Reproducibilty is an essential aspect of data analysis. Especially for sequencing data for microbial community profilling. There are lack of standards to analysing and reporting data. This is mostly because there are simply some aspects of the data that we do not control. In such cases it is important to have rational decision making and documenting it. In almost all microbiome profilling data choices regarding filtering parameters (OTU count, OTU prevalences), normalisation, transformation are routine practice. Pre-processing is an essential step and it is crucial that it is documented. There are some common approaches to look at the data.
1. How many reads/samples? if there is large difference this may affect the choices you make for alpha diversity, betadiversity and differential abundance testing.
2. How many reads/OTU ? this is aslo important to see if there are lot of OTUs with small mean & large coefficient of variation (C.V.).
This tempate for microbiome data analysis, handling and visualisation. This makes it easy to follow some standard procedures for looking at your data and reporting your results. This first-hand look at your data will aid in making different choices for more thorough and in-dept investigation.
Check this link on Organize your data and code.
You can find the source code at microbiomeutilities github repo.
A step wise guide can be found at microbiomeutilities wiki.
This step will check for installed packages and dependencies.
# List of packages for session .packages <- c("data.table", "plyr", "dplyr", "tidyr", "ggplot2", "viridis", "phyloseq", "ggrepel", "Matrix", "RColorBrewer", "devtools", "picante") # Install CRAN packages .inst <- .packages %in% installed.packages() if(any(!.inst)) { install.packages(.packages[!.inst], repos = "http://cran.rstudio.com/") } if (!("microbiome" %in% installed.packages())) { devtools::install_github("microbiome/microbiome") # Install the package } # install.packages("picante",repos="http://R-Forge.R-project.org") allneededpackages <- c(.packages, "microbiome") # Load packages into session sapply(allneededpackages, require, character.only = TRUE)
We have to load libraries click on the code tab on the left hand die of this doc for looking at the libraries that are installed. It is good to install these packages before.
library(microbiome) library(microbiomeutilities) # install.packages("picante",repos="http://R-Forge.R-project.org") # library(picante) library(picante) library(data.table) library(DT) library(RColorBrewer)
Fill the details accordingly.
Project title : Create a standard workflow for data reporting.
Main author : Yourname
contributor(s) : XXXX
Principal Investigator (s) : XXXX
Co-supervisor(s) : XXXXX
The next code chunk is where you put all the parameters for analysis. This is the basis for your prelimanry analysis.
#generate a random number n <- runif(1, 0, 10^6) # keep this number stored as below instead of XXX when re running the codes. message("random number set.seed is below") print(n) set.seed(n)
seed number(XXX) check this link for more information
otufile = "myBiom.biom" # input biom/shared/otutable as *csv file. mapping = "myMapping.csv" # input mapping file in *.csv format. taxonomy = NULL # taxonomy file in *.csv format. treefilename = "myTree.tre" # tree file as *.tre and not *.tree. type = "biom" # type of input either biom, mothur .. # .. or csv file as *.tre and not *.tree. out_dir = "F:/PATH/to/my/output/directory" # path to the output directory. VariableA = "MY_MainVariable" # main variable of interest. VariableB = "MY_SecondaryVariable" # secondary variable of interest. UnConstOrd = TRUE # Unconstrained ORDINATION TRUE or FALSE. heatmap = TRUE # TRUE or FALSE filterCount = 4 # Filter OTUs below this count number. filterPrev = 0.01 # filter parameter for OTUs, prevalence in dataset 0.01 = 1%. col.palette = "Paired" # choice of color (RColorbrewer) filterpseq = TRUE # filter phyloseq will be saved samsize=NA # rarefying depth
Check the code chunk next to this of left side to see folder names where files are stored.
oripath <- getwd() message("Current working directory is ", oripath) setwd(out_dir) message("Otuput directory is set to ", out_dir) if(file.exists("QC")) { message("QC folder already exists, data will be overwritten") } else{ message("QC folder will be created in ", out_dir) dir.create("QC") } if(file.exists("AlphaDiversity")) { message("AlphaDiversity folder already exists, data will be overwritten") } else{ message("AlphaDiversity folder will be created in ", out_dir) dir.create("AlphaDiversity") } if(file.exists("BetaDiversity")) { message("BetaDiversity folder already exists, data will be overwritten") } else{ message("BetaDiversity folder will be created in ", out_dir) dir.create("BetaDiversity") } if(file.exists("Others")) { message("Others folder already exists, data will be overwritten") } else{ message("Others folder will be created in ", out_dir) dir.create("Others") } if(file.exists("PhyloseqObjects")) { message("PhyloseqObjects folder already exists, data will be overwritten") } else{ message("PhyloseqObjects folder will be created in ", out_dir) dir.create("PhyloseqObjects") }
This code chunk reads the input files and creates a phyloseq object.
ps0 <- read_phyloseq(otu.file = otufile, metadata.file = mapping, taxonomy.file = taxonomy, type = type) tree <- read.tree(treefilename) ps0 <- merge_phyloseq(ps0, tree) saveRDS(ps0, "./PhyloseqObjects/ps_raw.rds") message("Raw phyloseq object, confirm the number of samples and variables (as in columns of mapping file)") message("Below is the content of raw phyloseqobject stored as ps_raw.rds") print(ps0)
print("Below is the summary of the phyloseq object") summarize_phyloseq(ps0)
Check the library sizes for each of the samples within the main variable. Do you think it is okay or there is difference in library sizes and will this affect you downstream statistical analysis?
qc_plot1 <- plot_ReadDistribution(ps0, groups = VariableB, plot.type = "histogram") ggsave("./QC/ReadDistribution.pdf") message("QC plots for ReadDistribution stored in QC folder as ReadDistribution.pdf")
message("Investigating library sizes") barplot(sort(sample_sums(ps0)), horiz = TRUE, las = 2, sub = "Check if there is large differeence in library sizes")
message("Investigating library sizes") hist(sort(sample_sums(ps0)), las = 2, sub = "Check if there is large differeence in library sizes") message("QC plots for library sizes stored in QC folder as LibrarySizePerSample.pdf") pdf("./QC/LibrarySizePerSample.pdf") barplot(sort(sample_sums(ps0)), horiz = TRUE, las = 2, sub = "Check if there is large differeence in library sizes") hist(sort(sample_sums(ps0)), las = 2) dev.off()
If any sample has less than 2000 reads, it will be removed
if(min(sample_sums(ps0)) < 2000){ print("There are sample(s) less than 2000 reads, these will be removed") ps0 = prune_samples(sample_sums(ps0)>=2000, ps0) } else { print("No samples below 2000 reads") print(ps0) }
message("Investigating OTU counts distribution") barplot(sort(taxa_sums(ps0)), horiz = TRUE, las = 2, sub = "Check if there is large differeence in otu counts")
message("Investigating OTU counts distribution") hist(taxa_sums(ps0), las = 2, main = "raw") pdf("./QC/Distribution_OTU_Counts.pdf") barplot(sort(taxa_sums(ps0)), horiz = TRUE, las = 2, sub = "Check if there is large differeence in otu counts") hist(taxa_sums(ps0), las = 2, main = "raw") dev.off() message("QC plots for library sizes stored in QC folder as Distribution_OTU_Counts.pdf")
# for sanity ps1 <- prune_taxa(taxa_sums(ps0) > 0, ps0)
This is variance for all OTU counts without filtering for min number of reads/OTU and prevalence.
Variance.plot.a <- qplot(log10(apply(otu_table(ps1), 1, var)), xlab = "log10(variance)", main = "Variance in OTUs") + ggtitle("before filtering") + theme_minimal() print(Variance.plot.a) ggsave("./QC/Variance before filtering.pdf") message("QC plots for OTU variance stored in QC folder as Variance before filtering.pdf")
This is variance for all OTU counts after filtering for min number of reads/OTU and prevalence.
if (filterpseq == TRUE) { message(paste0("Filtering OTUs with less than ", filterCount, " counts")) message(paste0("in at least ", filterPrev*100, " % of the samples ")) ps2 = filter_taxa(ps1, function(x) sum(x > filterCount) > (filterPrev * length(x)), TRUE) message("Saving the transformed phyloseq object as ps_filtered.rds") saveRDS(ps2, "./PhyloseqObjects/ps_filtered.rds") message("Below is the content of filtered phyloseqobject (based on filterCount and filterPrev) stored as ps_filtered.rds") print(ps2) Variance.plot.b <- qplot(log10(apply(otu_table(ps2), 1, var)), xlab = "log10(variance)", main = "Variance in OTUs") + ggtitle("after filtering") + theme_minimal() print(Variance.plot.b) ggsave("./QC/Variance After filtering.pdf") } else { message("filterpseq was false. Did not filter and hence will not save the filtered phyloseq") ps2 <- ps1 }
Kurtosis is essentially a measure of how much weight is at the tails of the distribution relative to the weight around the location.
If you have large differences in your library sizes then you may have to normalise your data.
message("Using the raw phyloseq to check for kurtosis in library size") df <- data.table(NumberReads = sample_sums(ps0), SampleID = sample_names(ps0)) require(moments) n <- kurtosis(df$NumberReads) if (n > 3) { message("Your library size is heavily tailed, considering normalising them for further analysis") } else { message("The variation in library sizes is below kurtosis value of 3 may indicate no need for rarefying") }
Alpha diversity measures are standard calculations in microbial ecology. The differences in richness and eveness between groups may have importance to understanding the ecology. There are numerous measures we use the defaults from microbiome
R package and also the phylogenetic diversity from picante
R package.
The caculations can be done on rarefied or non-rarefied data which can be specified by the samsize
option in "Set project attributes" option above.
For more on this check Microbiome:Diversities
if (!is.na(samsize)) { ps3 <- rarefy_even_depth(ps2, sample.size = samsize) saveRDS(ps.rare, "./phyloseqObjects/ps_rarefyied.rds") } else{ ps3 <- ps2 } metadf <- meta(ps3) alpha_div <- plot_richness(ps3, color = VariableA, shape = VariableB, measures = c("Observed", "Chao1", "Shannon", "InvSimpson")) alpha_div <- alpha_div + geom_boxplot() + ggtitle("Non phylogenetic diversity") + scale_fill_brewer(palette = col.palette) print(alpha_div) if (!is.na(samsize)){ message("Non-phylogenetic_alpha_diversity on RAREFIED data stored in AlphaDiversity folder as") message("Non-phylogenetic_alpha_diversity.pdf") } else{ message("Non-phylogenetic_alpha_diversity on NON-RAREFIED data stored in AlphaDiversity folder as") message("Non-phylogenetic_alpha_diversity.pdf") } ggsave("./AlphaDiversity/Non-phylogenetic_alpha_diversity.pdf", height = 6, width = 18)
For more on this check Picante.
if (!is.na(samsize)) { message("Rarefyied phyloseq object will be used to calculate PD") print(ps3) saveRDS(ps.rare, "./phyloseqObjects/ps_rarefyied.rds") } else{ message("Non-Rarefyied phyloseq object will be used to calcualte PD") print(ps3) } otu_table_ps3 <- as.data.frame(ps3@otu_table) metadata_table_ps3 <- meta(ps3) message("include.root in pd is set to FALSE by default") df.pd <- pd(t(otu_table_ps3), tree,include.root=F) # t(ou_table) transposes the table for use in picante and the tre file comes from the first code chunck we used to read tree file (see making a phyloseq object section). datatable(df.pd) # now we need to plot PD # check above how to get the metadata file from phyloseq object. # We will add the results of PD to this file and then plot. select.meta <- metadata_table_ps3[,c(VariableA,VariableB)] #, "Phyogenetic_diversity"] select.meta$Phyogenetic_diversity <- df.pd$PD colnames(select.meta) <- c("VariableA", "VariableB", "Phyogenetic_diversity") plot.pd <- ggplot(select.meta, aes(VariableA, Phyogenetic_diversity)) + geom_boxplot(aes(fill = VariableB)) + geom_point(size = 2) + theme(axis.text.x = element_text(size=14, angle = 90)) + theme_bw() + scale_fill_brewer(palette = col.palette) print(plot.pd) if (!is.na(samsize)) { message("Non-phylogenetic_alpha_diversity.pdf") ggsave("./AlphaDiversity/Phylogenetic_diversityon_Rafrefied_data.pdf", plot = plot.pd, height = 6, width = 18) } else{ message("Non-phylogenetic_alpha_diversity.pdf") ggsave("./AlphaDiversity/Phylogenetic_diversityon_nonRafrefied_data.pdf", plot = plot.pd, height = 6, width = 18) }
Having a look at the phyloegnetic composition of you data is useful for many reasons. FOr this purpose we have two Phylum and Family level plots.
ps3.com <- ps3 # create a new pseq object # We need to set Palette taxic <- as.data.frame(ps3.com@tax_table) # this will help in setting large color options colourCount = length(unique(taxic$Phylum)) #define number of variable colors based on number of Family (change the level accordingly to phylum/class/order) getPalette = colorRampPalette(brewer.pal(12, col.palette)) # change the palette as well as the number of colors will change according to palette. # now edit the unclassified taxa tax_table(ps3.com)[tax_table(ps3.com)[, "Phylum"] == "f__", "Phylum"] <- "f__Unclassified Phylum" # We will also remove the 'f__' patterns for cleaner labels tax_table(ps3.com)[, colnames(tax_table(ps3.com))] <- gsub(tax_table(ps3.com)[, colnames(tax_table(ps3.com))], pattern = "[a-z]__", replacement = "") otu.df <- as.data.frame(otu_table(ps3.com)) # make a dataframe for OTU information. # head(otu.df) # check the rows and columns taxic$OTU <- row.names.data.frame(otu.df) # Add the OTU ids from OTU table into the taxa table at the end. colnames(taxic) # You can see that we now have extra taxonomy levels. taxmat <- as.matrix(taxic) # convert it into a matrix. new.tax <- tax_table(taxmat) # convert into phyloseq compaitble file. tax_table(ps3.com) <- new.tax # incroporate into phyloseq Object # it would be nice to have the Taxonomic names in italics. # for that we set this guide_italics <- guides(fill = guide_legend(label.theme = element_text(size = 15, face = "italic", colour = "Black", angle = 0))) ## Now we need to plot at family level, We can do it as follows: # first remove the phy_tree ps3.com@phy_tree <- NULL ps3.com.phy <- aggregate_taxa(ps3.com, "Phylum") ps3.com.phy.rel <- microbiome::transform(ps3.com.phy, "compositional") plot.composition.relAbun.phy <- plot_composition(ps3.com.phy.rel) + theme(legend.position = "bottom") + scale_fill_manual(values = getPalette(colourCount)) + theme_bw() + theme(axis.text.x = element_text(angle = 90)) + ggtitle("Relative abundance Phylum level") + guide_italics plot.composition.relAbun.phy if (nrow(metadf) > 30) { ggsave("./Others/compositionbarplot_Phylum.pdf", plot = plot.composition.relAbun.phy, height = 8, width = 28) } else { ggsave("./Others/compositionbarplot_Phylum.pdf", plot = plot.composition.relAbun.phy, height = 8, width = 18) }
colourCount = length(unique(taxic$Family)) #define number of variable colors based on number of Family (change the level accordingly to phylum/class/order) getPalette = colorRampPalette(brewer.pal(12, col.palette)) # change the palette as well as the number of colors will change according to palette. ps3.com.fam <- aggregate_taxa(ps3.com, "Family") ps3.com.fam.rel <- microbiome::transform(ps3.com.fam, "compositional") plot.composition.relAbun.fam <- plot_composition(ps3.com.fam.rel) + theme(legend.position = "bottom") + scale_fill_manual(values = getPalette(colourCount)) + theme_bw() + theme(axis.text.x = element_text(angle = 90)) + ggtitle("Relative abundance Phylum level") + guide_italics plot.composition.relAbun.fam if (nrow(metadf) > 30) { ggsave("./Others/compositionbarplot_Family.pdf", plot = plot.composition.relAbun.fam, height = 8, width = 28) } else { ggsave("./Others/compositionbarplot_Family.pdf", plot = plot.composition.relAbun.fam, height = 8, width = 18) }
The counts are compositionally tranformed and then used for ordinations.
ps3.rel <- microbiome::transform(ps3, "compositional") bc.pcoa <- ordinate(ps3.rel, method = "PCoA", distance = "bray") bc.pcoa.plot <- plot_ordination(ps3.rel, bc.pcoa, type = "split", axes = 1:2, color = VariableA, shape = VariableB, label = NULL, title = "Bray-Curtis distance PCoA", justDF = FALSE) bc.pcoa.plot <- bc.pcoa.plot + theme_bw() + geom_point(size = 2) print(bc.pcoa.plot) ggsave("./BetaDiversity/Bray-Curtis distance PCoA.pdf", plot = bc.pcoa.plot, height = 6, width = 10) # Calculate bray curtis distance matrix ps3_bray <- phyloseq::distance(ps3.rel, method = "bray") # use meta data from phylogenetic div code chunk. # Adonis test adonis(ps3_bray ~ VariableA, data = select.meta) # Homogeneity of dispersion test beta.bray <- betadisper(ps3_bray, select.meta$VariableA) permutest(beta.bray)
The counts are converted to relative abundance and then used for ordinations.
wunifrac.pcoa <- ordinate(ps3.rel, method = "PCoA", distance = "wunifrac") wunifrac.pcoa.plot <- plot_ordination(ps3.rel, wunifrac.pcoa, type = "split", axes = 1:2, color = VariableA, shape = VariableB, label = NULL, title = "Weighted Unifrac distance PCoA", justDF = FALSE) wunifrac.pcoa.plot <- wunifrac.pcoa.plot + theme_bw() + geom_point(size = 2) print(wunifrac.pcoa.plot) ggsave("./BetaDiversity/Weighted Unifrac distance PCoA.pdf", plot = wunifrac.pcoa.plot, height = 6, width = 10) # Calculate bray curtis distance matrix ps3_wunifrac <- phyloseq::distance(ps3.rel, method = "wunifrac") # use meta data from phylogenetic div code chunk. # Adonis test adonis(ps3_wunifrac ~ VariableA, data = select.meta) # Homogeneity of dispersion test beta.wunifrac <- betadisper(ps3_wunifrac, select.meta$VariableA) permutest(beta.wunifrac)
The counts are converted to relative abundnces and then used for ordinations.
unifrac.pcoa <- ordinate(ps3.rel, method = "PCoA", distance = "unifrac") unifrac.pcoa.plot <- plot_ordination(ps3.rel, unifrac.pcoa, type = "split", axes = 1:2, color = VariableA, shape = VariableB, label = NULL, title = "Unweighted Unifrac distance PCoA", justDF = FALSE) unifrac.pcoa.plot <- unifrac.pcoa.plot + theme_bw() + geom_point(size = 2) print(unifrac.pcoa.plot) ggsave("./BetaDiversity/Unweighted Unifrac distance PCoA.pdf", plot = unifrac.pcoa.plot, height = 6, width = 10) message("Unweighted Unifrac distance gives negative values and standard anova and anoism cannot be used")
if (heatmap == TRUE) { pseq.fam <- aggregate_taxa(ps3.com, "Family") pseq.fam.rel <- transform(pseq.fam, "compositional") ps3.rel.heatmap <- plot_composition(pseq.fam.rel, sample.sort = VariableB, otu.sort = NULL, x.label = VariableA, plot.type = "heatmap", verbose = FALSE) ps3.rel.heatmap <- ps3.rel.heatmap + theme(legend.position = "bottom") + theme_bw() + theme(axis.text.x = element_text(angle = 90)) + ggtitle("Heatmap relative abundance") + theme(axis.text = element_text(face = "italic")) ps3.rel.heatmap <- ps3.rel.heatmap + scale_fill_distiller(palette = "YlOrRd", trans = "reverse") print(ps3.rel.heatmap) } ggsave("./Others/Heatmap rel abun tranformed Family.pdf", plot = ps3.rel.heatmap, height = 6, width = 10)
sessionInfo()
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.