α多样性
Alpha多样性(α多样性)是指在特定生境或生态系统内部的物种多样性。它通常包括两个方面:物种丰富度(种类的数量)和物种均匀度(物种分布的均匀性)。α多样性的测量可以帮助我们了解一个生态系统的复杂性和生物群落的健康状况。
以下是一些常用的α多样性指数及其作用和异同:
Shannon指数:
- 综合考虑了物种的丰富度和均匀度。
- 值越高,表明群落的多样性越高。
- 计算公式为 \(H = -\sum (p_i \ln p_i)\),其中\(( p_i )\)是第\(i\)个物种的相对丰度。
Richness:
- 简单地衡量生态群落中物种的数量。
- 不考虑物种的丰度或其分布。
Pielou均匀度指数:
- 衡量物种分布均匀度的指标。
- 计算公式为 \(E = H / \ln(S)\),其中\(H\)是Shannon指数,\(S\)是物种数。
Simpson指数:
衡量物种丰富度,但考虑物种的丰度权重。
值越大,多样性越低(因为它实际上计算的是均匀度的倒数)。
计算公式为 \(D=∑(n_i/N)^2\)
,其中( n_i )是第i个物种的个体数,N是群落中个体的总数。
Chao1指数:
- 估计群落中未被观察到的物种丰富度。
- 特别适用于样本中有很多稀有物种的情况。
这些指数的主要区别在于它们考虑的多样性方面不同。例如,Shannon和Simpson指数都考虑了物种的丰富度和均匀度,但Shannon更强调均匀度,而Simpson更强调丰富度。Pielou均匀度指数专注于物种分布的均匀性,而Chao1则用于估计可能未被检测到的物种的数量。这些指数通常结合使用,以提供关于生态群落多样性的更全面的视图。
通过下列代码可以绘制α多样性相关指数图:
rm(list = ls())
## 加载必要的包
library(vegan)
library(ggplot2)
library(reshape2)
library(patchwork)
## 读入数据并进行转置
otu <- read.table("final_resample_otu_table_sediment_taxa.csv", sep = ",", header = TRUE, row.names = 1)
otu <- otu[ , -ncol(otu)] # 移除taxa列
otu <- t(otu)
## 假设每个样本的组信息存储在数据行名中
groups <- sub("_.*", "", rownames(otu)) # 从行名中提取组信息,例如 "D151_2" -> "D151"
## 计算丰富度、Shannon 指数、Simpson 指数、Pielou 均匀度、Chao1 指数、ACE 指数、goods_coverage 指数
alpha <- function(x, base = exp(1)) {
est <- estimateR(x)
Richness <- est[1, ]
Chao1 <- est[2, ]
ACE <- est[4, ]
Shannon <- diversity(x, index = "shannon", base = base)
Simpson <- diversity(x, index = "simpson")
Pielou <- Shannon / log(Richness, base)
goods_coverage <- 1 - rowSums(x == 1) / rowSums(x)
result <- data.frame(SITE = rownames(x), GROUP = groups, Richness, Shannon, Simpson, Pielou, Chao1, ACE, goods_coverage)
colnames(result) <- c("SITE", "GROUP", "RICHNESS", "SHANNON", "SIMPSON", "PIELOU", "CHAO1", "ACE", "GOODS_COVERAGE")
result
}
alpha_all <- alpha(otu, base = 2)
## 保存为 CSV 文件
write.csv(alpha_all, file = "alpha_all.csv", row.names = FALSE, quote = FALSE)
## 绘制各指数箱线图
indices <- c("SHANNON", "RICHNESS", "PIELOU", "SIMPSON", "CHAO1")
# 创建一个空列表来存储图形对象
plots <- list()
# 使用循环创建每个指数的箱线图
for (i in seq_along(indices)) {
index <- indices[i]
data_melted <- melt(alpha_all, id.vars = c("GROUP", "SITE"), measure.vars = index)
colnames(data_melted)[which(colnames(data_melted) == "value")] <- "value"
p <- ggplot(data_melted, aes(x = GROUP, y = value, fill = GROUP)) +
geom_boxplot() +
labs(x = paste(index, "Index"), y = "") +
theme_bw() +
theme(axis.text.x = element_blank())
if (i == length(indices)) {
p <- p + labs(fill = "Group") # 仅在最后一个图表上添加图例
} else {
p <- p + guides(fill = FALSE) # 隐藏其余图表的图例
}
plots[[index]] <- p
}
# 使用patchwork将所有图表组合成一个1行5列的布局
p_combined <- wrap_plots(plots, ncol = 5)
# 保存图像
ggsave(paste0('alpha_all_', "Alpha_Diversity", '.svg'), p_combined, width = 6 * 0.6 * 5, height = 6)
其中,alpha_all.csv
的格式如:
SITE GROUP RICHNESS SHANNON SIMPSON PIELOU CHAO1 ACE GOODS_COVERAGE
D151_2 Raw 456 5.985928225 0.966626149 0.677686263 908.7631579 921.6164037 0.994927596
结果图:
β多样性
基于OTU表
library(vegan)
library(ape)
library(ggplot2)
library(grid)
library(ggalt)
library(dplyr)
library(multcomp)
library(patchwork)
library(xlsx)
rm(list = ls())
galaxy <- "33"
method <-"jaccard" #https://rdocumentation.org/packages/vegan/versions/2.6-4/topics/vegdist #bray jaccard
groups <- read.xlsx2("Group3233.xlsx", sheetName =galaxy, row.names=1)
Legend_column <- "Group" #分组依据是csv中的哪一列
output_url <- paste0("Galaxy",galaxy,"-PCoA-",method)
data = read.table(paste0("Galaxy",galaxy,".csv"), header=T, row.names=1, sep=",")
data=t(data)#每行为一个样品
#在otu和gp中删除Legend_column为空的行
# empty_rows <- which( (is.na(groups[[paste0(Legend_column)]])) | (groups[[paste0(Legend_column)]]=="") )
# groups <- groups[-empty_rows,]
# data <- data[-empty_rows,]
# rm(empty_rows)
# groups=subset(groups, select = c(paste0(Legend_column),"Mark"))#只保留所需列
#只保留某些mark,此段代码选择启用
# rows <- c("G52","C52","Raw")
# remain_rows <- which( groups[["Mark"]] %in% rows )
# groups <- groups[remain_rows,]
# data <- data[remain_rows,]
# rm(remain_rows)
length=nrow(groups)
times1=length
res1=length
times2=length
res2=length
col1=rep(1:8,times1)
col=c(col1,1:res1)
pich1=rep(c(21:24),times2)
pich=c(pich1,15:(15+res2))
data <- vegdist(data, method = method)
pcoa<- pcoa(data, correction = "none", rn = NULL)
PC1 = pcoa$vectors[,1]
PC2 = pcoa$vectors[,2]
plotdata <- data.frame(rownames(pcoa$vectors),PC1,PC2,groups)
colnames(plotdata)[1:3] <-c("sample","PC1","PC2") #重命名1~3列
names(plotdata)[names(plotdata) == paste0(Legend_column)] <- 'Group' #重命Group列
pc1 <-floor(pcoa$values$Relative_eig[1]*100)
pc2 <-floor(pcoa$values$Relative_eig[2]*100)
plotdata$Group <- factor(plotdata$Group)
yf <- plotdata
yd1 <- yf %>% group_by(Group) %>% summarise(Max = max(PC1))
yd2 <- yf %>% group_by(Group) %>% summarise(Max = max(PC2))
yd1$Max <- yd1$Max + max(yd1$Max)*0.1
yd2$Max <- yd2$Max + max(yd2$Max)*0.1
fit1 <- aov(PC1~Group,data = plotdata)
tuk1<-glht(fit1,linfct=mcp(Group="Tukey"))
res1 <- cld(tuk1,alpah=0.05)
fit2 <- aov(PC2~Group,data = plotdata)
tuk2<-glht(fit2,linfct=mcp(Group="Tukey"))
res2 <- cld(tuk2,alpah=0.05)
test <- data.frame(PC1 = res1$mcletters$Letters,PC2 = res2$mcletters$Letters,
yd1 = yd1$Max,yd2 = yd2$Max,Group = yd1$Group)
test$Group <- factor(test$Group)
p1 <- ggplot(plotdata,aes(Group,PC1)) +
geom_boxplot(aes(fill = Group)) +
geom_text(data = test,aes(x = Group,y = yd1,label = PC1),size = 7) +
coord_flip() +
theme_bw()+
theme(axis.ticks.length = unit(.4,"lines"),
axis.title.x=element_blank(),
axis.title.y=element_blank(),
axis.text.y=element_text(size=15),
axis.text.x=element_blank(),
legend.position = "none")
yrange <- ggplot_build(p1)$layout$panel_scales_y[[1]]$range$range
p1
p3 <- ggplot(plotdata,aes(Group,PC2)) +
geom_boxplot(aes(fill = Group)) +
geom_text(data = test,aes(x = Group,y = yd2,label = PC2),
size = 7,color = "black",fontface = "bold") +
theme_bw()+
theme(axis.ticks.length = unit(0.4,"lines"),
axis.ticks = element_line(color='black'),
axis.line = element_line(colour = "black"),
axis.title.x=element_blank(),
axis.title.y=element_blank(),
axis.text.x=element_text(colour='black',size=15,angle = 45,
vjust = 1,hjust = 1),
axis.text.y=element_blank(),
legend.position = "none")
yrange <- ggplot_build(p3)$layout$panel_scales_y[[1]]$range$range
p3 <- p3
p3
p2<-ggplot(plotdata, aes(PC1, PC2)) +
geom_encircle(aes(fill=Group,colour=Group),alpha=.1, show.legend=F,size=2,expand=0.05)+ #同组阴影底色
#stat_ellipse(level=.6,aes(colour=Group))+ #画置信圈,alpha是透明度
geom_point(aes(fill=Group,colour=Group),size=5,pch =21, stroke=2, alpha=.8)+ #点
#geom_text(aes(label = Mark), size = 3)+
xlab(paste0("PC1 ( ",pc1,"%"," )")) +
ylab(paste0("PC2 ( ",pc2,"%"," )"))+
xlim(ggplot_build(p1)$layout$panel_scales_y[[1]]$range$range) +
ylim(ggplot_build(p3)$layout$panel_scales_y[[1]]$range$range) +
theme(text=element_text(size=18))+
geom_vline(aes(xintercept = 0),linetype="dotted")+
geom_hline(aes(yintercept = 0),linetype="dotted")+
theme_classic()+
theme(axis.ticks.length = unit(0.4,"lines"), axis.ticks = element_line(color='black'),
axis.line = element_line(colour = "black"),
axis.title.x=element_text(colour='black', size=18,vjust=3),
axis.title.y=element_text(colour='black', size=18,vjust=-1),
axis.text=element_text(colour='black',size=14),
legend.title=element_text(size = 14),
legend.text=element_text(size=12),
legend.key=element_blank(),legend.position = c(1, 1),legend.justification = c(1,1),
legend.background = element_rect(colour = "black",fill=alpha("white",0.4)),
)
p2
otu.adonis=adonis2(data~groups[[paste0(Legend_column)]],data = groups,distance = method)
p4 <- ggplot(plotdata, aes(PC1, PC2)) +
geom_text(aes(x = -0.5,y = 0.6,label = paste(output_url,"\ndf = ",otu.adonis$Df[1], "\nR² = ",round(otu.adonis$R2[1],4), "\nP-value = ",otu.adonis$`Pr(>F)`[1],sep = "")),
size = 5,family="serif") +
theme_bw() +
xlab("") + ylab("") +
theme(panel.grid=element_blank(),
axis.title = element_blank(),
axis.line = element_blank(),
axis.ticks = element_blank(),
axis.text = element_blank())
p4
p5 <- p1 + p4 + p2 + p3 +
plot_layout(heights = c(1,4),widths = c(4,1),ncol = 2,nrow = 2)
p5
# info <- paste0("【数据】选用",unlist(strsplit(data_url, "-"))[1],"压强为",paste(unique(groups$Pressure),collapse ="、"),"第",paste(unique(groups$Generation),collapse ="、"),"代,营养浓度",paste(unique(groups$Nutrition),collapse ="、"),"的数据样本。\n","【结论】",Legend_column,"对于本样本有一定影响。","\n【说明】P值<0.05时说明不同组间差异显著。R²越大说明分组对差异的解释度越高。\n\tGalaxy32使用UPARSE方法,Galaxy33使用DADA2方法。")
# library(showtext)
# library(Cairo)
# library(stringr)
# font_add("STSong","STSONG.TTF")
# showtext_auto(enable=T)
# p2 <- p2+labs(caption = info)+
# theme(plot.caption = element_text(size=15, family="STSong",hjust=0,vjust=3,lineheight=1.25))
# p2
# p5 <- p1 + p4 + p2 + p3 +
# plot_layout(heights = c(1,4),widths = c(4,1),ncol = 2,nrow = 2)
# p5
ggsave(paste0(output_url,'.pdf'), p5, width = 12, height =12/sqrt(2))
print ("end")
library(vegan)
library(ape)
library(ggplot2)
library(grid)
library(ggalt)
library(dplyr)
library(multcomp)
library(patchwork)
library(xlsx)
rm(list = ls())
galaxy <- "33"
method <-"jaccard" #https://rdocumentation.org/packages/vegan/versions/2.6-4/topics/vegdist #bray jaccard
groups <- read.xlsx2("Group3233.xlsx", sheetName =galaxy, row.names=1)
Legend_column <- "Group" #分组依据是csv中的哪一列
output_url <- paste0("Galaxy",galaxy,"-NMDS-",method)
data = read.table(paste0("Galaxy",galaxy,".csv"), header=T, row.names=1, sep=",")
data=t(data)#每行为一个样品
#在otu和gp中删除Legend_column为空的行
# empty_rows <- which( (is.na(groups[[paste0(Legend_column)]])) | (groups[[paste0(Legend_column)]]=="") )
# groups <- groups[-empty_rows,]
# data <- data[-empty_rows,]
# rm(empty_rows)
# groups=subset(groups, select = c(paste0(Legend_column),"Mark"))#只保留所需列
#只保留某些mark,此段代码选择启用
# rows <- c("G52","C52","Raw")
# remain_rows <- which( groups[["Mark"]] %in% rows )
# groups <- groups[remain_rows,]
# data <- data[remain_rows,]
# rm(remain_rows)
data <- vegdist(data, method = method)
nmds <- metaMDS(data, distance = method, k = 2)
NMDS1 = nmds$points[,1]
NMDS2 = nmds$points[,2]
# 创建包含NMDS坐标和分组信息的数据框架
plotdata <- data.frame(sample=rownames(nmds$points), NMDS1, NMDS2, groups)
colnames(plotdata)[1:3] <- c("sample","NMDS1","NMDS2") # 重命名1~3列
names(plotdata)[names(plotdata) == paste0(Legend_column)] <- 'Group' # 重命Group列
plotdata$Group <- factor(plotdata$Group)
p1 <- ggplot(plotdata, aes(Group, NMDS1)) +
geom_boxplot(aes(fill = Group)) +
coord_flip() +
theme_bw() +
theme(axis.ticks.length = unit(.4,"lines"),
axis.title.x=element_blank(),
axis.title.y=element_blank(),
axis.text.y=element_text(size=15),
axis.text.x=element_blank(),
legend.position = "none")
p1
p3 <- ggplot(plotdata, aes(Group, NMDS2)) +
geom_boxplot(aes(fill = Group)) +
theme_bw() +
theme(axis.ticks.length = unit(0.4,"lines"),
axis.ticks = element_line(color='black'),
axis.line = element_line(colour = "black"),
axis.title.x=element_blank(),
axis.title.y=element_blank(),
axis.text.x=element_text(colour='black',size=15,angle = 45,
vjust = 1,hjust = 1),
axis.text.y=element_blank(),
legend.position = "none")
p3
p2 <- ggplot(plotdata, aes(NMDS1, NMDS2)) +
geom_point(aes(fill=Group,color=Group), size=5, pch = 21, stroke=2, alpha=.8) +
#geom_encircle(aes(fill=sample_data(ps)$Group,colour=sample_data(ps)$Group),alpha=.1, show.legend=F,size=2,expand=0.05)+ #同组阴影底色
stat_ellipse(level=.8,aes(colour=Group))+ #画置信圈,alpha是透明度
xlab("NMDS 1") +
ylab("NMDS 2") +
theme_classic() +
theme(axis.ticks.length = unit(0.4,"lines"), axis.ticks = element_line(color='black'),
axis.line = element_line(colour = "black"),
axis.title.x=element_text(colour='black', size=18,vjust=3),
axis.title.y=element_text(colour='black', size=18,vjust=-1),
axis.text=element_text(colour='black',size=14),
legend.title=element_text(size = 14),
legend.text=element_text(size=12),
legend.key=element_blank(),legend.position = c(1, 1),legend.justification = c(1,1),
legend.background = element_rect(colour = "black",fill=alpha("white",0.4)),
)
p2
otu.adonis <- adonis2(data ~ groups[[paste0(Legend_column)]], data = groups, distance = method)
p4 <- ggplot(plotdata, aes(PC1, PC2)) +
geom_text(aes(x = -0.5,y = 0.6,label = paste(output_url,"\ndf = ",otu.adonis$Df[1], "\nR² = ",round(otu.adonis$R2[1],4), "\nP-value = ",otu.adonis$`Pr(>F)`[1],sep = "")),
size = 5,family="serif") +
theme_bw() +
xlab("") + ylab("") +
theme(panel.grid=element_blank(),
axis.title = element_blank(),
axis.line = element_blank(),
axis.ticks = element_blank(),
axis.text = element_blank())
p4
p5 <- p1 + p4 + p2 + p3 +
plot_layout(heights = c(1,4),widths = c(4,1),ncol = 2,nrow = 2)
p5
# info <- paste0("【数据】选用",unlist(strsplit(data_url, "-"))[1],"压强为",paste(unique(groups$Pressure),collapse ="、"),"第",paste(unique(groups$Generation),collapse ="、"),"代,营养浓度",paste(unique(groups$Nutrition),collapse ="、"),"的数据样本。\n","【结论】",Legend_column,"对于本样本有一定影响。","\n【说明】P值<0.05时说明不同组间差异显著。R²越大说明分组对差异的解释度越高。\n\tGalaxy32使用UPARSE方法,Galaxy33使用DADA2方法。")
# library(showtext)
# library(Cairo)
# library(stringr)
# font_add("STSong","STSONG.TTF")
# showtext_auto(enable=T)
# p2 <- p2+labs(caption = info)+
# theme(plot.caption = element_text(size=15, family="STSong",hjust=0,vjust=3,lineheight=1.25))
# p2
# p5 <- p1 + p4 + p2 + p3 +
# plot_layout(heights = c(1,4),widths = c(4,1),ncol = 2,nrow = 2)
# p5
ggsave(paste0(output_url,'.pdf'), p5, width = 12, height =12/sqrt(2))
print ("end")
基于系统发生树
rm(list = ls())
library(phyloseq)
library(vegan)
library(ggalt)
library(ggplot2)
library(ape)
library(xlsx)
library(patchwork)
tree <- read.tree("G33-Galaxy51-[FastTree.nwk].nwk")
# 读取分组文件
Group <- read.xlsx2("Group3233.xlsx", sheetName ="33", row.names=1)
# 读取物种丰度表,假设文件名为otu_table.csv
otu_url <- "Galaxy33-[DADA2_OTU_table].csv"
otu <- read.csv(otu_url, row.names = 1)
method <- "wunifrac" #wunifrac unifrac
output_url<-paste0(unlist(strsplit(otu_url, "-"))[1],"-NMDS-",method)
# 构建phyloseq对象
ps <- phyloseq(otu_table(as.matrix(otu), taxa_are_rows = TRUE),
sample_data(Group),
phy_tree(tree))
# 从phyloseq对象中提取元数据
meta <- as.data.frame(sample_data(ps))
# 计算加权UniFrac距离矩阵
dist_matrix <- as.matrix(distance(ps, method = method))
# 计算NMDS
nmds <- metaMDS(dist_matrix, distance = method, k = 2)
# 将NMDS结果转换为数据框架
nmds_df <- as.data.frame(nmds$points)
meta <- as.data.frame(sample_data(ps))
otu.adonis <- adonis2(dist_matrix ~ Group, data=Group)
NMDS1 = nmds$points[,1]
NMDS2 = nmds$points[,2]
plotdata <- data.frame(sample=rownames(nmds$points), NMDS1, NMDS2, Group=meta$Group)
plotdata$Group <- factor(plotdata$Group) # 确保分组变量是因子类型
p1 <- ggplot(plotdata,aes(Group,NMDS1)) +
geom_boxplot(aes(fill = Group)) +
coord_flip() +
theme_bw()+
theme(axis.ticks.length = unit(.4,"lines"),
axis.title.x=element_blank(),
axis.title.y=element_blank(),
axis.text.y=element_text(size=15),
axis.text.x=element_blank(),
legend.position = "none")
yrange <- ggplot_build(p1)$layout$panel_scales_y[[1]]$range$range
p1
p3 <- ggplot(plotdata,aes(Group,NMDS2)) +
geom_boxplot(aes(fill = Group)) +
theme_bw()+
theme(axis.ticks.length = unit(0.4,"lines"),
axis.ticks = element_line(color='black'),
axis.line = element_line(colour = "black"),
axis.title.x=element_blank(),
axis.title.y=element_blank(),
axis.text.x=element_text(colour='black',size=15,angle = 45,
vjust = 1,hjust = 1),
axis.text.y=element_blank(),
legend.position = "none")
yrange <- ggplot_build(p3)$layout$panel_scales_y[[1]]$range$range
p3
# 使用ggplot2包进行绘图
p2<-ggplot(nmds_df, aes(MDS1, MDS2, color = meta$Group)) +
labs(color = "Group",fill="Group")+
#geom_encircle(aes(fill=sample_data(ps)$Group,colour=sample_data(ps)$Group),alpha=.1, show.legend=F,size=2,expand=0.05)+ #同组阴影底色
stat_ellipse(level=.8,aes(colour=sample_data(ps)$Group))+ #画置信圈,alpha是透明度
geom_point(aes(fill=sample_data(ps)$Group,colour=sample_data(ps)$Group),size=5,pch =21, stroke=2, alpha=.8)+ #点
#geom_text(aes(label = Mark), size = 3)+
xlab("NMDS 1") +
ylab("NMDS 2")+
xlim(ggplot_build(p1)$layout$panel_scales_y[[1]]$range$range) +
ylim(ggplot_build(p3)$layout$panel_scales_y[[1]]$range$range) +
theme(text=element_text(size=18))+
geom_vline(aes(xintercept = 0),linetype="dotted")+
geom_hline(aes(yintercept = 0),linetype="dotted")+
theme_classic()+
theme(axis.ticks.length = unit(0.4,"lines"), axis.ticks = element_line(color='black'),
axis.line = element_line(colour = "black"),
axis.title.x=element_text(colour='black', size=18,vjust=3),
axis.title.y=element_text(colour='black', size=18,vjust=-1),
axis.text=element_text(colour='black',size=14),
legend.title=element_text(size = 14),
legend.text=element_text(size=12),
legend.key=element_blank(),legend.position = c(1, 1),legend.justification = c(1,1),
legend.background = element_rect(colour = "black",fill=alpha("white",0.4)),
)
p2
p4 <- ggplot(plotdata, aes(PC1, PC2)) +
geom_text(aes(x = -0.5,y = 0.6,label = paste0(output_url,"\ndf = ",otu.adonis$Df[1], "\nR² = ",round(otu.adonis$R2[1],4), "\nP-value = ",otu.adonis$`Pr(>F)`[1],sep = "")),
size = 4.5,family="serif") +
theme_bw() +
xlab("") + ylab("") +
theme(panel.grid=element_blank(),
axis.title = element_blank(),
axis.line = element_blank(),
axis.ticks = element_blank(),
axis.text = element_blank())
p4
p5 <- p1 + p4 + p2 + p3 +
plot_layout(heights = c(1,4),widths = c(4,1),ncol = 2,nrow = 2)
p5
ggsave(paste0("output/",output_url,".pdf"), p5, width = 12, height =12/sqrt(2))
print ("end")
rm(list = ls())
library(phyloseq)
library(vegan)
library(ggalt)
library(ggplot2)
library(ape)
library(xlsx)
library(patchwork)
tree <- read.tree("G33-Galaxy51-[FastTree.nwk].nwk")
# 读取分组文件
Group <- read.xlsx2("Group3233.xlsx", sheetName ="33", row.names=1)
# 读取物种丰度表,假设文件名为otu_table.csv
otu_url <- "Galaxy33-[DADA2_OTU_table].csv"
otu <- read.csv(otu_url, row.names = 1)
method <- "unifrac"
output_url<-paste0(unlist(strsplit(otu_url, "-"))[1],"-PCoA-",method)
# 构建phyloseq对象
ps <- phyloseq(otu_table(as.matrix(otu), taxa_are_rows = TRUE),
sample_data(Group),
phy_tree(tree))
# 计算PCoA
pcoa <- ordinate(ps, method = "PCoA", distance = method)
PC1 = pcoa$vectors[,1]
PC2 = pcoa$vectors[,2]
plotdata <- data.frame(rownames(pcoa$vectors),PC1,PC2,Group)
colnames(plotdata)[1:4] <-c("sample","PC1","PC2","Group") #重命名1~3列
pc1 <-floor(pcoa$values$Relative_eig[1]*100)
pc2 <-floor(pcoa$values$Relative_eig[2]*100)
plotdata$Group <- factor(plotdata$Group)
dist_matrix <- as.matrix(distance(ps, method = method))
meta <- as.data.frame(sample_data(ps))
otu.adonis <- adonis2(dist_matrix ~ Group, data=Group)
p1 <- ggplot(plotdata,aes(Group,PC1)) +
geom_boxplot(aes(fill = Group)) +
coord_flip() +
theme_bw()+
theme(axis.ticks.length = unit(.4,"lines"),
axis.title.x=element_blank(),
axis.title.y=element_blank(),
axis.text.y=element_text(size=15),
axis.text.x=element_blank(),
legend.position = "none")
yrange <- ggplot_build(p1)$layout$panel_scales_y[[1]]$range$range
p1
p3 <- ggplot(plotdata,aes(Group,PC2)) +
geom_boxplot(aes(fill = Group)) +
theme_bw()+
theme(axis.ticks.length = unit(0.4,"lines"),
axis.ticks = element_line(color='black'),
axis.line = element_line(colour = "black"),
axis.title.x=element_blank(),
axis.title.y=element_blank(),
axis.text.x=element_text(colour='black',size=15,angle = 45,
vjust = 1,hjust = 1),
axis.text.y=element_blank(),
legend.position = "none")
yrange <- ggplot_build(p3)$layout$panel_scales_y[[1]]$range$range
p3
# 使用ggplot2包进行绘图
p2<-plot_ordination(ps, pcoa) +
labs(color = "Group",fill="Group")+
#geom_encircle(aes(fill=sample_data(ps)$Group,colour=sample_data(ps)$Group),alpha=.1, show.legend=F,size=2,expand=0.05)+ #同组阴影底色
stat_ellipse(level=.6,aes(colour=sample_data(ps)$Group))+ #画置信圈,alpha是透明度
geom_point(aes(fill=sample_data(ps)$Group,colour=sample_data(ps)$Group),size=5,pch =21, stroke=2, alpha=.8)+ #点
#geom_text(aes(label = Mark), size = 3)+
xlab(paste0("PC1 ( ",pc1,"%"," )")) +
ylab(paste0("PC2 ( ",pc2,"%"," )"))+
xlim(ggplot_build(p1)$layout$panel_scales_y[[1]]$range$range) +
ylim(ggplot_build(p3)$layout$panel_scales_y[[1]]$range$range) +
theme(text=element_text(size=18))+
geom_vline(aes(xintercept = 0),linetype="dotted")+
geom_hline(aes(yintercept = 0),linetype="dotted")+
theme_classic()+
theme(axis.ticks.length = unit(0.4,"lines"), axis.ticks = element_line(color='black'),
axis.line = element_line(colour = "black"),
axis.title.x=element_text(colour='black', size=18,vjust=3),
axis.title.y=element_text(colour='black', size=18,vjust=-1),
axis.text=element_text(colour='black',size=14),
legend.title=element_text(size = 14),
legend.text=element_text(size=12),
legend.key=element_blank(),legend.position = c(1, 1),legend.justification = c(1,1),
legend.background = element_rect(colour = "black",fill=alpha("white",0.4)),
)
p2
dist_matrix <- as.matrix(distance(ps, method = method))
# 从phyloseq对象中提取元数据
meta <- as.data.frame(sample_data(ps))
# 进行ADONIS分析
otu.adonis <- adonis2(dist_matrix ~ Group, data=Group)
p4 <- ggplot(plotdata, aes(PC1, PC2)) +
geom_text(aes(x = -0.5,y = 0.6,label = paste(output_url,"\ndf = ",otu.adonis$Df[1], "\nR² = ",round(otu.adonis$R2[1],4), "\nP-value = ",otu.adonis$`Pr(>F)`[1],sep = "")),
size = 4.5,family="serif") +
theme_bw() +
xlab("") + ylab("") +
theme(panel.grid=element_blank(),
axis.title = element_blank(),
axis.line = element_blank(),
axis.ticks = element_blank(),
axis.text = element_blank())
p4
p5 <- p1 + p4 + p2 + p3 +
plot_layout(heights = c(1,4),widths = c(4,1),ncol = 2,nrow = 2)
p5
ggsave(paste0("output\",output_url,'.pdf'), p5, width = 12, height =12/sqrt(2))
print ("end")
发表您的看法