iMapPESTS local metabarcoding workflow

Demultiplex sequencing reads

For this workflow to run, we need to first demultiplex the miseq run again as the miseq does not put indexes in fasta headers by default, and also obtain some necessary files from the sequencing folder. The below code is written for the Agriculture Victoria BASC server, and the locations will be different if you are using a different HPC cluster.

The output directory should be unique for each sequencing run, named as the flowcell id, within a directory called data

For example:

root/
  ├── data/
     ├── CJL7D/

BASH:

#load module
module load bcl2fastq2/2.20.0-foss-2018b

#raise amount of available file handles
ulimit -n 4000

###Run1

#Set up input and outputs
inputdir=/group/sequencing/190412_M03633_0313_000000000-CGK9B #CHANGE TO YOUR SEQ RUN
outputdir=/group/pathogens/Alexp/Metabarcoding/imappests/data/CGK9B #CHANGE TO YOUR DATA FOLDER RUN
samplesheet=/group/pathogens/Alexp/Metabarcoding/imappests/data/CGK9B/SampleSheet_CGK9B.csv #CHANGE TO YOUR SAMPLESHEET

# convert samplesheet to unix format
dos2unix $samplesheet

#Demultiplex
bcl2fastq -p 12 --runfolder-dir $inputdir \
--output-dir $outputdir \
--sample-sheet $samplesheet \
--no-lane-splitting --barcode-mismatches 1

# Copy other necessary files and move fastqs
cd $outputdir
cp -r $inputdir/InterOp $outputdir
cp $inputdir/RunInfo.xml $outputdir
cp $inputdir/runInfo.xml $outputdir
cp $inputdir/runParameters.xml $outputdir
cp $inputdir/RunParameters.xml $outputdir
cp $samplesheet $outputdir
mv **/*.fastq.gz $outputdir

# Append fcid to start of sample names if missing
fcid=$(echo $inputdir | sed 's/^.*-//')
for i in *.fastq.gz; do
  if ! [[ $i == $fcid* ]]; then
  new=$(echo ${fcid} ${i}) #append together
  new=$(echo ${new// /_}) #remove any whitespace
  mv -v "$i" "$new"
  fi
done

Optional: Run R on BASC

You may wish to run this workflow through the BASC command line in order to take advantage of more processing power. To do this, you can start a new SLURM interactive session. Press the CODE button to the lower right to display the code for this optional step.

# Create new interactive SLURM session
sinteractive --ntasks=1 --cpus-per-task=10 --mem-per-cpu=10GB --time=72:00:00

module load R/4.1.0-foss-2021a
module load pkgconfig/1.5.1-GCCcore-9.3.0-Python-3.8.2
module load GDAL/3.3.0-foss-2021a
module load BLAST+/2.11.0-gompi-2020a
module load Pandoc/2.5

# Load R
R

# Run quit() to quit R once you are finished

Install and load R packages and setup directories

This pipeline depends on various R packages to be installed prior to running. There are now two ways of doing this, the recommended “Renv” option ensures that the exact same software versions are installed to what was used during development. The alternative “Manual” option represents the older default for this pipeline prior to June 2021 and may lead to errors due to packages changing over time.

Renv

This approach downloads all required package versions to a local cache. This can take a while the first time you run it, and it may be best to restart afterwards

The seqateurs R package also provides wrappers around other software packages for QC. For convenience we will download and install these software in a new folder called “bin”

install.packages("renv")
library(renv)

# Download the renv lockfile, which keeps track of package versions
rlock <- readLines("https://github.com/alexpiper/iMapPESTS/blob/master/renv.lock?raw=true")
writeLines(rlock, "renv.lock")

# Restore the renv to install package dependencies
renv::restore(prompt = FALSE)

# Packages to load
.packages <- c(
  "devtools",
  "ggplot2",
  "gridExtra",
  "data.table",
  "tidyverse", 
  "stringdist",
  "patchwork",
  "vegan",
  "seqinr",
  "patchwork",
  "stringi",
  "phangorn",
  "magrittr",
  "galah",
  "dada2",
  "phyloseq",
  "DECIPHER",
  "Biostrings",
  "ShortRead",
  "ggtree",
  "savR",
  "ngsReports",
  "seqateurs",
  "taxreturn",
  "afdscraper",
  "speedyseq"
  )

# Load all packages
sapply(.packages, require, character.only = TRUE)

#Install bbmap if its not in $path or in bin folder
if(Sys.which("bbduk") == "" & !file.exists("bin/bbmap/bbduk.sh")){
  seqateurs::bbmap_install(dest_dir = "bin")
}

#Install fastqc if its not in $path or in bin folder
if(Sys.which("fastqc") == "" & !file.exists("bin/FastQC/fastqc")){
  seqateurs::fastqc_install(dest_dir = "bin")
}

#Install BLAST if its not in $path or in bin folder
if(.findExecutable("blastn") == "" & (length(fs::dir_ls("bin", glob="*blastn.exe",recurse = TRUE)) ==0)){
  taxreturn::blast_install(dest_dir = "bin")
}

# Create directories
if(!dir.exists("data")){dir.create("data", recursive = TRUE)}
if(!dir.exists("reference")){dir.create("reference", recursive = TRUE)}
if(!dir.exists("output/logs")){dir.create("output/logs", recursive = TRUE)}
if(!dir.exists("output/results")){dir.create("output/results", recursive = TRUE)}
if(!dir.exists("output/rds")){dir.create("output/rds", recursive = TRUE)}
if(!dir.exists("sample_data")){dir.create("sample_data", recursive = TRUE)}
if(!dir.exists("output/results/final")) {dir.create("output/results/final", recursive = TRUE)}
if(!dir.exists("output/results/unfiltered")) {dir.create("output/results/unfiltered", recursive = TRUE)}
if(!dir.exists("output/results/filtered")) {dir.create("output/results/filtered", recursive = TRUE)}

Manual

This approach manually obtains packages from CRAN, Bioconductor and Github. This method is no longer recomended.

The seqateurs R package also provides wrappers around other software packages for QC. For convenience we will download and install these software in a new folder called “bin”

#Set required packages
.cran_packages <- c(
  "devtools",
  "ggplot2",
  "gridExtra",
  "data.table",
  "tidyverse", 
  "stringdist",
  "patchwork",
  "vegan",
  "seqinr",
  "patchwork",
  "stringi",
  "phangorn",
  "magrittr",
  "galah"
  )

.bioc_packages <- c(
  "phyloseq",
  "DECIPHER",
  "Biostrings",
  "ShortRead",
  "ggtree",
  "savR",
  "dada2",
  "ngsReports"
  )

.inst <- .cran_packages %in% installed.packages()
if(any(!.inst)) {
   install.packages(.cran_packages[!.inst])
}
.inst <- .bioc_packages %in% installed.packages()
if(any(!.inst)) {
  if (!requireNamespace("BiocManager", quietly = TRUE)){
    install.packages("BiocManager")
  }
  BiocManager::install(.bioc_packages[!.inst], ask = F)
}

#Load all published packages
sapply(c(.cran_packages,.bioc_packages), require, character.only = TRUE)

# Install and load github packages
devtools::install_github("alexpiper/seqateurs", dependencies = TRUE)
library(seqateurs)

devtools::install_github("alexpiper/taxreturn", dependencies = TRUE)
library(taxreturn)

devtools::install_github("alexpiper/afdscraper", dependencies = TRUE)
library(afdscraper)

devtools::install_github("mikemc/speedyseq", dependencies = TRUE)
library(speedyseq)

#Install bbmap if its not in $path or in bin folder
if(Sys.which("bbduk") == "" & !file.exists("bin/bbmap/bbduk.sh")){
  seqateurs::bbmap_install(dest_dir = "bin")
}

#Install fastqc if its not in $path or in bin folder
if(Sys.which("fastqc") == "" & !file.exists("bin/FastQC/fastqc")){
  seqateurs::fastqc_install(dest_dir = "bin")
}

#Install BLAST if its not in $path or in bin folder
if(.findExecutable("blastn") == "" & (length(fs::dir_ls("bin", glob="*blastn.exe",recurse = TRUE)) ==0)){
  taxreturn::blast_install(dest_dir = "bin")
}

# Create directories
if(!dir.exists("data")){dir.create("data", recursive = TRUE)}
if(!dir.exists("reference")){dir.create("reference", recursive = TRUE)}
if(!dir.exists("output/logs")){dir.create("output/logs", recursive = TRUE)}
if(!dir.exists("output/results")){dir.create("output/results", recursive = TRUE)}
if(!dir.exists("output/rds")){dir.create("output/rds", recursive = TRUE)}
if(!dir.exists("sample_data")){dir.create("sample_data", recursive = TRUE)}
if(!dir.exists("output/results/final")) {dir.create("output/results/final", recursive = TRUE)}
if(!dir.exists("output/results/unfiltered")) {dir.create("output/results/unfiltered", recursive = TRUE)}
if(!dir.exists("output/results/filtered")) {dir.create("output/results/filtered", recursive = TRUE)}

Create sample sheet

The directory structure should now look something like this:

root/
├── data/
│   ├── CJL7D/
│   │  ├── R1.fastq.gz
│   │  ├── R2.fastq.gz
│   │  ├── runInfo.xml
│   │  ├── runParameters.xml
│   │  ├── SampleSheet.csv
│   │  └── InterOp/
│   └── fcid2/
├── sample_data/
├── reference
├── bin
├── output/
└── doc/

The reference and bin folders can be copied from previous runs.

In order to track samples and relevant QC statistics throughout the metabarcoding pipeline, we will first create a new samplesheet from our input samplesheets. This function requires both the SampleSheet.csv used for the sequencing run, and the runParameters.xml, both of which should have been automatically obtained from the demultiplexed sequencing run folder in the bash step above

runs <- dir("data/") #Find all directories within data
SampleSheet <- list.files(paste0("data/", runs), pattern= "SampleSheet", full.names = TRUE)
runParameters <- list.files(paste0("data/", runs), pattern= "[Rr]unParameters.xml", full.names = TRUE)

# Create samplesheet containing samples and run parameters for all runs
samdf <- create_samplesheet(SampleSheet = SampleSheet, runParameters = runParameters, template = "V4") %>%
  distinct()

# Merge in existing sample_info metadata file if you have one
sample_info <- readxl::read_excel("sample_data/sample_infoV4.xlsx", sheet="SAMPLESHEET")
samdf <- coalesce_join(samdf, sample_info, by="sample_id")

# Create logfile containing samples and run parameters for all runs
logdf <- create_logsheet(SampleSheet = SampleSheet, runParameters = runParameters) %>%
  distinct()

#Check logdf and samdf are the same length
if(!nrow(samdf) == nrow(logdf)){
  warning("Samdf and logdf do not contain the same number of rows!")
}

# Check if sampleids contain fcid, if not; attatch
samdf <- samdf %>%
  mutate(sample_id = case_when(
    !str_detect(sample_id, fcid) ~ paste0(fcid,"_",sample_id),
    TRUE ~ sample_id
  ))
logdf <- logdf %>%
  mutate(sample_id = case_when(
    !str_detect(sample_id, fcid) ~ paste0(fcid,"_",sample_id),
    TRUE ~ sample_id
  ))

# Check if samples match samplesheet
fastqFs <- purrr::map(list.dirs("data", recursive=FALSE),
                      list.files, pattern="_R1_", full.names = TRUE) %>%
  unlist() %>%
  str_remove(pattern = "^(.*)\\/") %>%
  str_remove(pattern = "(?:.(?!_S))+$")
fastqFs <- fastqFs[!str_detect(fastqFs, "Undetermined")]

#Check missing in samplesheet
if (length(setdiff(fastqFs, samdf$sample_id)) > 0) {warning("The fastq file/s: ", setdiff(fastqFs, samdf$sample_id), " are not in the sample sheet") }

#Check missing fastqs
if (length(setdiff(samdf$sample_id, fastqFs)) > 0) {
  warning(paste0("The fastq file: ",
                 setdiff(samdf$sample_id, fastqFs),
                 " is missing, dropping from samplesheet \n")) 
  samdf <- samdf %>%
    filter(!sample_id %in% setdiff(samdf$sample_id, fastqFs))
  logdf <- logdf %>%
    filter(!sample_id %in% setdiff(logdf$sample_id, fastqFs))
}

#Write out updated sample CSV for use
write_csv(samdf, "sample_data/Sample_info.csv")
write_csv(logdf, "output/logs/logdf.csv")

Quality checks:

We will conduct 3 quality checks. Firstly a check of the entire sequence run, followed by a sample level quality check to identify potential issues with specific samples. And then a calculation of the index switching rate by summarising correctly assigned vs missasigned indices.

#Load sample sheet
samdf <- read.csv("sample_data/Sample_info.csv", stringsAsFactors = FALSE)
runs <- unique(samdf$fcid)

flowcells <- vector("list", length=length(runs))
for (i in 1:length(runs)){
  ## Run level quality check using savR
  path <- paste0("data/", runs[i], "/")
  flowcells[[i]] <- savR(path)
  fc <- flowcells[[i]]
  qc.dir <- paste0("output/logs/", runs[i],"/" )
  dir.create(qc.dir, recursive = TRUE)
  write_csv(correctedIntensities(fc), paste0(qc.dir, "correctedIntensities.csv"))
  write_csv(errorMetrics(fc), paste0(qc.dir, "errorMetrics.csv"))
  write_csv(extractionMetrics(fc), paste0(qc.dir, "extractionMetrics.csv"))
  write_csv(qualityMetrics(fc), paste0(qc.dir, "qualityMetrics.csv"))
  write_csv(tileMetrics(fc), paste0(qc.dir, "tileMetrics.csv"))

  avg_intensity <- fc@parsedData[["savCorrectedIntensityFormat"]]@data %>%
    group_by(tile, lane) %>%
    summarise(Average_intensity = mean(avg_intensity)) %>% 
    ungroup() %>%
    mutate(side = case_when(
      str_detect(tile, "^11") ~ "Top",
      str_detect(tile, "^21") ~ "Bottom"
        ))%>%
    ggplot(aes(x=lane, y=as.factor(tile), fill=Average_intensity)) +
    geom_tile() +
    facet_wrap(~side, scales="free") +
    scale_fill_viridis_c()
  
  pdf(file=paste(qc.dir, "/avgintensity.pdf", sep=""), width = 11, height = 8 , paper="a4r")
  plot(avg_intensity)
  try(dev.off(), silent=TRUE)
  
  pdf(file=paste(qc.dir, "/PFclusters.pdf", sep=""), width = 11, height = 8 , paper="a4r")
  pfBoxplot(fc)
  try(dev.off(), silent=TRUE)

  for (lane in 1:fc@layout@lanecount) {
  pdf(file=paste(qc.dir, "/QScore_L", lane, ".pdf", sep=""), width = 11, height = 8 , paper="a4r")
      qualityHeatmap(fc, lane, 1:fc@directions)
  try(dev.off(), silent=TRUE)
  } 
}
#Update log DF
logdf <- read_csv("output/logs/logdf.csv")

# Track reads
logdf <- logdf %>% 
  left_join(
    flowcells %>%
  purrr::map(~{.x@parsedData[["savTileFormat"]]@data %>%
  dplyr::filter(code %in% c(100,101)) %>%
  dplyr::mutate(code = case_when(
    code == 100 ~ "reads_total",
    code == 101 ~ "reads_pf"
  ))}) %>%
  purrr::set_names(runs) %>%
  bind_rows(.id="fcid") %>% 
  group_by(fcid, code) %>%
  summarise(reads = sum(value)) %>%
  pivot_wider(names_from = code,
              values_from = reads),
  by="fcid")
  
write_csv(logdf, "output/logs/logdf.csv")

## Sample level quality check using fastqc
for (i in 1:length(runs)){
  path <- paste0("data/", runs[i], "/")
  qc.dir <- paste0("output/logs/", runs[i],"/FASTQC" )
  dir.create(qc.dir, recursive=TRUE)
  qc_out <- seqateurs::fastqc(fq.dir = path, qc.dir = qc.dir, fastqc.path = "bin/FastQC/fastqc", threads=2)
  writeHtmlReport(qc.dir, overwrite = TRUE, gcType ="Genome",  quiet=FALSE)
}

## Calculate index switching
for (i in 1:length(runs)){
  path <- paste0("data/", runs[i], "/")
  qc.dir <- paste0("output/logs/", runs[i] )
  run_data <- samdf %>%
    filter(fcid == runs[i])

  indices <- sort(list.files(path, pattern="_R1_", full.names = TRUE)) %>%
    purrr::set_names() %>%
    purrr::map(seqateurs::summarise_index) %>%
    bind_rows(.id="Sample_Name")%>%
    arrange(desc(Freq)) %>% 
    dplyr::mutate(Sample_Name = Sample_Name %>% 
                    str_remove(pattern = "^(.*)\\/") %>%
                    str_remove(pattern = "(?:.(?!_S))+$"))

  if(!any(str_detect(indices$Sample_Name, "Undetermined"))){
    stop("Error, an Undetermined reads fastq must be present to calculate index switching")
    }
  
  combos <- indices %>% 
    dplyr::filter(!str_detect(Sample_Name, "Undetermined")) %>%
    dplyr::select(index, index2) %>%
    tidyr::expand(index, index2)

  #get unused combinations resulting from index switching
  switched <- left_join(combos, indices, by=c("index", "index2")) %>%
    drop_na()
  
  other_reads <- anti_join(indices,combos, by=c("index", "index2")) %>%
    summarise(sum = sum(Freq)) %>%
    pull(sum)
  
  #Summary of index switching rate
  exp_rate <- switched %>% 
    filter(!str_detect(Sample_Name, "Undetermined"))
  obs_rate <- switched %>% 
    filter(str_detect(Sample_Name,"Undetermined"))
  switch_rate <- (sum(obs_rate$Freq)/sum(exp_rate$Freq))
    message("Index switching rate calculated as: ", switch_rate)

  #Plot switching
    gg.switch <- switched %>%
     # mutate(index = factor(index, levels = index),
     #        index2 = factor(index2, levels = index)) %>%
      ggplot(aes(x = index, y = index2), stat="identity") +
    geom_tile(aes(fill = Freq),alpha=0.8)  + 
    scale_fill_viridis_c(name="log10 Reads", begin=0.1, trans="log10")+
    theme(axis.text.x = element_text(angle=90, hjust=1), 
          plot.title=element_text(hjust = 0.5),
          plot.subtitle =element_text(hjust = 0.5)#,
          #legend.position = "none"
          ) +
    labs(title= runs[i], subtitle = paste0(
      "Total Reads: ", sum(indices$Freq),
      ", Switch rate: ", sprintf("%1.4f%%", switch_rate*100),
      ", other Reads: ", other_reads)) 
  pdf(file=paste(qc.dir, "/switchrate.pdf", sep=""), width = 11, height = 8 , paper="a4r")
      plot(gg.switch)
  try(dev.off(), silent=TRUE)
  
  }

Trim Primers

DADA2 requires Non-biological nucleotides i.e. primers, adapters, linkers, etc to be removed. Following demultiplexing however primer sequences still remain in the reads and must be removed prior to use with the DADA2 algorithm. For this workflow we will be using the Kmer based adapter trimming software BBDuk (Part of BBTools package https://jgi.doe.gov/data-and-tools/bbtools/) to trim the primers from our raw data files. the seqateurs R package contains a wrapper fucntion to call bbduk from R to trim primers.

If multiple primers have been multiplexed per library, use the multiplexed primer option below, otherwise proceed with the regular single primer workflow.

Single primer

This workflow is for a single primer-pair per library. For this workflow to run, the pcr_primers, for_primer_seq and rev_primer_seq fields in the sample sheet must contain the primer information.

#Load sample sheet
samdf <- read.csv("sample_data/Sample_info.csv", stringsAsFactors = FALSE)
logdf <- read_csv("output/logs/logdf.csv")
runs <- unique(samdf$fcid)

#Create lists to track reads
trimmed <- vector("list", length = length(runs))
demux <- vector("list", length = length(runs))

#Check primers are present
if(any(is.na(samdf$for_primer_seq), is.na(samdf$rev_primer_seq))){warning("Some primer sequences are missing from samdf, check manually")}

i=1
for (i in 1:length(runs)){
  path <- paste0("data/", runs[i])
  qc.dir <- paste0("output/logs/", runs[i])

  run_data <- samdf %>%
    dplyr::filter(fcid == runs[i])
  
  #Get primer sequences
  primers <- na.omit(c(unique(run_data$for_primer_seq), unique(run_data$rev_primer_seq)))
  
  # check if any samples need a second round of demultiplexing
  twintagged <- any(str_detect(primers, ";"))
  if (twintagged == TRUE) stop("Multiple primers are listed per sample in the sample data sheet, use the multi-primer workflow instead")

  fastqFs <- sort(list.files(paste0(path), pattern="_R1_", full.names = TRUE))
  fastqRs <- sort(list.files(paste0(path), pattern="_R2_", full.names = TRUE))
  if(length(fastqFs) != length(fastqRs)) stop(paste0("Forward and reverse files for ",runs[i]," do not match."))
    
  # If there is multiple primer combination per run, do each seperately
  primer_runs <- unique(run_data$pcr_primers)
  
  for (p in 1:length(primer_runs)){
    primer_data <- run_data %>% dplyr::filter(pcr_primers == primer_runs[p])
    
    Fprimers <- unlist(str_split(unique(primer_data$for_primer_seq), ";"))
    Rprimers <- unlist(str_split(unique(primer_data$rev_primer_seq), ";"))
    
    # Subset fastqs to only those samples with target primers
    fastqFs_primer <- fastqFs[sapply(fastqFs, function(x){any(str_detect(x, primer_data$sample_id))})]
    fastqRs_primer <- fastqRs[sapply(fastqRs, function(x){any(str_detect(x, primer_data$sample_id))})]

    trimmed[[i]] <- purrr::map2_dfr(fastqFs_primer, fastqRs_primer, function(x,y){
        bbtrim2(install="bin/bbmap", fwd = x, rev = y,
              primers = c(Fprimers, Rprimers), checkpairs = FALSE,
              degenerate = TRUE, out.dir=file.path(path, "01_trimmed"), trim.end = "left", 
              kmer=NULL, tpe=TRUE, tbo=TRUE,
              ordered = TRUE, mink = FALSE, hdist = 2,
              maxlength =(max(run_data$for_read_length,
                              run_data$rev_read_length) - sort(nchar(c(Fprimers, Rprimers)),
                              decreasing = FALSE)[1]) +5, 
              force = TRUE, quiet=FALSE)
      })
  }
  
  # Check sequence lengths
  pre_trim <- plot_lengths(dir=path, aggregate=TRUE, sample=1e5) + 
    labs(title = runs[i], subtitle = "Pre-trimming")
  post_trim <- plot_lengths(dir=paste0(path, "/01_trimmed/"), aggregate=TRUE, sample=1e5)+ 
    labs(title = runs[i], subtitle = "Post-trimming")

  pdf(file=file.path(qc.dir, "readlengths.pdf"), width = 11, height = 8 , paper="a4r")
    plot(pre_trim)
    plot(post_trim)
  try(dev.off(), silent=TRUE)
  
  trim_summary <- trimmed[[i]] %>% 
    mutate(perc_reads_remaining = signif(((output_reads / input_reads) * 100), 2),
           perc_bases_remaining = signif(((output_bases / input_bases) * 100), 2)
           ) %>%
    filter(!is.na(perc_reads_remaining))
    
  message(paste0(signif(mean(trim_summary$perc_reads_remaining, na.rm = TRUE), 2),
                 "% of reads and ",
                 signif(mean(trim_summary$perc_bases_remaining, na.rm = TRUE), 2),
                 "% of bases remaining for ", runs[i]," after trimming"))
  
  # Print warning for each sample
  for(w in 1:nrow(trim_summary)){
    if (trim_summary[w,]$perc_reads_remaining < 10) {message(paste0("WARNING: Less than 10% bases remaining for ",trim_summary[w,]$sample), ", check primer sequences are correct")}
  }
}

# Track reads
logdf <- logdf %>% 
  left_join(
    trimmed %>%
    purrr::set_names(runs) %>%
    bind_rows(.id="fcid") %>%
    mutate(sample_id = str_replace(basename(sample), pattern="_S.*$", replacement=""),
           reads_demulti = input_reads/2,
           reads_trimmed = output_reads/2) %>%
    dplyr::select(fcid, sample_id, reads_demulti, reads_trimmed),
  by=c("sample_id", "fcid"))

write_csv(logdf, "output/logs/logdf.csv")

Multiplexed primers

This workflow is for multiple primer-pairs per library. These could either be targeting different gene regions, taxa, or be twin-tagged replicate primers.

For the multiplexed workflow to run, the pcr_primers, for_primer_seq and rev_primer_seq fields in the sample sheet must contain all primers separated by a semicolon ‘;’ For example:

pcr_primers Sterno18SF2-Sterno18SR2;Sterno12SF2-Sterno12SR2;SternoCOIF1-SternoCOIR1

for_primer_seq ATGCATGTCTCAGTGCAAG;CAYCTTGACYTAACAT;ATTGGWGGWTTYGGAAAYTG

rev_primer_seq TCGACAGTTGATAAGGCAGAC;TAAAYYAGGATTAGATACCC;TATRAARTTRATWGCTCCTA

#Load sample sheet
samdf <- read.csv("sample_data/Sample_info.csv", stringsAsFactors = FALSE)
logdf <- read_csv("output/logs/logdf.csv")
runs <- unique(samdf$fcid)

#Create lists to track reads
trimmed <- vector("list", length = length(runs))
demux <- vector("list", length = length(runs))

#Check primers are present
if(any(is.na(samdf$for_primer_seq), is.na(samdf$rev_primer_seq))){warning("Some primer sequences are missing from samdf, check manually")}

# check if any samples need a second round of demultiplexing
i=1
for (i in 1:length(runs)){
  path <- paste0("data/", runs[i])
  qc.dir <- paste0("output/logs/", runs[i])

  run_data <- samdf %>%
    dplyr::filter(fcid == runs[i])
  
  #Get primer sequences
  primers <- na.omit(c(unique(run_data$for_primer_seq), unique(run_data$rev_primer_seq)))

  #Check if samples were twin tagged - these require extra round of demultiplexing
  twintagged <- any(str_detect(primers, ";"))
  if (twintagged == TRUE) {
      # Create output directory
      demuxpath <- file.path(path, "00_demux")
      dir.create(demuxpath)

      # Get list of fasta files
      fastqFs <- sort(list.files(path, pattern="*R1_001.*", full.names = TRUE))
      fastqRs <- sort(list.files(path, pattern="*R2_001.*", full.names = TRUE))
        
      # If there is multiple primer combination per run, do each seperately
      primer_runs <- unique(run_data$pcr_primers)
      for (p in 1:length(primer_runs)){
        primer_data <- run_data %>% dplyr::filter(pcr_primers == primer_runs[p])
        
        Fprimers <- unlist(str_split(unique(primer_data$for_primer_seq), ";"))
        Rprimers <- unlist(str_split(unique(primer_data$rev_primer_seq), ";"))
        primer_names <- unlist(str_split(unique(primer_data$pcr_primers), ";"))
        
        # Subset fastqs to only those samples with target primers
        fastqFs_primer <- fastqFs[sapply(fastqFs, function(x){any(str_detect(x, primer_data$sample_id))})]
        fastqRs_primer <- fastqRs[sapply(fastqRs, function(x){any(str_detect(x, primer_data$sample_id))})]
        
        demux <- purrr::map2_dfr(fastqFs_primer, fastqRs_primer, function(x,y){
          bbdemux2(install="bin/bbmap", fwd=x, rev=y, Fbarcodes = Fprimers, 
                      Rbarcodes = Rprimers, names=primer_names, degenerate=TRUE, out.dir=demuxpath,
                      threads=1 , mem=4,  hdist=0, force=TRUE)
        })
        
        # Rename output files to match imappests format (FCID_sample_ext1_pcr1_CALngsF1-CALngsR1_S12_R1R2_001.fastq.gz)
        old_names <- sort(list.files(paste0(demuxpath), pattern="_R1R2_", full.names = TRUE)) 
        old_names <- old_names[sapply(old_names, function(x){any(str_detect(x, primer_data$sample_id))})]
        new_names <- old_names %>%
          basename() %>%
          str_remove(".fastq.gz") %>%
          str_split("_", n=Inf) %>%
          purrr::map_chr(function(x){
            paste0(paste0( c(x[1:4], x[length(x)], x[5:(length(x)-1)]), collapse = "_"),".fastq.gz")
                      }) 
        file.remove(file.path(dirname(old_names), new_names))
        file.rename(old_names, file.path(dirname(old_names), new_names))

        # Trim primers from demultiplexed fastq
        demux_fastqs <- sort(list.files(paste0(demuxpath), pattern="_R1R2_", full.names = TRUE))
        demux_fastqs <- demux_fastqs[sapply(demux_fastqs, function(x){any(str_detect(x, primer_data$sample_id))})]

        trimmed[[i]] <- purrr::map_dfr(demux_fastqs, function(x){
          bbtrim2(install="bin/bbmap", fwd = x,
                primers = c(Fprimers, Rprimers), checkpairs = FALSE,
                degenerate = TRUE, out.dir=file.path(path, "01_trimmed"), trim.end = "left", 
                kmer=NULL, tpe=TRUE, tbo=TRUE,
                ordered = TRUE, mink = FALSE, hdist = 2,
                maxlength =(max(run_data$for_read_length, run_data$rev_read_length) - sort(nchar(c(Fprimers, Rprimers)), decreasing = FALSE)[1]) +5, force = TRUE, quiet=FALSE)
        })
        
        # Re-split interleaved fastq's
        trimmedpath <- file.path(path, "01_trimmed") 
        trimmed_fastqs <- sort(list.files(trimmedpath, pattern="_R1R2_", full.names = TRUE))
        purrr::walk(trimmed_fastqs, function(x){
          bbsplit2(install="bin/bbmap", file=x, force=TRUE)
        })
        }
    } else if (twintagged == FALSE) {
      
      fastqFs <- sort(list.files(paste0(path), pattern="_R1_", full.names = TRUE))
      fastqRs <- sort(list.files(paste0(path), pattern="_R2_", full.names = TRUE))
      if(length(fastqFs) != length(fastqRs)) stop(paste0("Forward and reverse files for ",runs[i]," do not match."))
    
      # If there is multiple primer combination per run, do each seperately
      primer_runs <- unique(run_data$pcr_primers)
      
      for (p in 1:length(primer_runs)){
        primer_data <- run_data %>% dplyr::filter(pcr_primers == primer_runs[p])
        
        Fprimers <- unlist(str_split(unique(primer_data$for_primer_seq), ";"))
        Rprimers <- unlist(str_split(unique(primer_data$rev_primer_seq), ";"))
        
        # Subset fastqs to only those samples with target primers
        fastqFs_primer <- fastqFs[sapply(fastqFs, function(x){any(str_detect(x, primer_data$sample_id))})]
        fastqRs_primer <- fastqRs[sapply(fastqRs, function(x){any(str_detect(x, primer_data$sample_id))})]

        trimmed[[i]] <- purrr::map2_dfr(fastqFs_primer, fastqRs_primer, function(x,y){
            bbtrim2(install="bin/bbmap", fwd = x, rev = y,
                  primers = c(Fprimers, Rprimers), checkpairs = FALSE,
                  degenerate = TRUE, out.dir=file.path(path, "01_trimmed"), trim.end = "left", 
                  kmer=NULL, tpe=TRUE, tbo=TRUE,
                  ordered = TRUE, mink = FALSE, hdist = 2,
                  maxlength =(max(run_data$for_read_length, run_data$rev_read_length) - sort(nchar(c(Fprimers, Rprimers)), decreasing = FALSE)[1]) +5, force = TRUE, quiet=FALSE)
          })
      }
  }
  
  # Check sequence lengths
  pre_trim <- plot_lengths(dir=path, aggregate=TRUE, sample=1e5) + 
    labs(title = runs[i], subtitle = "Pre-trimming")
  post_trim <- plot_lengths(dir=paste0(path, "/01_trimmed/"), aggregate=TRUE, sample=1e5)+ 
    labs(title = runs[i], subtitle = "Post-trimming")

  pdf(file=file.path(qc.dir, "readlengths.pdf"), width = 11, height = 8 , paper="a4r")
    plot(pre_trim)
    plot(post_trim)
  try(dev.off(), silent=TRUE)
  
  trim_summary <- trimmed[[i]] %>% 
    mutate(perc_reads_remaining = signif(((output_reads / input_reads) * 100), 2),
           perc_bases_remaining = signif(((output_bases / input_bases) * 100), 2)
           ) %>%
    filter(!is.na(perc_reads_remaining))
    
  message(paste0(signif(mean(trim_summary$perc_reads_remaining, na.rm = TRUE), 2),
                 "% of reads and ",
                 signif(mean(trim_summary$perc_bases_remaining, na.rm = TRUE), 2),
                 "% of bases remaining for ", runs[i]," after trimming"))
  
  # Print warning for each sample
  for(w in 1:nrow(trim_summary)){
    if (trim_summary[w,]$perc_reads_remaining < 10) {message(paste0("WARNING: Less than 10% bases remaining for ",trim_summary[w,]$sample), ", check primer sequences are correct")}
  }
}

#Update the sample sheet and logging sheet to deal with any newly demultiplexed files
demultiplexed_samples <- samdf %>%
  dplyr::select(sample_id, pcr_primers)

samdf <- samdf %>%
 group_by(sample_id) %>%
 group_split() %>%
 purrr::map(function(x){
   if(any(str_detect(x$pcr_primers, ";"))){
     primer_names <- unlist(str_split(unique(x$pcr_primers), ";")) 
     x %>% 
       mutate(count = length(primer_names)) %>% #Replicate the samples
       uncount(count) %>%
       mutate(pcr_primers = unlist(str_split(unique(x$pcr_primers), ";")),
              for_primer_seq = unlist(str_split(unique(x$for_primer_seq), ";")),
              rev_primer_seq = unlist(str_split(unique(x$rev_primer_seq), ";")),
              sample_id = paste0(sample_id, "_",pcr_primers)
            ) 
   } else (x)
 }) %>%
  bind_rows()

logdf <- logdf %>%
 left_join(demultiplexed_samples) %>%
 group_by(sample_id) %>%
 group_split() %>%
 purrr::map(function(x){
   if(any(str_detect(x$pcr_primers, ";"))){
   x %>% 
     mutate(count = length(primer_names)) %>% #Replicate the samples
     uncount(count) %>%
     mutate(pcr_primers = unlist(str_split(unique(x$pcr_primers), ";")),
            sample_id = paste0(sample_id, "_",pcr_primers)
            ) 
   } else (x)
 }) %>%
  bind_rows()

# Track reads
logdf <- logdf %>% 
  left_join(
    trimmed %>%
    purrr::set_names(runs) %>%
    bind_rows(.id="fcid") %>%
    mutate(sample_id = str_replace(basename(sample), pattern="_S.*$", replacement=""),
           reads_demulti = input_reads/2,
           reads_trimmed = output_reads/2) %>%
    dplyr::select(fcid, sample_id, reads_demulti, reads_trimmed),
  by=c("sample_id", "fcid"))

write_csv(logdf, "output/logs/logdf.csv")

Plot read quality & lengths

#Load sample sheet
samdf <- read.csv("sample_data/Sample_info.csv", stringsAsFactors = FALSE)
runs <- unique(samdf$fcid)

# Plotting parameters
readQC_aggregate <- TRUE
readQC_subsample <-  12

amplicon = 205 # Set to maximum size between the two primers. If working with variable barcode lengths, set to the expected or average amplicon length

for (i in 1:length(runs)){
  run_data <- samdf %>%
    filter(fcid == runs[i])

  path <- paste0("data/", runs[i], "/01_trimmed" )
 
  ##Get trimmed files, accounting for empty files (28 indicates empty sample)
  trimmedFs <- sort(list.files(path, pattern="_R1_", full.names = TRUE))
  trimmedFs <- trimmedFs[!str_detect(trimmedFs, "Undetermined")]
  trimmedFs <- trimmedFs[file.size(trimmedFs) > 28]

  #Choose a random subsample for quality checks
  sampleF <- sample(trimmedFs, readQC_subsample) #NOTE - need to have option to pass
  sampleR <- sampleF %>% str_replace(pattern="_R1_", replacement = "_R2_")
  
  #Estimate an optimat trunclen
  truncLen <- estimate_trunclen(sampleF, sampleR, maxlength=amplicon)

  #Plot qualities
  gg.Fqual <- plot_quality(sampleF) +
    geom_vline(aes(xintercept=truncLen[1]), colour="blue") +
    annotate("text", x = truncLen[1]-10, y =2, label = paste0("Suggested truncLen = ", truncLen[1]), colour="blue") +
    ggtitle(paste0(runs[i], " Forward Reads")) +
    scale_x_continuous(breaks=seq(0,300,25))
  gg.Fee <- plot_maxEE(sampleF) + 
    geom_vline(aes(xintercept=truncLen[1]), colour="blue")+
    annotate("text", x = truncLen[1]-10, y =-3, label = paste0("Suggested truncLen = ", truncLen[1]), colour="blue") +
    ggtitle(paste0(runs[i], " Forward Reads")) +
    scale_x_continuous(breaks=seq(0,300,25)) +
    theme(legend.position = "bottom")
  gg.Rqual <- plot_quality(sampleR) + 
    geom_vline(aes(xintercept=truncLen[2]), colour="blue")+
    annotate("text", x = truncLen[1]-10, y =2, label = paste0("Suggested truncLen = ", truncLen[2]), colour="blue") +
    ggtitle(paste0(runs[i], " Reverse Reads")) +
    scale_x_continuous(breaks=seq(0,300,25)) +
    theme(legend.position = "bottom")
  gg.Ree <- plot_maxEE(sampleR) +
    geom_vline(aes(xintercept=truncLen[2]), colour="blue")+
    annotate("text", x = truncLen[1]-10, y =-3, label = paste0("Suggested truncLen = ", truncLen[2]), colour="blue") +
    ggtitle(paste0(runs[i], " Reverse Reads")) +
    scale_x_continuous(breaks=seq(0,300,25)) +
    theme(legend.position = "bottom")

  Qualplots <- (gg.Fqual + gg.Rqual) / (gg.Fee + gg.Ree)
  
  #output plots
  pdf(paste0("output/logs/",runs[i],"/",runs[i], "_prefilt_quality.pdf"), width = 11, height = 8 , paper="a4r")
  plot(Qualplots)
  try(dev.off(), silent=TRUE)
}

This has output a prefilt_quality.pdf plot for each of the runs analysed in the logs folder. On the top is the quality score per cycle, and on the bottom is the cumulative expected errors (calculated as EE = sum(10^(-Q/10)) on a log scale. For the quality plot, the median quality score at each position is shown by the green line, and the quartiles of the quality score distribution by the orange lines. For the maxEE lines, the red lines showing the expected error filter options. The blue vertical line on both plots shows the suggested truncLen option automatically determined.

Ensure that the blue suggested trunclen looks reasonable before continuing. Truncating length will reduce the number of reads violating the expected error filter, and therefore increase the number of reads proceding through the pipeline. The reverse reads will generally have lower quality, and therefore a lower truncLen than the forward reads.

Filter reads

This stage will use read truncation and max expected error function to remove low quality reads and read tails. All reads containing N bases will also be removed. this will output _postfilt_quality.pdf in the logs folder to determine how sucessfull it has been in cleaning up the quality.

Non-length variable

samdf <- read.csv("sample_data/Sample_info.csv", stringsAsFactors = FALSE)
runs <- unique(samdf$fcid)
filtered_out <- vector("list", length=length(runs))

# Set important variables for trimming
maxEE <- 1 #Filter reads above Expected errors (EE = sum(10^(-Q/10))). Set higher for poor quality sequences.
rm.lowcomplex <- 0 # Remove low-complexity, set higher for NovaSeq and other 2 colour platforms
amplicon = 205 # Set to maximum size between the two primers. If working with variable barcode lengths, set to readlength

# Estimate best length to truncate forward and reverse reads to
#truncLen <- estimate_trunclen(sampleF, sampleR, maxlength=amplicon)

for (i in 1:length(runs)){
  
   run_data <- samdf %>%
    filter(fcid == runs[i])
  
  path <- paste0("data/", runs[i], "/01_trimmed" )
  
  filtpath <- paste0("data/", runs[i], "/02_filtered" ) # Filtered forward files go into the path/filtered/ subdirectory
  dir.create(filtpath)
  
  fastqFs <- sort(list.files(path, pattern="_R1_001.*"))
  fastqRs <- sort(list.files(path, pattern="_R2_001.*"))
  
  if(length(fastqFs) != length(fastqRs)) stop(paste0("Forward and reverse files for ",runs[i]," do not match."))
  
  filtered_out[[i]] <- filterAndTrim(fwd = file.path(path, fastqFs), filt = file.path(filtpath, fastqFs),
                                      rev = file.path(path, fastqRs), filt.rev = file.path(filtpath, fastqRs),
                                      maxEE = maxEE, truncLen = truncLen, rm.lowcomplex = rm.lowcomplex,
                                      rm.phix = TRUE, matchIDs = TRUE, id.sep = "\\s",
                                      multithread = TRUE, compress = TRUE, verbose = TRUE)

  # post filtering plot
  filtFs <- sort(list.files(filtpath, pattern="R1_001.*", full.names = TRUE))
  sampleF <- sample(filtFs, readQC_subsample)
  sampleR <- sampleF %>% str_replace(pattern="R1_001", replacement = "R2_001")
  
  p1 <- plotQualityProfile(sampleF, aggregate = readQC_aggregate) +
    ggtitle(paste0(runs[i]," Forward Reads")) +
    scale_x_continuous(breaks=seq(0,300,25))
  p2 <- plotQualityProfile(sampleR, aggregate = readQC_aggregate) + 
    ggtitle(paste0(runs[i]," Reverse Reads"))+
    scale_x_continuous(breaks=seq(0,300,25))
  
  #output plots
  if (!dir.exists("output/logs/")){ dir.create("output/logs/")}
  pdf(paste0("output/logs/", runs[i],"/",runs[i], "_postfilt_quality.pdf"), width = 11, height = 8 , paper="a4r")
  plot(p1)
  plot(p2)
  try(dev.off(), silent=TRUE)
  
  filtered_summary <- filtered_out[[i]] %>% 
    as.data.frame() %>%
    rownames_to_column("sample") %>%
    mutate(reads_remaining = signif(((reads.out / reads.in) * 100), 2)) %>%
    filter(!is.na(reads_remaining))
    
  message(paste0(signif(mean(filtered_summary$reads_remaining, na.rm = TRUE), 2), "% of reads remaining for ", runs[i]," after filtering"))
  
  # Print warning for each sample
  for(w in 1:nrow(filtered_summary)){
    if (filtered_summary[w,]$reads_remaining < 10) {
      message(paste0("WARNING: Less than 10% reads remaining for ", filtered_summary[w,]$sample), "Check filtering parameters ")
    } 
  }
  
}

#Update log DF
logdf <- read_csv("output/logs/logdf.csv")

logdf <- logdf %>%
  left_join(filtered_out %>%
    map(as_tibble, rownames=NA) %>%
    map(rownames_to_column, var="sample_id") %>%
    purrr::set_names(runs) %>%
    bind_rows(.id="fcid") %>%
    mutate(sample_id = str_replace(basename(sample_id), pattern="_S.*$", replacement="")) %>%
    dplyr::select(fcid, sample_id, reads_qualfilt = reads.out),
  by=c("sample_id", "fcid"))

write_csv(logdf, "output/logs/logdf.csv")

length variable marker

The main difference between the filtering and trimming for a length variable marker, is we do not want to truncate the reads to a certain length, and instead use a minimum and maximum length.

samdf <- read.csv("sample_data/Sample_info.csv", stringsAsFactors = FALSE)
runs <- unique(samdf$fcid)
filtered_out <- vector("list", length=length(runs))

# Set important variables for trimming
maxEE <- 1 #Filter reads above Expected errors (EE = sum(10^(-Q/10))). Set higher for poor quality sequences.
rm.lowcomplex <- 0 # Remove low-complexity, set higher for NovaSeq and other 2 colour platforms
minlength <- 50
maxlength <- 225

for (i in 1:length(runs)){
  
   run_data <- samdf %>%
    filter(fcid == runs[i])
  
  path <- paste0("data/", runs[i], "/01_trimmed" )
  
  filtpath <- paste0("data/", runs[i], "/02_filtered" ) # Filtered forward files go into the path/filtered/ subdirectory
  dir.create(filtpath)
  
  fastqFs <- sort(list.files(path, pattern="R1_001.*"))
  fastqRs <- sort(list.files(path, pattern="R2_001.*"))
  
  if(length(fastqFs) != length(fastqRs)) stop(paste0("Forward and reverse files for ",runs[i]," do not match."))
  
  filtered_out[[i]] <- filterAndTrim(fwd = file.path(path, fastqFs), filt = file.path(filtpath, fastqFs),
                                      rev = file.path(path, fastqRs), filt.rev = file.path(filtpath, fastqRs),
                                      maxEE = maxEE, truncLen = 0, minLen = minlength, maxLen = maxlength ,
                                      rm.lowcomplex = rm.lowcomplex, rm.phix = TRUE, 
                                      matchIDs = TRUE, id.sep = "\\s",
                                      multithread = TRUE, compress = TRUE, verbose = TRUE)

  # post filtering plot
  filtFs <- sort(list.files(filtpath, pattern="R1_001.*", full.names = TRUE))
  sampleF <- sample(filtFs, readQC_subsample)
  sampleR <- sampleF %>% str_replace(pattern="R1_001", replacement = "R2_001")
  
  p1 <- plotQualityProfile(sampleF, aggregate = readQC_aggregate) +
    ggtitle(paste0(runs[i]," Forward Reads")) +
    scale_x_continuous(breaks=seq(0,300,25))
  p2 <- plotQualityProfile(sampleR, aggregate = readQC_aggregate) + 
    ggtitle(paste0(runs[i]," Reverse Reads"))+
    scale_x_continuous(breaks=seq(0,300,25))
  
  #output plots
  if (!dir.exists("output/logs/")){ dir.create("output/logs/")}
  pdf(paste0("output/logs/", runs[i],"/",runs[i], "_postfilt_quality.pdf"), width = 11, height = 8 , paper="a4r")
  plot(p1)
  plot(p2)
  try(dev.off(), silent=TRUE)
  
  filtered_summary <- filtered_out[[i]] %>% 
    as.data.frame() %>%
    rownames_to_column("sample") %>%
    mutate(reads_remaining = signif(((reads.out / reads.in) * 100), 2)) %>%
    filter(!is.na(reads_remaining))
    
  message(paste0(signif(mean(filtered_summary$reads_remaining, na.rm = TRUE), 2), "% of reads remaining for ", runs[i]," after filtering"))
  
  # Print warning for each sample
  for(w in 1:nrow(filtered_summary)){
    if (filtered_summary[w,]$reads_remaining < 10) {
      message(paste0("WARNING: Less than 10% reads remaining for ", trim_summary[w,]$sample), "Check filtering parameters ")
    } 
  }
  
}

#Update log DF
logdf <- read_csv("output/logs/logdf.csv")

logdf <- logdf %>%
  left_join(filtered_out %>%
    map(as_tibble, rownames=NA) %>%
    map(rownames_to_column, var="sample_id") %>%
    purrr::set_names(runs) %>%
    bind_rows(.id="fcid") %>%
    mutate(sample_id = str_replace(basename(sample_id), pattern="_S.*$", replacement="")) %>%
    dplyr::select(fcid, sample_id, reads_qualfilt = reads.out),
  by=c("sample_id", "fcid"))

write_csv(logdf, "output/logs/logdf.csv")

Infer sequence variants for each run

This workflow uses the DADA2 algorithm to differentiate real sequences from error using their abundance and co-occurance patters. This relies on the assumption of a random error process where base errors are introduced randomly by either PCR polymerase or sequencing, real sequences will be high quality in the same way, while bad sequences are bad in different individual ways. DADA2 depends on a parameterized error model (the 16(possible bases) × 41(phred score) transition probabilities, for example, p(A→C, 35)), which is estimated from the data. DADA2’s default parameter estimation method is to perform a weighted loess fit to the regularized log of the observed mismatch rates as a function of their quality, separately for each transition type (for example, A→C mismatches are fit separately from A→G mismatches). Following error model learning, all identical sequencing reads are dereplicated into into “Amplicon sequence variants” (ASVs) with a corresponding abundance equal to the number of reads with that unique sequence. The forward and reverse reads are then merged together by aligning the denoised forward reads with the reverse-complement of the corresponding reverse reads, and then constructing the merged “contig” sequences. Following this step, a sequence variant table is constructed and saved as an RDS file.

set.seed(100) # set random seed for reproducability
samdf <- read.csv("sample_data/Sample_info.csv", stringsAsFactors = FALSE)
runs <- unique(samdf$fcid)

# Set parameters
nbases = 1e+08 # Minimum number of total bases to use for error rate - increase if samples are deep sequenced (>1M reads per sample)
randomize = TRUE # Pick samples randomly to learn errors
pool = "pseudo" # Higher accuracy for low abundance at expense of runtime. Set to FALSE for a faster run

dada_out <- vector("list", length=length(runs))
i=1
for (i in 1:length(runs)){
  
  run_data <- samdf %>%
    filter(fcid == runs[i])
  
  #Check if run used twin tags
  filtpath <- paste0("data/", runs[i], "/02_filtered" )
  
  filtFs <- list.files(filtpath, pattern="R1_001.*", full.names = TRUE)
  filtRs <- list.files(filtpath, pattern="R2_001.*", full.names = TRUE)
  
  # Learn error rates from a subset of the samples and reads (rather than running self-consist with full dataset)
  errF <- learnErrors(filtFs, multithread = TRUE, nbases = nbases, randomize = randomize, qualityType = "FastqQuality", verbose=TRUE)
  errR <- learnErrors(filtRs, multithread = TRUE, nbases = nbases, randomize = randomize, qualityType = "FastqQuality", verbose=TRUE)
  
  #write out errors for diagnostics
  write_csv(as.data.frame(errF$trans), paste0("output/logs/", runs[i],"/",runs[i],"_errF_observed_transitions.csv"))
  write_csv(as.data.frame(errF$err_out), paste0("output/logs/", runs[i],"/",runs[i],"_errF_inferred_errors.csv"))
  write_csv(as.data.frame(errR$trans), paste0("output/logs/", runs[i],"/",runs[i],"_errR_observed_transitions.csv"))
  write_csv(as.data.frame(errR$err_out), paste0("output/logs/", runs[i],"/",runs[i],"_errR_inferred_errors.csv"))
  
  ##output error plots to see how well the algorithm modelled the errors in the different runs
  p1 <- plotErrors(errF, nominalQ = TRUE) + ggtitle(paste0(runs[i], " Forward Reads"))
  p2 <- plotErrors(errR, nominalQ = TRUE) + ggtitle(paste0(runs[i], " Reverse Reads"))
  pdf(paste0("output/logs/", runs[i],"/",runs[i],"_errormodel.pdf"), width = 11, height = 8 , paper="a4r")
  plot(p1)
  plot(p2)
  try(dev.off(), silent=TRUE)
  
  #Error inference and merger of reads
  dadaFs <- dada(filtFs, err = errF, multithread = TRUE, pool = pool, verbose = TRUE)
  dadaRs <- dada(filtRs, err = errR, multithread = TRUE, pool = pool, verbose = TRUE)
  
  # merge reads
  mergers <- mergePairs(dadaFs, filtFs, dadaRs, filtRs, verbose = TRUE, minOverlap = 12, trimOverhang = TRUE) 
  mergers <- mergers[sapply(mergers, nrow) > 0]
  bind_rows(mergers, .id="Sample") %>%
    mutate(Sample = str_replace(Sample, pattern="_S.*$", replacement="")) %>%
    write_csv(paste0("output/logs/",runs[i],"/",runs[i], "_mergers.csv"))
  
  #Construct sequence table
  seqtab <- makeSequenceTable(mergers)
  saveRDS(seqtab, paste0("output/rds/", runs[i], "_seqtab.rds"))

  # Track reads
  getN <- function(x) sum(getUniques(x))
  dada_out[[i]] <- cbind(sapply(dadaFs, getN), sapply(dadaRs, getN), sapply(mergers, getN)) %>%
    magrittr::set_colnames(c("dadaFs", "dadaRs", "merged")) %>%
    as.data.frame() %>%
    rownames_to_column("sample_id") %>%
    mutate(sample_id = str_replace(basename(sample_id), pattern="_S.*$", replacement=""))
}

#Update log DF
logdf <- read_csv("output/logs/logdf.csv")

logdf <- logdf  %>% 
  left_join(dada_out %>%
            purrr::set_names(runs) %>%
            bind_rows(.id="fcid") %>%
            mutate(reads_denoised = case_when(
              dadaFs < dadaRs ~ dadaFs,
              dadaFs > dadaRs ~ dadaRs)) %>%
            dplyr::select(fcid, sample_id, reads_denoised, reads_merged = merged),
  by=c("sample_id", "fcid"))

write_csv(logdf, "output/logs/logdf.csv")

Merge Runs, Remove Chimeras and filter

Following denoising and merging of reads, if there were multiple flowcells of data analyse the sequence tables from these will be merged together. Next the sequences are checked for chimeras, and all sequences containing stop codons are removed. The final cleaned sequence table is saved as output/rds/seqtab_final.rds

Note: this will change if you are using a coding marker or not

Coding marker

seqtabs <- list.files("output/rds/", pattern="seqtab.rds", full.names = TRUE)

# If multiple seqtabs present, merge.
if(length(seqtabs) > 1){
  st.all <- mergeSequenceTables(tables=seqtabs)
} else if(length(seqtabs) == 1) {
  st.all <- readRDS(seqtabs)
}

#Remove chimeras
seqtab_nochim <- removeBimeraDenovo(st.all, method="consensus", multithread=TRUE, verbose=TRUE)
message(paste(sum(seqtab_nochim)/sum(st.all),"of the abundance remaining after chimera removal"))

#cut to expected size allowing for some codon indels
seqtab_cut <- seqtab_nochim[,nchar(colnames(seqtab_nochim)) %in% 200:210]
message(paste0("Identified ",
               length(colnames(seqtab_nochim))  - length(colnames(seqtab_cut)),
               " incorrectly sized sequences out of ", length(colnames(seqtab_nochim)) , " input sequences."))
message(paste(sum(seqtab_cut)/sum(seqtab_nochim),"of the abundance remaining after cutting to expected size"))

#Filter sequences containing stop codons
seqs <- DNAStringSet(getSequences(seqtab_cut))
codon_filt <- codon_filter(seqs, genetic_code = 'SGC4') # Default is invertebrate mitochondial code
seqtab_final <- seqtab_cut[,colnames(seqtab_cut) %in% codon_filt]
message(paste0("Identified ",
               length(colnames(seqtab_cut))  - length(colnames(seqtab_final)),
               " sequences containing stop codon out of ", length(colnames(seqtab_cut)) , " input sequences."))
message(paste(sum(seqtab_final)/sum(seqtab_cut),"of the abundance remaining after removing seqs with stop codons "))

saveRDS(seqtab_final, "output/rds/seqtab_final.rds")

# summarise cleanup
cleanup <- st.all %>%
  as.data.frame() %>%
  pivot_longer( everything(),
    names_to = "OTU",
    values_to = "Abundance") %>%
  group_by(OTU) %>%
  summarise(Abundance = sum(Abundance)) %>%
  mutate(length  = nchar(OTU)) %>%
  mutate(type = case_when(
    !OTU %in% getSequences(seqtab_nochim) ~ "Chimera",
    !OTU %in% getSequences(seqtab_cut) ~ "Incorrect size",
    !OTU %in% getSequences(seqtab_final) ~ "Stop codons",
    TRUE ~ "Real"
  )) 
write_csv(cleanup, "output/logs/ASV_cleanup_summary.csv")

#Update log DF
logdf <- read_csv("output/logs/logdf.csv")

logdf <- logdf %>% 
  left_join(as.data.frame(cbind(rowSums(st.all),
                                rowSums(seqtab_nochim),
                                rowSums(seqtab_cut),
                                rowSums(seqtab_final))) %>%
                            rownames_to_column("sample_id") %>%
              mutate(sample_id = str_replace(basename(sample_id), pattern="_S.*$", replacement="")) %>%
              dplyr::select(sample_id, reads_chimerafilt = V2, reads_sizefilt = V3, reads_codonfilt = V4),
  by=c("sample_id"))

write_csv(logdf, "output/logs/logdf.csv")

# Output length distribution plots
gg.abundance <- ggplot(cleanup, aes(x=length, y=Abundance, fill=type))+
              geom_bar(stat="identity") + 
              labs(title = "Abundance of sequences")

gg.unique <- ggplot(cleanup, aes(x=length, fill=type))+
            geom_histogram() + 
            labs(title = "Number of unique sequences")

pdf(paste0("output/logs/seqtab_length_dist.pdf"), width = 11, height = 8 , paper="a4r")
  plot(gg.abundance / gg.unique)
try(dev.off(), silent=TRUE)

Non-coding marker

If you are not using a coding marker, then stop codons should not be checked for

seqtabs <- list.files("output/rds/", pattern="seqtab.rds", full.names = TRUE)

# If multiple seqtabs present, merge.
if(length(seqtabs) > 1){
  st.all <- mergeSequenceTables(tables=seqtabs)
} else if(length(seqtabs) == 1) {
  st.all <- readRDS(seqtabs)
}

#Remove chimeras
seqtab_nochim <- removeBimeraDenovo(st.all, method="consensus", multithread=TRUE, verbose=TRUE)
message(paste(sum(seqtab_nochim)/sum(st.all),"of the abundance remaining after chimera removal"))

#cut to expected size allowing for some codon indels
seqtab_final <- seqtab_nochim[,nchar(colnames(seqtab_nochim)) %in% 200:210]
message(paste0("Identified ",
               length(colnames(seqtab_nochim))  - length(colnames(seqtab_cut)),
               " incorrectly sized sequences out of ", length(colnames(seqtab_nochim)) , " input sequences."))
message(paste(sum(seqtab_final)/sum(seqtab_nochim),"of the abundance remaining after cutting to expected size"))

saveRDS(seqtab_final, "output/rds/seqtab_final.rds")

# summarise cleanup
cleanup <- st.all %>%
  as.data.frame() %>%
  pivot_longer( everything(),
    names_to = "OTU",
    values_to = "Abundance") %>%
  group_by(OTU) %>%
  summarise(Abundance = sum(Abundance)) %>%
  mutate(length  = nchar(OTU)) %>%
  mutate(type = case_when(
    !OTU %in% getSequences(seqtab_nochim) ~ "Chimera",
    !OTU %in% getSequences(seqtab_final) ~ "Incorrect size",
    TRUE ~ "Real"
  )) 
write_csv(cleanup, "output/logs/chimera_summary.csv")

#Update log DF
logdf <- read_csv("output/logs/logdf.csv")

logdf <- logdf %>% 
  left_join(as.data.frame(cbind(rowSums(st.all),
                                rowSums(seqtab_nochim),
                                rowSums(seqtab_final))) %>%
                            rownames_to_column("sample_id") %>%
              mutate(sample_id = str_replace(basename(sample_id), pattern="_S.*$", replacement="")) %>%
              dplyr::select(sample_id, reads_chimerafilt = V2, reads_sizefilt = V3),
  by=c("sample_id"))

write_csv(logdf, "output/logs/logdf.csv")

# Output length distribution plots
gg.abundance <- ggplot(cleanup, aes(x=length, y=Abundance, fill=type))+
              geom_bar(stat="identity") + 
              labs(title = "Abundance of sequences")

gg.unique <- ggplot(cleanup, aes(x=length, fill=type))+
            geom_histogram() + 
            labs(title = "Number of unique sequences")

pdf(paste0("output/logs/seqtab_length_dist.pdf"), width = 11, height = 8 , paper="a4r")
  plot(gg.abundance / gg.unique)
try(dev.off(), silent=TRUE)

Assign taxonomy

Now that we have a cleaned table of sequences and their abundances across samples, we need to assign taxonomy to the sequences in order to identify taxa. The default approach is currently to use IDTAXA to assign heirarchial taxonomy, followed by a BLAST search for increased species level assignment. However there are a number of alternative classifiers you can use to do this, a few of which are represented in the tabs below.

IDTAXA + BLAST

We will use the IDTAXA algorithm of Murali et al 2018 to assign taxonomy to the ASVs. IDTAXA requires a pre-trained classifier, which can be found in the reference folder, alternatively see the taxreturn r package if you wish to curate a reference database and train a new classifier.

To increase classification to species level, we will also incorporate a BLAST search. However as top hit assignment methods such as BLAST do not take the context of other sequences into account, to reduce the risk of over-classification we will only assign an ASV to species rank if the BLAST search agrees with IDTAXA at the Genus rank.

seqtab_final <- readRDS("output/rds/seqtab_final.rds")
# NOTE: these ranks may differ for different training sets. Check your training set to avoid an error
ranks <-  c("Root", "Kingdom", "Phylum","Class", "Order", "Family", "Genus","Species") 

#Classify using IDTAXA
trainingSet <- readRDS("reference/idtaxa_bftrimmed.rds")
dna <- DNAStringSet(getSequences(seqtab_final)) # Create a DNAStringSet from the ASVs
ids <- IdTaxa(dna, trainingSet, processors=1, threshold = 60, verbose=TRUE, strand = "top") 
saveRDS(ids, "output/rds/idtaxa.rds")

# Output plot of ids
pdf(paste0("output/logs/idtaxa.pdf"), width = 11, height = 8 , paper="a4r")
  plot(ids)
try(dev.off(), silent=TRUE)

#Convert the output object of class "Taxa" to a matrix analogous to the output from assignTaxonomy
tax <- t(sapply(ids, function(x) {
  taxa <- paste0(x$taxon,"_", x$confidence)
  taxa[startsWith(taxa, "unclassified_")] <- NA
  taxa
})) %>%
  purrr::map(unlist) %>%
  stri_list2matrix(byrow=TRUE, fill=NA) %>%
  magrittr::set_colnames(ranks) %>%
  as.data.frame() %>%
magrittr::set_rownames(getSequences(seqtab_final)) %T>%
  write.csv("output/logs/idtaxa_results.csv") %>%  #Write out logfile with confidence levels
  mutate_all(str_replace,pattern="(?:.(?!_))+$", replacement="") %>%
  magrittr::set_rownames(getSequences(seqtab_final)) 

# Top hit with BLAST
seqs <- taxreturn::char2DNAbin(colnames(seqtab_final))
names(seqs) <- colnames(seqtab_final) 

blast_spp <- blast_assign_species(query=seqs,db="reference/insecta_hierarchial_bftrimmed.fa.gz", identity=97, coverage=95, evalue=1e06, max_target_seqs=5, max_hsp=5, ranks=ranks, delim=";") %>%
  dplyr::rename(blast_genus = Genus, blast_spp = Species) %>%
  dplyr::filter(!is.na(blast_spp))

#Join together
tax_blast <- tax %>%
  as_tibble(rownames = "OTU") %>%
  left_join(blast_spp , by="OTU") %>%
  dplyr::mutate(Species = case_when(
    is.na(Species) & Genus == blast_genus ~ blast_spp,
    !is.na(Species) ~ Species
  )) %>%
  dplyr::select(OTU, ranks) %>%
  column_to_rownames("OTU") %>%
  seqateurs::na_to_unclassified() %>% #Propagate high order ranks to unassigned ASVs
  as.matrix()

# Write taxonomy table to disk
saveRDS(tax_blast, "output/rds/tax.rds") 

IDTAXA + Exact Matching

As an alternative to using BLAST, we can use exact 100% matches only to assign additional sequences to the species rank. This is particularly useful for bacterial metabarcoding as 100% matches have been shown to be the only valid method of assigning species to short bacterial 16s sequences.

seqtab_final <- readRDS("output/rds/seqtab_final.rds")
# NOTE: these ranks may differ for different training sets. Check your training set to avoid an error
ranks <-  c("Root", "Kingdom", "Phylum","Class", "Order", "Family", "Genus","Species") 

#Classify using IDTAXA
trainingSet <- readRDS("reference/idtaxa_bftrimmed.rds")
dna <- DNAStringSet(getSequences(seqtab_final)) # Create a DNAStringSet from the ASVs
ids <- IdTaxa(dna, trainingSet, processors=1, threshold = 60, verbose=TRUE, strand = "top") 
saveRDS(ids, "output/rds/idtaxa.rds")

# Output plot of ids
pdf(paste0("output/logs/idtaxa.pdf"), width = 11, height = 8 , paper="a4r")
  plot(ids)
try(dev.off(), silent=TRUE)

#Convert the output object of class "Taxa" to a matrix analogous to the output from assignTaxonomy
tax <- t(sapply(ids, function(x) {
  taxa <- paste0(x$taxon,"_", x$confidence)
  taxa[startsWith(taxa, "unclassified_")] <- NA
  taxa
})) %>%
  purrr::map(unlist) %>%
  stri_list2matrix(byrow=TRUE, fill=NA) %>%
  magrittr::set_colnames(ranks) %>%
  as.data.frame() %>%
magrittr::set_rownames(getSequences(seqtab_final)) %T>%
  write.csv("output/logs/idtaxa_results.csv") %>%  #Write out logfile with confidence levels
  mutate_all(str_replace,pattern="(?:.(?!_))+$", replacement="") %>%
  magrittr::set_rownames(getSequences(seqtab_final)) 

#Further assign to species rank using exact matching
exact <- assignSpecies(seqtab_final, "reference/insecta_binomial_bftrimmed.fa.gz", allowMultiple = TRUE, tryRC = TRUE, verbose = FALSE) %>%
    as_tibble(rownames = "OTU") %>%
    filter(!is.na(Species)) %>%
    dplyr::mutate(binomial = paste0(Genus," ",Species)) %>%
     dplyr::rename(exact_genus = Genus, exact_species = Species)

#Merge together
tax_exact <- tax %>%
  as_tibble(rownames = "OTU") %>%
  left_join(exact, by="OTU") %>%
  dplyr::mutate(Species = case_when(
    is.na(Species) & Genus == exact_genus ~ binomial,
    !is.na(Species) ~ Species
  )) %>%
dplyr::select(OTU, ranks) %>%
  column_to_rownames("OTU") %>%
  seqateurs::na_to_unclassified() %>% #Propagate high order ranks to unassigned ASVs
  as.matrix()

# Write taxonomy table to disk
saveRDS(tax_exact, "output/rds/tax.rds") 

IDTAXA

Alternatively, we can use the IDTAXA classifier by itself with no supplementary assignment

seqtab_final <- readRDS("output/rds/seqtab_final.rds")
# NOTE: these ranks may differ for different training sets. Check your training set to avoid an error
ranks <-  c("Root","Kingdom", "Phylum","Class", "Order", "Family", "Genus","Species") 

#Classify using IDTAXA
trainingSet <- readRDS("reference/idtaxa_bftrimmed.rds")
dna <- DNAStringSet(getSequences(seqtab_final)) # Create a DNAStringSet from the ASVs
ids <- IdTaxa(dna, trainingSet, processors=1, threshold = 60, verbose=TRUE, strand = "top") 
saveRDS(ids, "output/rds/idtaxa.rds")

# Output plot of ids
pdf(paste0("output/logs/idtaxa.pdf"), width = 11, height = 8 , paper="a4r")
  plot(ids)
try(dev.off(), silent=TRUE)

#Convert the output object of class "Taxa" to a matrix analogous to the output from assignTaxonomy
tax <- t(sapply(ids, function(x) {
  taxa <- paste0(x$taxon,"_", x$confidence)
  taxa[startsWith(taxa, "unclassified_")] <- NA
  taxa
})) %>%
  purrr::map(unlist) %>%
  stri_list2matrix(byrow=TRUE, fill=NA) %>%
  magrittr::set_colnames(ranks) %>%
  as.data.frame() %>%
  magrittr::set_rownames(getSequences(seqtab_final)) %T>%
  write.csv("output/logs/idtaxa_results.csv") %>%  #Write out logfile with confidence levels
  mutate_all(str_replace,pattern="(?:.(?!_))+$", replacement="") %>%
  magrittr::set_rownames(getSequences(seqtab_final)) %>%
  seqateurs::na_to_unclassified() %>% #Propagate high order ranks to unassigned ASVs
  as.matrix()

# Write taxonomy table to disk
saveRDS(tax, "output/rds/tax.rds") 

RDP + Exact Matching

Alternatively, the RDP classifier can be used to assign heirarchial taxonomy to the ASVs using the convenient assigntaxonomy wrapper function within DADA2

seqtab_final <- readRDS("output/rds/seqtab_final.rds")

# Assign Kingdom:Genus taxonomy using RDP classifier
tax <- assignTaxonomy(seqtab_final, "reference/insecta_hierarchial_bftrimmed.fa.gz", multithread=FALSE, minBoot=60, outputBootstraps=FALSE)
colnames(tax) <- c("Root","Kingdom", "Phylum","Class", "Order", "Family", "Genus", "Species") 
saveRDS(tax, "output/rds/rdp.rds")

##add species to taxtable using exact matching
tax_plus <- addSpecies(tax, "reference/insecta_binomial_bftrimmed.fa.gz", allowMultiple=TRUE)

tax_plus <- na_to_unclassified(tax_plus)

# Write taxonomy table to disk
saveRDS(tax_plus, "output/rds/tax.rds") 

Optional - Top hit identity distribution

This is an optional step to produce a Top Hit Identity Distribution (THID) plot. This displays the taxonomy assigned to each ASV compared to its genetic distance from the most similar sequence in the reference database. This provides a good overview of how well the taxonomic assignment algorithm has performed, as well as the representation of the target taxa within the reference database. Press the CODE button to the lower right to display the code for this optional step.

seqtab_final <- readRDS("output/rds/seqtab_final.rds")
tax <- readRDS("output/rds/tax.rds")

seqs <- taxreturn::char2DNAbin(colnames(seqtab_final))
names(seqs) <- colnames(seqtab_final)

out <- blast_top_hit(query=seqs, db="reference/insecta_hierarchial_bftrimmed.fa.gz", identity=60, coverage=80)

joint <- out %>% 
  dplyr::select(OTU = qseqid, acc, blastspp = Species, pident, length, evalue, qcovs) %>%
  left_join(tax %>% 
              seqateurs::unclassified_to_na(rownames=FALSE) %>%
              mutate(lowest = seqateurs::lowest_classified(.)), by="OTU")

#Write out comparison between BLAST and Heiarchial assignment
write_csv(joint, "output/logs/tax_assignment_comparison.csv")

gg.tophit <- joint %>%
  dplyr::select(pident, rank = lowest) %>%
  mutate(rank = factor(rank, levels = c("Root","Kingdom","Phylum","Class","Order","Family","Genus","Species"))) %>%
  ggplot(aes(x=pident, fill=rank))+ 
  geom_histogram(colour="black", binwidth = 1, position = "stack") + 
  labs(title = "Top hit identity distribution",
       x = "BLAST top hit % identity",
       y = "OTUs") + 
  scale_x_continuous(breaks=seq(60,100,2)) +
  scale_fill_brewer(name = "Taxonomic \nAssignment", palette = "Spectral")

gg.tophit

pdf(paste0("output/logs/top_hit_tax_assignment.pdf"), width = 11, height = 8 , paper="a4r")
  gg.tophit
try(dev.off(), silent=TRUE)

Make phylogenetic tree

In addition to taxonomic assignment, we will create a phylogenetic tree from the identified sequences to allow interpetation within a phylognetic context.

seqtab_final <- readRDS("output/rds/seqtab_final.rds")

seqs <- getSequences(seqtab_final)
names(seqs) <- seqs # This propagates to the tip labels of the tree
alignment <- AlignSeqs(DNAStringSet(seqs), anchor=NA)

library(phangorn)
phang.align <- phyDat(as(alignment, "matrix"), type="DNA")
dm <- dist.ml(phang.align)

#Fit NJ tree
treeNJ <- NJ(dm) # Note, tip order != sequence order
fit <- pml(treeNJ, data=phang.align)

#Fit ML tree
fitGTR <- update(fit, k=4, inv=0.2)
fitGTR <- optim.pml(fitGTR, model="GTR", optInv=TRUE, optGamma=TRUE,
                      rearrangement = "stochastic", control = pml.control(trace = 0))

# Write phytree to disk
saveRDS(fitGTR, "output/rds/phytree.rds") 

#Output newick tree
write.tree(fitGTR$tree, file="output/results/unfiltered/tree_unfiltered.nwk")

Make Phyloseq object & Output final csvs

Finally, we will merge the sequence table, taxonomy table, phylogenetic tree, and sample data into a single phyloseq object, filter low abundance taxa, and output summary CSV files and fasta files of the identified taxa

seqtab_final <- readRDS("output/rds/seqtab_final.rds")

#Extract start of sequence names
rownames(seqtab_final) <- str_replace(rownames(seqtab_final), pattern="_S[0-9].*$", replacement="")

tax <- readRDS("output/rds/tax.rds") 
phy <- readRDS("output/rds/phytree.rds")$tree
seqs <- DNAStringSet(colnames(seqtab_final))
names(seqs) <- seqs

#Load sample information
samdf <- read.csv("sample_data/Sample_info.csv", header=TRUE) %>%
  filter(!duplicated(sample_id)) %>%
  magrittr::set_rownames(.$sample_id) 

# Create phyloseq object
ps <- phyloseq(tax_table(tax),
               sample_data(samdf),
               otu_table(seqtab_final, taxa_are_rows = FALSE),
               phy_tree(phy),
               refseq(seqs))

if(nrow(seqtab_final) > nrow(sample_data(ps))){
  message("Warning: the following samples were not included in phyloseq object, check sample names match the sample metadata")
  message(rownames(seqtab_final)[!rownames(seqtab_final) %in% sample_names(ps)])
}

saveRDS(ps, "output/rds/ps.rds") 

#Export raw csv
speedyseq::psmelt(ps) %>%
  filter(Abundance > 0) %>%
  dplyr::select(-Sample) %>%
  write_csv("output/results/unfiltered/raw_combined.csv")
  
#Summary export
seqateurs::summarise_taxa(ps, "Species", "sample_id") %>%
  spread(key="sample_id", value="totalRA") %>%
  write.csv(file = "output/results/unfiltered/spp_sum_unfiltered.csv")

seqateurs::summarise_taxa(ps, "Genus", "sample_id") %>%
  spread(key="sample_id", value="totalRA") %>%
  write.csv(file = "output/results/unfiltered/gen_sum_unfiltered.csv")

#Output fasta of all ASV's
seqateurs::ps_to_fasta(ps, out.file ="output/results/unfiltered/asvs_unfiltered.fasta", seqnames = "Species")

Taxon & Sample filtering

Here we will remove all taxa that were not classified to Arthropoda, as these most likely represent residual erroneous sequences. This will be followed by removing all samples which are under a minimum read threshold. In this case, 1000.

#Set a threshold for minimum reads per sample
threshold <- 1000

ps0 <- ps %>%
  subset_taxa(
    Phylum == "Arthropoda"
  ) %>%
  filter_taxa(function(x) mean(x) > 0, TRUE) %>%
  prune_samples(sample_sums(.) >0, .) 

#Create rarefaction curve

rare <- otu_table(ps0) %>%
  as("matrix") %>%
  rarecurve(step=max(sample_sums(ps0))/100) %>%
  purrr::map(function(x){
    b <- as.data.frame(x)
    b <- data.frame(OTU = b[,1], count = rownames(b))
    b$count <- as.numeric(gsub("N", "",  b$count))
    return(b)
  }) %>%
  purrr::set_names(sample_names(ps0)) %>%
  bind_rows(.id="sample_id")

gg.rare <- ggplot(data = rare)+
  geom_line(aes(x = count, y = OTU, group=sample_id), alpha=0.5)+
  geom_point(data = rare %>% 
               group_by(sample_id) %>% 
               top_n(1, count),
             aes(x = count, y = OTU, colour=(count > threshold))) +
  geom_label(data = rare %>% 
               group_by(sample_id) %>% 
               top_n(1, count),
             aes(x = count, y = OTU,label=sample_id, colour=(count > threshold)),
             hjust=-0.05)+
  scale_x_continuous(labels =  scales::scientific_format()) +
  geom_vline(xintercept=threshold, linetype="dashed") +
  labs(colour = "Sample kept?") +
  xlab("Sequence reads") +
  ylab("Observed ASV's")

gg.rare

#Write out figure
pdf(file="fig/rarefaction.pdf", width = 11, height = 8 , paper="a4r")
  plot(gg.rare)
try(dev.off(), silent=TRUE)

#Remove all samples under the minimum read threshold 
ps1 <- ps0 %>%
  prune_samples(sample_sums(.)>=threshold, .) %>% 
  filter_taxa(function(x) mean(x) > 0, TRUE) #Drop missing taxa from table

#Message how many were removed
message(nsamples(ps) - nsamples(ps1), " Samples and ", ntaxa(ps) - ntaxa(ps1), " ASVs dropped")

# Export summary of filtered results
seqateurs::summarise_taxa(ps1, "Species", "sample_id") %>%
  spread(key="sample_id", value="totalRA") %>%
  write.csv(file = "output/results/filtered/spp_sum_filtered.csv")

seqateurs::summarise_taxa(ps1, "Genus", "sample_id") %>%
  spread(key="sample_id", value="totalRA") %>%
  write.csv(file = "output/results/filtered/gen_sum_filtered.csv")

#Output fasta of all ASV's
seqateurs::ps_to_fasta(ps1, "output/results/filtered/asvs_filtered.fasta", seqnames="Species")

#Output newick tree
write.tree(phy_tree(ps1), file="output/results/filtered/tree_filtered.nwk")

# output filtered phyloseq object
saveRDS(ps1, "output/rds/ps_filtered.rds") 

Output final CSVs

We will output the final 3 filtered CSVs which will be uploaded to the imappests staging point database

• seqtab.csv
• taxtab.csv
• samdf.csv

seqtab <- otu_table(ps1) %>%
  as("matrix") %>%
  as_tibble(rownames = "sample_id")

taxtab <- tax_table(ps1) %>%
  as("matrix") %>%
  as_tibble(rownames = "OTU") %>%
  unclassified_to_na(rownames = FALSE)

#Check taxonomy table outputs
if(!all(colnames(taxtab) == c("OTU", "Root", "Kingdom", "Phylum", "Class", "Order", "Family", "Genus", "Species"))){
  message("Warning: Taxonomy table columns do not meet expectations for the staging database \n
          Database requires the columns: OTU, Root, Kingdom, Phylum, Class, Order, Family, Genus, Species ")
}

if(any(str_detect(taxtab$Species, "/"))){
  message("Warning: Taxonomy table contains taxa with clashes at the species level, these should be corrected before upload:")
  clashes <- taxtab$Species[str_detect(taxtab$Species, "/")]
  print(clashes[!is.na(clashes)])
}

samdf <- sample_data(ps1) %>%
  as("matrix") %>%
  as_tibble()

# Write out
write_csv(seqtab, "output/results/final/seqtab.csv")
write_csv(taxtab, "output/results/final/taxtab.csv")
write_csv(samdf, "output/results/final/samdf.csv")

#Write out combined
speedyseq::psmelt(ps1) %>%
  filter(Abundance > 0) %>%
  dplyr::select(-Sample) %>%
  write_csv("output/results/filtered/combined.csv")

Optional - Check presence of taxa in Australia

This is an optional step to run an automated search against the Australian Faunal Directory ato see if the the detected species have been recorded in Australia before.

Press the CODE button to the lower right to display the code for this optional step.

ps1 <- readRDS("output/rds/ps_filtered.rds")

# Check presence on AFD
afd_check <- ps1 %>%
  speedyseq::psmelt() %>%
  dplyr::group_by(Family, Genus, Species) %>%
  summarise(metabarcoding_reads = sum(Abundance)) %>%
  filter(!str_detect(Species, "__")) %>%
  mutate(Species = Species %>% str_replace_all("_", " ")) %>%
  mutate(
    Family_present = afdscraper::check_afd_presence(Family),
    Genus_present = afdscraper::check_afd_presence(Genus),
    Species_present = afdscraper::check_afd_presence(Species)
  ) %>%
  dplyr::select(Family, Family_present, Genus, Genus_present, 
                Species, Species_present, metabarcoding_reads)
    
write_csv(afd_check, "output/results/final/afd_check.csv")

# Check presence on ALA
# First we need to set some data quality filters for ALA
# To view available filters, run: find_field_values("basis_of_record")
ala_quality_filter <- galah::select_filters(
      basisOfRecord = c("PreservedSpecimen", "LivingSpecimen",
                      "MaterialSample", "NomenclaturalChecklist"),
      profile = "ALA")

ala_quality_filter <- galah::select_filters(
      profile = "ALA")

ala_check <- ps1 %>%
  speedyseq::psmelt() %>%
  dplyr::group_by(Family, Genus, Species) %>%
  summarise(metabarcoding_reads = sum(Abundance)) %>%
  filter(!str_detect(Species, "__")) %>%
  mutate(Species = Species %>% str_replace_all("_", " ")) %>%
  mutate(
    species_present = purrr::map(Species, function(x){
    # first check name
    query <- select_taxa(x) %>% 
      as_tibble()%>%
      dplyr::filter(across(any_of("match_type"), ~!.x == "higherMatch"))
    # Then get occurance counts
    if(!is.null(query$scientific_name)){
      ala_occur <- ala_counts(taxa=query, filters=ala_quality_filter)
      return(data.frame(Species_present = ifelse(ala_occur > 0, TRUE, FALSE), ALA_counts = ala_occur))
    } else {
      return(data.frame(Species_present = FALSE, ALA_counts = 0))
    }
    })) %>%
  unnest(species_present) %>%
  dplyr::select(Family, Genus, Species, Species_present, ALA_counts, metabarcoding_reads)

write_csv(ala_check, "output/results/final/ala_check.csv")

Output fate of reads through pipeline

#Fraction of reads assigned to each taxonomic rank
sum_reads <- speedyseq::psmelt(ps) %>%
  gather("Rank","Name", rank_names(ps)) %>%
  group_by(Rank, sample_id) %>% 
  mutate(Name = replace(Name, str_detect(Name, "__"),NA)) %>% # This line turns the "__" we added to lower ranks back to NA's
  summarise(Reads_classified = sum(Abundance * !is.na(Name))) %>% 
  pivot_wider(names_from = "Rank",
              values_from = "Reads_classified") %>%
  dplyr::select(sample_id, rank_names(ps)) %>%
  dplyr::rename_at(rank_names(ps), ~paste0("reads_", .))

#Update log DF
logdf <- read_csv("output/logs/logdf.csv")

logdf <- logdf %>% 
  left_join(sum_reads,
  by=c("sample_id"))

write_csv(logdf, "output/logs/logdf.csv")


gg.all_reads <- logdf %>%
  dplyr::select(sample_id, fcid, starts_with("reads_"), -reads_total, -reads_pf) %>%
  pivot_longer(starts_with("reads_"),
               names_to = "type",
               values_to = "value") %>%
  group_by(fcid, type) %>% 
  summarise(value = sum(value, na.rm = TRUE)) %>%
  bind_rows(logdf %>%
  dplyr::select(fcid, reads_total, reads_pf) %>%
    distinct()%>%
  pivot_longer(starts_with("reads_"),
               names_to = "type",
               values_to = "value")
  ) %>%
  mutate(type = str_remove(type, "reads_")) %>%
  mutate(type = factor(type, levels = c(
  "total", "pf", "demulti",
  "trimmed", "qualfilt",
  "denoised", "merged",
  "chimerafilt", "sizefilt", "codonfilt",
  "Root", "Kingdom", "Phylum",
  "Class", "Order","Family",
  "Genus", "Species"))) %>% 
  ggplot(aes(x=type, y=value, fill=fcid)) +
  geom_bar(stat="identity") +
  facet_wrap(~fcid) +
  theme(axis.text.x = element_text(angle=90, hjust = 1, vjust=0.5)) 

gg.all_reads

pdf(paste0("output/logs/read_tracker_all.pdf"), width = 11, height = 8 , paper="a4r")
  gg.all_reads
try(dev.off(), silent=TRUE)
  
# Read tracker per sample
  
gg.separate_reads <- logdf %>%
  dplyr::select(sample_id, fcid, starts_with("reads_"), -reads_total, -reads_pf) %>%
  pivot_longer(starts_with("reads_"),
               names_to = "type",
               values_to = "value") %>%
  mutate(type = str_remove(type, "reads_")) %>%
  mutate(type = factor(type, levels = c(
  "total", "pf", "demulti",
  "trimmed", "qualfilt",
  "denoised", "merged",
  "chimerafilt", "sizefilt", "codonfilt",
  "Root", "Kingdom", "Phylum",
  "Class", "Order","Family",
  "Genus", "Species"))) %>% 
  ggplot(aes(x=type, y=value, fill=fcid)) +
  geom_bar(stat="identity") +
  facet_wrap(~sample_id) +
  theme(axis.text.x = element_text(angle=90, hjust = 1, vjust=0.5)) 

gg.separate_reads

pdf(paste0("output/logs/read_tracker_separate.pdf"), width = 11, height = 8 , paper="a4r")
  gg.separate_reads
try(dev.off(), silent=TRUE)

Further analysis

From here, the dataset can be further analysed in software of your choice. I suggest the use of phyloseq