R语言并行计算beta-NTI值

一、beta-NTI（nearest taxon index）简介

1.首先简单描述一下下图一，这棵树就是遗传发育树，每个圈代表一个OTU，数字可认为是OTU ID，每个颜色代表一个样点，其中灰色是我们实验观察到的结果，红色和蓝色是我们进行1000次在发育树上随机分配的结果。
2.我们可根据实验观察和随机分配的结果计算MNTD（mean nearest taxon distance），然后将观察的结果与模拟的结果进行比较，得到NTI。
3.这里介绍的是样点内的NTI，可以理解为alpha-NTI，理解了alpha-NTI，再理解beta-NTI也是同样的道理（见图二），只不过是（2）中的MNTD换成了beta-MNTD。
4.一定要注意，只有OTU具有遗传发育信号（见图三），才能用beta-NTI去估算构建过程是决定性的还是随机的。

图一
图二图三

二、系统发育信号（phylogenetic signal）检测

步骤一：
大致判断OTU栖息地偏好性与OTU遗传发育距离之间的关系

phylosignal <- function(otu_niche,otu_tree,nclass){
  library(picante)
  library(ape)

  prune_tree<-prune.sample(otu_niche,otu_tree)
  tip<-prune_tree$tip.label
  coln<-colnames(otu_niche)
  m<-NULL
  for(i in 1:length(coln)){
    if(!coln[i]%in%tip){
      print(coln[i])
      m<-cbind(m,coln[i])
    }
  }
  m<-as.vector(m)
  otu_niche<-otu_niche[,!colnames(otu_niche)%in%m]
  otu_phydist <- cophenetic(prune_tree)
  #otu_bray<-vegdist(log1p(otu_niche), method="bray")
  otu_bray<-vegdist(t(otu_niche), method="bray")
  otu_bray<-as.matrix(otu_bray)
  length(otu_bray) == length(otu_phydist)
  #mntd(otu_niche,cophenetic(prune_tree))
  otu_bray <- otu_bray[colnames(otu_phydist),colnames(otu_phydist)]
  data1 <- data.frame(cor = c(unlist(otu_bray)),phydist=c(unlist(otu_phydist)))
  colnames(data1) <- c("sorted_otu_bray_3","sorted_otu_phydist_3")
  data1<-data1[order(data1$sorted_otu_phydist_3),]
  fac1<-cut(data1$sorted_otu_phydist_3,nclass)
  data2<-cbind.data.frame(fac1,data1)
  library(plyr)
  data3<-ddply(data2,.(fac1),colwise(median))
  plot(data3$sorted_otu_phydist_3,data3$sorted_otu_bray_3)
  physin <- data.frame(phydist=data3$sorted_otu_phydist_3,habpre=data3$sorted_otu_bray_3)
  physin2 <- list(physin,otu_bray,otu_phydist)
  names(physin2) <- c("physin","otu_bray","otu_phydist")
  return(physin2)
}

步骤二：mantel correlogram

physin <- phylosignal(otu_niche,otu_tree,600)
physin$otu_bray[upper.tri(physin$otu_bray, diag=TRUE)] = NA
physin$otu_phydist[upper.tri(physin$otu_phydist, diag=TRUE)] = NA
man2 <- mantel.correlog(physin$otu_bray,physin$otu_phydist)
plot(1:6,man2$mantel.res[,5],lty = 1,type = "b")
abline(h = 0.05,lty = 5)

三、R代码并行实现beta-NTI计算

#本函数依赖picant、ape、doParrallel、foreach四个包
#threads：使用的线程数，可用detectCores()函数查看可使用的CPU数。
#使用的CPU数并不是越多越好，还要顾及自己的内存是否足够，因为每开一个线程都会占据相同的内存。
#2021年7月6日
beta_nti <- function(otu_niche,otu_tree,reps,threads){
  library(picante)
  library(ape)
  
  prune_tree<-prune.sample(otu_niche,otu_tree)
  tip<-prune_tree$tip.label
  coln<-colnames(otu_niche)
  m<-NULL
  for(i in 1:length(coln)){
    if(!coln[i]%in%tip){
      #print(coln[i])
      m<-cbind(m,coln[i])
    }
  }
  m<-as.vector(m)
  otu_niche<-otu_niche[,-which(colnames(otu_niche)%in%m)]
  otu_phydist <- cophenetic(prune_tree)
  match.phylo.otu = match.phylo.comm(prune_tree, otu_niche)
  #str(match.phylo.otu)
  ## calculate empirical betaMNTD
  #（算betamntd的）
  beta.mntd.weighted = as.matrix(comdistnt(match.phylo.otu$comm,cophenetic(match.phylo.otu$phy),abundance.weighted=T));
  
  #identical(rownames(match.phylo.otu$comm),colnames(beta.mntd.weighted)); # just a check, should be TRUE
  #identical(rownames(match.phylo.otu$comm),rownames(beta.mntd.weighted)); # just a check, should be TRUE
  
  # calculate randomized betaMNTD
  
  beta.reps = reps; # number of randomizations
  
  rand.weighted.bMNTD.comp = NULL
  dim(rand.weighted.bMNTD.comp);
  library(abind)
  arraybind <- function(...){
    abind(...,along = 3,force.array=TRUE)
  }

  library(foreach)
  library(doParallel)
  registerDoParallel(cores = threads)
  rand.weighted.bMNTD.comp <- foreach (rep = 1:beta.reps, .combine = "arraybind") %dopar%{
    #print(c(date(),rep))
    as.matrix(comdistnt(match.phylo.otu$comm,taxaShuffle(cophenetic(match.phylo.otu$phy)),abundance.weighted=T,exclude.conspecifics = F))
  }
  weighted.bNTI = matrix(c(NA),nrow=nrow(match.phylo.otu$comm),ncol=nrow(match.phylo.otu$comm));
  dim(weighted.bNTI);
  
  for (columns in 1:(nrow(match.phylo.otu$comm)-1)) {
    for (rows in (columns+1):nrow(match.phylo.otu$comm)) {
      
      rand.vals = rand.weighted.bMNTD.comp[rows,columns,];
      weighted.bNTI[rows,columns] = (beta.mntd.weighted[rows,columns] - mean(rand.vals)) / sd(rand.vals);
      rm("rand.vals");
      
    };
  };
  
  rownames(weighted.bNTI) = rownames(match.phylo.otu$comm);
  colnames(weighted.bNTI) = rownames(match.phylo.otu$comm);
  return(weighted.bNTI)
  }

参考文献

[1] 原核微生物群落随机性和确定性装配过程的计算方法
[2] Stochastic and deterministic assembly processes in subsurface microbial communities

测试文件请参考付费资源，两者内容相同
面包多：https://mianbaoduo.com/o/bread/YpmTlZhx
CSDN：https://download.csdn.net/download/weixin_43367441/22468236
更多R语言分析微生物生态学的资源可参考如下链接：
https://mbd.pub/o/bread/mbd-YpmTlZpr