL, curr.lnL)

    284. 

  284. 			ll$rates=rbind(ll$rates, cur.rates) 

    285. 

  285. 			if(cur.model=="OU") {

    286. 

  286. 				ll$theta=rbind(ll$theta, cur.theta)

    287. 

  287. 				ll$alpha=rbind(ll$alpha, cur.alpha)

    288. 

  288. 				ll$root=rbind(ll$root, cur.root.theta)

    289. 

  289. 			}

    290. 

  290. 			if(cur.model=="BM") {

    291. 

  291. 				ll$root=rbind(ll$root, cur.root)

    292. 

  292. 			}

    293. 

  293. 			if(i==ngen) {

    294. 

  294. 				if(cur.model=="OU") dimnames(ll$theta)[[2]]=ape.tre$edge[,2]

    295. 

  295. 				dimnames(ll$rates)[[2]]=ape.tre$edge[,2]

    296. 

  296. 			}

    297. 

  297. 		}

    298. 

  298. 	}

    299. 

  299. 	

    300. 

  300. # End rjMCMC

    301. 

  301. 	close(errorLog)

    302. 

  302. 	close(runLog)

    303. 

  303. 	summarize.run(cOK, endparms, cur.model)	

    304. 

  304. 	return(list(results=ll,data=orig.dat,tree=ape.tre))

    305. 

  305. }

    306. 

  306. 

    307. 

  307. 

    308. 

  308. ## AUXILIARY FUNCTIONS

    309. 

  309. 

    310. 

  310. 

    311. 

  311. new.ou.lik.fn <- function(rates, alpha, regimes, theta, ape.tre, ouch.dat) { # from OUCH 2.7-1:::hansen()

    312. 

  312. # mod 09.13.2010 JM Eastman

    313. 

  313. 	ouch.tre=ape2ouch(ape.tre, scale=FALSE)

    314. 

  314. 	ouch.dat$regimes=as.factor(regimes)

    315. 

  315. 	theta.tmp=as.numeric(levels(as.factor(theta)))

    316. 

  316. 	theta=theta.tmp[order(match(theta.tmp,theta))]

    317. 

  317. 	alpha.ouch=as.matrix(sqrt(alpha))

    318. 

  318. 	beta=ouch:::regime.spec(ouch.tre, ouch.dat['regimes'])

    319. 

  319. 	dat=do.call(c,lapply(ouch.dat['trait'],function(y)y[ouch.tre@term]))

    320. 

  320. 	names(dat)=ouch.dat$labels[Ntip(ape.tre):nrow(ouch.dat)]

    321. 

  321. 	n <- length(dat)

    322. 

  322. 	otree.df=as(ouch.tre,"data.frame")

    323. 

  323. 	ordering=otree.df[n:nrow(otree.df),'labels']

    324. 

  324. 	ev <- eigen(alpha.ouch,symmetric=TRUE)

    325. 

  325. 	w <- .Call(ouch:::ouch_weights,object=ouch.tre,lambda=ev$values,S=ev$vectors,beta=beta)

    326. 

  326. 	v <- geiger:::ouMatrix(vcv.phylo(ape.tre), alpha)

    327. 

  327. 	v.match <- match(row.names(v),ordering)

    328. 

  328. 	v <- v[v.match,v.match]

    329. 

  329. 		## GEIGER call does not assume ultrametricity (and thus allows different 'rates' under BM)

    330. 

  330. 		# -old-	v <- .Call(ouch:::ouch_covar,object=ouch.tre,lambda=ev$values,S=ev$vectors,sigma.sq=1)

    331. 

  331. 	

    332. 

  332. 	y <- matrix(theta,ncol=1)

    333. 

  333. 	e <- w%*%y-dat

    334. 

  334. 	dim(y) <- dim(y)[1]

    335. 

  335. 	dim(e) <- n

    336. 

  336. 	q <- e%*%solve(v,e)

    337. 

  337. 	det.v <- determinant(v,logarithm=TRUE)

    338. 

  338. 	if (det.v$sign!=1)stop("ou.lik.fn error: non-positive determinant",call.=FALSE)

    339. 

  339. 	dev <- n*log(2*pi)+as.numeric(det.v$modulus)+q[1,1]

    340. 

  340. 	list(

    341. 

  341. 		 theta=theta,

    342. 

  342. 		 weight=w,

    343. 

  343. 		 vcov=v,

    344. 

  344. 		 resids=e,

    345. 

  345. 		 lnL=-0.5*dev

    346. 

  346. 		 )

    347. 

  347. }

    348. 

  348. 

    349. 

  349. new.bm.lik.fn <- function(rates, root, ape.tre, orig.dat) { # from OUCH 2.7-1:::brown()	

    350. 

  350. # mod 09.16.2010 JM Eastman

    351. 

  351. 

    352. 

  352. 	tree=ape.tre

    353. 

  353. 	data=orig.dat

    354. 

  354. 

    355. 

  355. 	nterm=Ntip(tree)

    356. 

  356. 	nchar=length(data)

    357. 

  357. 	dat=data[match(tree$tip.label,names(data))]

    358. 

  358. 

    359. 

  359. 	nchar <- 1

    360. 

  360. 	w <- matrix(data=1,nrow=nterm,ncol=1)

    361. 

  361. 	n <- length(dat)

    362. 

  362. 

    363. 

  363. 	b <- vcv.phylo(ape.tre)

    364. 

  364. 

    365. 

  365. 	y <- matrix(root)

    366. 

  366. 	e <- w%*%y-dat

    367. 

  367. 	q <- t(e)%*%solve(b,e)

    368. 

  368. 	v <- q/nterm

    369. 

  369. 	dev <- nchar*nterm*(1+log(2*pi))+nchar*log(det(b))+nterm*log(det(v))

    370. 

  370. 	list(

    371. 

  371. 		 root=root,

    372. 

  372. 		 lnL = -0.5*dev

    373. 

  373. 		 )	

    374. 

  374. }

    375. 

  375. 

    376. 

  376. split.or.merge <- function(cur.delta, cur.values, phy, node.des, theta=FALSE) { 

    377. 

  377. # mod 09.17.2010 JM Eastman

    378. 

  378. 	bb=cur.delta

    379. 

  379. 	vv=cur.values

    380. 

  380. 	names(vv)<-names(bb)<-phy$edge[,2]

    381. 

  381. 	new.bb=choose.one(bb)

    382. 

  382. 	new.vv=vv

    383. 

  383. 	

    384. 

  384. 	s=names(new.bb[bb!=new.bb])

    385. 

  385. 	all.shifts=as.numeric(names(bb[bb>0]))

    386. 

  386. 	all.D=node.des[[which(names(node.des)==s)]]

    387. 

  387. 	if(length(all.D)!=0) {

    388. 

  388. 		untouchable=unlist(lapply(all.shifts[all.shifts>s], function(x)node.des[[which(names(node.des)==x)]]))

    389. 

  389. 		remain.unchanged=union(all.shifts, untouchable)

    390. 

  390. 	} else {

    391. 

  391. 		untouchable=NULL

    392. 

  392. 		remain.unchanged=list()

    393. 

  393. 	}

    394. 

  394. 	

    395. 

  395. 	marker=match(s, names(new.vv))

    396. 

  396. 	nn=length(vv)

    397. 

  397. 	K=sum(bb)+1

    398. 

  398. 	N=Ntip(phy)

    399. 

  399. 	

    400. 

  400. 	if(sum(new.bb)>sum(bb)) {			# add transition: SPLIT

    401. 

  401. #		print("s")

    402. 

  402. 		if(theta) {							# assign new theta to current node

    403. 

  403. 			new.vv[marker] <- nr <- adjust.value(new.vv[marker])

    404. 

  404. 		} else {							# assign new rate to current node

    405. 

  405. 			new.vv[marker] <- nr <- adjust.rate(new.vv[marker])

    406. 

  406. 		}

    407. 

  407. 		if(length(remain.unchanged)==0) {	# assign new value to all descendants

    408. 

  408. 			new.vv[match(all.D,names(new.vv))] = nr

    409. 

  409. 		} else {							# assign new value to all 'open' descendants 

    410. 

  410. 			new.vv[match(all.D[!(all.D%in%remain.unchanged)],names(new.vv))] = nr

    411. 

  411. 		}

    412. 

  412. 		lnHastingsRatio = log((K+1)/(2*N-2-K)) ### from Drummond and Suchard 2010: where N is tips, K is number of local parms in tree

    413. 

  413. 		lnPriorRatio = log(ppois(K+1,log(2),lower.tail=FALSE)/ppois(K,log(2),lower.tail=FALSE))

    414. 

  414. 	} else {							# drop transition: MERGE

    415. 

  415. #		print("m")

    416. 

  416. 		anc = get.ancestor.of.node(s, phy)

    417. 

  417. 		if(!is.root(anc, phy)) {			# substitute ancestral value for current

    418. 

  418. 			new.vv[match(s, names(new.vv))] <- nr <- new.vv[match(anc, names(new.vv))] 

    419. 

  419. 		} else {

    420. 

  420. 			sister.tmp=get.desc.of.node(anc,phy)

    421. 

  421. 			sister=sister.tmp[sister.tmp!=s]

    422. 

  422. 			if(bb[match(sister, names(bb))]==0) {

    423. 

  423. 				nr=as.numeric(vv[match(sister, names(vv))])

    424. 

  424. 			} else {

    425. 

  425. 				if(theta) {						# assign new theta to first descendant of root

    426. 

  426. 					new.vv[marker] <- nr <- adjust.value(new.vv[marker])

    427. 

  427. 				} else {						# assign new rate to first descendant of root

    428. 

  428. 					new.vv[marker] <- nr <- adjust.rate(new.vv[marker])

    429. 

  429. 				}

    430. 

  430. 			}

    431. 

  431. 		}

    432. 

  432. 		if(length(remain.unchanged)==0) {	# assign new value to all descendants

    433. 

  433. 			new.vv[match(all.D, names(new.vv))] = nr

    434. 

  434. 		} else {							# assign new value to all 'open' descendants

    435. 

  435. 			new.vv[match(all.D[!(all.D%in%remain.unchanged)],names(new.vv))] = nr

    436. 

  436. 		}

    437. 

  437. 		lnHastingsRatio = log((2*N-2-K+1)/K) ### from Drummond and Suchard 2010: where N is tips, K is number of local parms in tree

    438. 

  438. 		lnPriorRatio = log(ppois(K-1,log(2),lower.tail=FALSE)/ppois(K,log(2),lower.tail=FALSE))

    439. 

  439. 		

    440. 

  440. 	}

    441. 

  441. 	if(((length(table(new.vv))-1)-sum(new.bb))!=0) fail=TRUE else fail=FALSE

    442. 

  442. 	return(list(new.delta=new.bb, new.values=new.vv, lnHastingsRatio=lnHastingsRatio, lnPriorRatio=lnPriorRatio, fail=fail))

    443. 

  443. }

    444. 

  444. 

    445. 

  445. switcheroo <- function(cur.delta, cur.values, phy, node.des, theta=FALSE) { 

    446. 

  446. 	bb=cur.delta

    447. 

  447. 	vv=cur.values

    448. 

  448. 	new.bb=sample(bb)

    449. 

  449. 	names(vv)<-names(bb)<-names(new.bb)<-phy$edge[,2]

    450. 

  450. 	new.vv=vv

    451. 

  451. 	

    452. 

  452. 	old.shifts=as.numeric(names(bb[bb!=0]))

    453. 

  453. 	old.values=vv[match(old.shifts, names(vv))]

    454. 

  454. 	

    455. 

  455. 	anc=lapply(old.shifts, function(x)get.ancestor.of.node(x,phy))

    456. 

  456. 	if(!is.root(min(unlist(anc)),phy)) {

    457. 

  457. 	# find most ancestral value in tree

    458. 

  458. 		anc.value=unname(vv[match(min(unlist(anc)),names(vv))]) 

    459. 

  459. 	} else {

    460. 

  460. 		if(theta) anc.value=mean(sapply(old.values, function(x) adjust.value(x))) else anc.value=mean(sapply(old.values, function(x) adjust.rate(x)))

    461. 

  461. 	}

    462. 

  462. 	all.desc.anc=node.des[[which(names(node.des)==min(unlist(anc)))]]

    463. 

  463. 	new.vv[match(all.desc.anc,names(new.vv))] = anc.value

    464. 

  464. 

    465. 

  465. 	new.shifts=sort(as.numeric(names(new.bb[new.bb!=0])))

    466. 

  466. 	new.tips=new.shifts[new.shifts<=Ntip(phy)]

    467. 

  467. 	new.ints=new.shifts[new.shifts>Ntip(phy)]

    468. 

  468. 	new.shifts=c(new.ints, new.tips)

    469. 

  469. 	

    470. 

  470. 	new.values=as.numeric(old.values)[sample(1:length(old.values))]

    471. 

  471. 	for(z in 1:length(new.shifts)) {

    472. 

  472. 		node=new.shifts[z]

    473. 

  473. 		new.vv[match(node, names(new.vv))]=new.values[z]

    474. 

  474. 		new.vv[match(node.des[[which(names(node.des)==node)]],names(new.vv))]=new.values[z]

    475. 

  475. 	} 

    476. 

  476. 	return(list(new.delta=new.bb, new.values=new.vv))

    477. 

  477. }

    478. 

  478. 

    479. 

  479. generate.starting.point <- function(data, phy, node.des, theta=FALSE) { 

    480. 

  480. 	nn=length(phy$edge.length)

    481. 

  481. 	ntip=Ntip(phy)

    482. 

  482. 	nshifts=rtpois(1,log(ntip),nn)

    483. 

  483. 	if(nshifts!=0) bb=sample(c(rep(1,nshifts),rep(0,nn-nshifts)),replace=FALSE) else bb=rep(0,nn)

    484. 

  484. 	names(bb)=phy$edge[,2]

    485. 

  485. 	shifts=as.numeric(names(bb)[bb==1])

    486. 

  486. 	if(theta) {

    487. 

  487. 		min.max=c(adjust.value(min(data)), adjust.value(max(data)))

    488. 

  488. 		values=runif(sum(bb)+1, min=min.max[min.max==min(min.max)], max=min.max[min.max==max(min.max)])

    489. 

  489. 	} else {

    490. 

  490. 		init.rate=fitContinuous(phy,data)[[1]]$beta

    491. 

  491. 		min.max=c(adjust.rate(init.rate), adjust.rate(init.rate))

    492. 

  492. 		values=runif(sum(bb)+1, min=min.max[min.max==min(min.max)], max=min.max[min.max==max(min.max)])

    493. 

  493. 	}

    494. 

  494. 	internal.shifts<-tip.shifts<-numeric(0)

    495. 

  495. 	internal.shifts=sort(shifts[shifts>ntip])

    496. 

  496. 	tip.shifts=shifts[shifts<=ntip]

    497. 

  497. 	

    498. 

  498. 	if(length(internal.shifts)==0 & length(tip.shifts)==0) {

    499. 

  499. 		vv=rep(values, nn)

    500. 

  500. 	} else {

    501. 

  501. 		vv=bb

    502. 

  502. 		vv[]=values[length(values)]

    503. 

  503. 		i=0

    504. 

  504. 		if(length(internal.shifts)!=0) {

    505. 

  505. 			for(i in 1:length(internal.shifts)) {

    506. 

  506. 				d=node.des[which(names(node.des)==internal.shifts[i])]

    507. 

  507. #		d=get.descendants.of.node(internal.shifts[i],phy)

    508. 

  508. 				vv[match(c(internal.shifts[i], d), names(vv))]=values[i]

    509. 

  509. 			}

    510. 

  510. 		}

    511. 

  511. 		if(length(tip.shifts)!=0) {

    512. 

  512. 			for(j in 1:length(tip.shifts)) {

    513. 

  513. 				vv[match(tip.shifts[j], names(vv))]=values[j+i]

    514. 

  514. 			}

    515. 

  515. 		}

    516. 

  516. 	}

    517. 

  517. 	return(list(delta=unname(bb), values=unname(vv)))

    518. 

  518. }

    519. 

  519. 

    520. 

  520. assign.root.theta <- function(cur.theta, phy) {

    521. 

  521. 	root=phy$edge[1,1]

    522. 

  522. 	desc=phy$edge[which(phy$edge[,1]==root),2]

    523. 

  523. 	root.theta=cur.theta[sample(which(phy$edge[,2]==desc),1)]

    524. 

  524. 	root.theta

    525. 

  525. }

    526. 

  526. 

    527. 

  527. get.descendants.of.node <- function(node, phy) {

    528. 

  528. 	storage=list()

    529. 

  529. 	if(node%in%phy$edge[,1]) {

    530. 

  530. 		desc=c(sapply(node, function(x) {return(phy$edge[which(phy$edge[,1]==x),2])}))

    531. 

  531. 		while(any(desc%in%phy$edge[,1])) {

    532. 

  532. 			desc.true=desc[which(desc%in%phy$edge[,1])]

    533. 

  533. 			desc.desc=lapply(desc.true, function(x) return(get.desc.of.node(x, phy)))

    534. 

  534. 			

    535. 

  535. 			storage=list(storage,union(desc,desc.desc))

    536. 

  536. 			

    537. 

  537. 			desc=unlist(desc.desc)

    538. 

  538. 		}

    539. 

  539. 		storage=sort(unique(unlist(union(desc,storage))))

    540. 

  540. 	}

    541. 

  541. 	return(storage)

    542. 

  542. }

    543. 

  543. 

    544. 

  544. get.desc.of.node <- function(node, phy) {

    545. 

  545. 	return(phy$edge[which(phy$edge[,1]==node),2])

    546. 

  546. }

    547. 

  547. 

    548. 

  548. get.ancestor.of.node<-function(node, phy) {

    549. 

  549. 	anc=phy$edge[which(phy$edge[,2]==node),1]

    550. 

  550. 	anc

    551. 

  551. }

    552. 

  552. 

    553. 

  553. choose.one <- function(cur.delta)

    554. 

  554. {

    555. 

  555. 	bb=cur.delta

    556. 

  556. 	s=sample(1:length(bb),1)

    557. 

  557. 	bb[s]=1-bb[s]

    558. 

  558. 	bb

    559. 

  559. }

    560. 

  560. 

    561. 

  561. is.root<-function(node,phy) {

    562. 

  562. 	if(node==phy$edge[1,1]) return(TRUE) else return(FALSE)

    563. 

  563. }

    564. 

  564. 

    565. 

  565. assess.lnR <- function(lnR) {

    566. 

  566. 	if(is.na(lnR) || abs(lnR)==Inf) {

    567. 

  567. 		error=TRUE

    568. 

  568. 		r=0

    569. 

  569. 	} else {

    570. 

  570. 		if(lnR < -20) {

    571. 

  571. 			r=0 

    572. 

  572. 		} else if(lnR > 0) {

    573. 

  573. 			r=1 

    574. 

  574. 		} else {

    575. 

  575. 			r=exp(lnR)

    576. 

  576. 		}

    577. 

  577. 		error=FALSE

    578. 

  578. 	}

    579. 

  579. 	return(list(r=r, error=error))

    580. 

  580. }

    581. 

  581. 

    582. 

  582. apply.BMM.to.apetree<-function(apetree, rates) {

    583. 

  583. 	apetree$edge.length=apetree$edge.length*rates

    584. 

  584. 	return(apetree)

    585. 

  585. }

    586. 

  586. 

    587. 

  587. adjust.value <- function(value) {

    588. 

  588. 	vv=value

    589. 

  589. 	vv.levels=levels(factor(vv))

    590. 

  590. 	rch <- runif(length(vv.levels), min = -1, max = 1)

    591. 

  591. 	for(i in 1:length(vv.levels)){

    592. 

  592. 		vv[which(vv==vv.levels[i])]=vv[which(vv==vv.levels[i])]+rch[i]

    593. 

  593. 	}

    594. 

  594. 	vv

    595. 

  595. }

    596. 

  596. 

    597. 

  597. adjust.rate <- function(rate) {

    598. 

  598. 	vv=rate

    599. 

  599. 	v=log(vv)

    600. 

  600. 	vv.levels=levels(factor(vv))

    601. 

  601. 	rch <- runif(length(vv.levels), min = -1, max = 1)

    602. 

  602. 	for(i in 1:length(vv.levels)){

    603. 

  603. 		v[which(vv==vv.levels[i])]=v[which(vv==vv.levels[i])]+rch[i]

    604. 

  604. 	}

    605. 

  605. 	v=exp(v)

    606. 

  606. 	v

    607. 

  607. }

    608. 

  608. 	

    609. 

  609. prepare.ouchdata <- function(phy, data) {

    610. 

  610. 	td <- treedata(phy, data, sort = T)	

    611. 

  611. 	ouch.tre <- ape2ouch(td$phy->ape.tre, scale=FALSE)		

    612. 

  612. 	ouch.dat=data.frame(as.numeric(ouch.tre@nodes), as.character(ouch.tre@nodelabels))

    613. 

  613. 	names(ouch.dat)=c("nodes","labels")

    614. 

  614. 	ouch.dat[ouch.dat[,2]=="",2]=NA

    615. 

  615. 	ouch.dat$trait=NA

    616. 

  616. 	mM=match(ouch.dat$labels,names(data))

    617. 

  617. 	for(i in 1:nrow(ouch.dat)){

    618. 

  618. 		if(!is.na(mM[i])){

    619. 

  619. 			ouch.dat$trait[i]=data[mM[i]]	

    620. 

  620. 		}

    621. 

  621. 	}

    622. 

  622. 	return(list(ouch.tre=ouch.tre, ouch.dat=ouch.dat, ape.tre=td$phy, orig.dat=td$data[,1]))

    623. 

  623. }

    624. 

  624. 

    625. 

  625. summarize.run<-function(cOK, endparms, cur.model) {

    626. 

  626. #	end = c(nRateProp, nRateSwapProp, nRcatProp, nRootProp, nAlphaProp, nThetaProp, nThetaSwapProp, nTcatProp)

    627. 

  627. 

    628. 

  628. 	names(endparms)<-names(cOK)<-c("rates", "rate.swap", "rate.catg", "root", "alpha", "theta", "theta.swap", "theta.catg")

    629. 

  629. 	names(endparms)=paste("pr.",names(endparms),sep="")

    630. 

  630. 	names(cOK)=paste("ok.",names(cOK),sep="")

    631. 

  631. 	cat("\n")

    632. 

  632. 	if(cur.model=="BM") {

    633. 

  633. 		print(endparms[paste("pr.", c("rates", "rate.swap", "rate.catg", "root"),sep="")])

    634. 

  634. 		print(cOK[paste("ok.", c("rates", "rate.swap", "rate.catg", "root"),sep="")])

    635. 

  635. 	} else {

    636. 

  636. 		print(endparms[paste("pr.", c("rates", "rate.swap", "rate.catg", "alpha", "theta", "theta.swap", "theta.catg"),sep="")])

    637. 

  637. 		print(cOK[paste("ok.", c("rates", "rate.swap", "rate.catg", "alpha", "theta", "theta.swap", "theta.catg"),sep="")])

    638. 

  638. 	}

    639. 

  639. }

    640. 

  640. 

    641. 

  641. generate.log<-function(bundled.parms, cur.model, file, init=FALSE) {

    642. 

  642. #	bundled.parms=list(gen=i, mrate=mean(cur.rates), cats=sum(cur.delta.rates)+1, root=cur.root, alpha=cur.alpha, reg=sum(cur.delta.theta)+1, theta=mean(cur.theta), lnL=curr.lnL)

    643. 

  643. 	res=bundled.parms

    644. 

  644. 

    645. 

  645. 	if(init) {

    646. 

  646. 		if(cur.model=="BM") msg=paste("gen", "model", "mean.rate", "rates", "root", "lnL", sep="\t") else msg=paste("gen", "model", "mean.rate", "rates", "alpha", "regimes", "mean.theta", "lnL", sep="\t")

    647. 

  647. 	} else {

    648. 

  648. 		if(cur.model=="BM") {

    649. 

  649. 			msg<-paste(res$gen, sQuote(cur.model), sprintf("%.3f", res$mrate), res$cats, sprintf("%.3f", res$root), sprintf("%.3f", res$lnL),sep="\t")

    650. 

  650. 		} else {

    651. 

  651. 			msg<-paste(res$gen, sQuote(cur.model), sprintf("%.3f", res$mrate), res$cats, sprintf("%.4f", res$alpha), res$reg, sprintf("%.3f", res$theta), sprintf("%.3f", res$lnL),sep="\t")

    652. 

  652. 		}

    653. 

  653. 	}

    654. 

  654. 	write(msg, file=file, append=TRUE)

    655. 

  655. }

    656. 

  656. 

    657. 

  657. generate.error.message<-function(i, mod.cur, mod.new, lnR, errorLog, init=FALSE) {

    658. 

  658. 	if(init) {

    659. 

  659. 		initmsg = write(paste("gen", "curr.lnL", "new.lnL", "lnR", sep="\t"), file=errorLog)

    660. 

  660. 	} else {

    661. 

  661. 		write(paste(i, sprintf("%.3f", mod.cur$lnL), sprintf("%.3f", mod.new$lnL), sprintf("%.3f", lnR), sep="\t"), file=errorLog)

    662. 

  662. 	}

    663. 

  663. }

    664. 

  664. 

    665. 

  665. determine.accepted.proposal<-function(startparms, endparms, cOK) {

    666. 

  666. 	which(startparms!=endparms)->marker

    667. 

  667. 	cOK[marker]=cOK[marker]+1

    668. 

  668. 	cOK

    669. 

  669. }

    670. 

  670. 

    671. 

  671. rtpois<-function(N, lambda, k) {

    672. 

  672. 	p=ppois(k, lambda, lower.tail=FALSE)

    673. 

  673. 	out=qpois(runif(N, min=0, max=1-p), lambda)

    674. 

  674. 	out

    675. 

  675. }
    676.