end of reads (where they occur most often) by calculating average
base quality across a sliding window of bases starting at either end
of the read. For the most comprehensive view of FASTQ files
before proceeding with quality control, FastQC (or NanoQC)
can be run a second time to quantify the effect of the read trimming
tools by comparing results of the trimmed FASTQ with the original
FASTQ.
3.5.5 Detecting
and Removing
Contaminating Reads
One critical but often overlooked step is ensuring the absence of
contaminating reads in a sample’s FASTQ file. Contamination of a
FASTQ can occur in two primary ways: (a) contaminated reads can
be derived from errors during demultiplexing (e.g., the phenomenon of index hopping in Illumina instruments [36] or misassignment of reads to barcode bins in Nanopore demultiplexing [37]) or
(b) contaminated reads can be derived from unintended organisms
inadvertently introduced into the extraction process or contaminating DNA entering gDNA vials. Regardless of the source, bioinformatics best practices are for FASTQ files treated as metagenomic
samples until demonstrated to be isolated samples. Tools that were
originally developed for reads-based metagenomic sample characterization can also be applied to demonstrate that a sequenced
sample is an isolate. For example, metagenomic identification
tools such as Kraken2 [38] assign a taxonomic rank to each read
in a FASTQ and bin these reads into individual taxa. Obtaining a
single bin would strongly suggest the sequenced sample was an
isolate rather than a sample composed of multiple different organisms. If more than a single bin is recovered after such an analysis,
one could either (a) proceed using only the bin matching the
intended sample or (b) resequence the sample beginning from the
wet lab protocol just prior to the likely introduction of the contaminant. However, this approach can be problematic in two cases.
First, if only closely related samples were multiplexed on the flow
cell, these algorithms would be unable to differentiate crosscontamination. For this reason, the best practice is for multiplexed
runs on an instrument to be composed of different species when
possible. Second, novel genomes or genomes without a closely
related publicly available counterpart in the Kraken2 database may
have their reads binned into more than one bin when they should
only be part of a single bin. An alternative approach not dependent
on the genomic databases of metagenomic identification programs
is to simply map the reads onto a genome derived from a verified
isolate of a closely related strain, typically performed using read
mapping software such as bowtie2 [39] or bwa-mem [40]. In this
case, detecting a substantial number of reads failing to map to the
closely related genome would indicate the presence of contaminating reads.
414
Marco A. Riojas et al.
base quality across a sliding window of bases starting at either end
of the read. For the most comprehensive view of FASTQ files
before proceeding with quality control, FastQC (or NanoQC)
can be run a second time to quantify the effect of the read trimming
tools by comparing results of the trimmed FASTQ with the original
FASTQ.
3.5.5 Detecting
and Removing
Contaminating Reads
One critical but often overlooked step is ensuring the absence of
contaminating reads in a sample’s FASTQ file. Contamination of a
FASTQ can occur in two primary ways: (a) contaminated reads can
be derived from errors during demultiplexing (e.g., the phenomenon of index hopping in Illumina instruments [36] or misassignment of reads to barcode bins in Nanopore demultiplexing [37]) or
(b) contaminated reads can be derived from unintended organisms
inadvertently introduced into the extraction process or contaminating DNA entering gDNA vials. Regardless of the source, bioinformatics best practices are for FASTQ files treated as metagenomic
samples until demonstrated to be isolated samples. Tools that were
originally developed for reads-based metagenomic sample characterization can also be applied to demonstrate that a sequenced
sample is an isolate. For example, metagenomic identification
tools such as Kraken2 [38] assign a taxonomic rank to each read
in a FASTQ and bin these reads into individual taxa. Obtaining a
single bin would strongly suggest the sequenced sample was an
isolate rather than a sample composed of multiple different organisms. If more than a single bin is recovered after such an analysis,
one could either (a) proceed using only the bin matching the
intended sample or (b) resequence the sample beginning from the
wet lab protocol just prior to the likely introduction of the contaminant. However, this approach can be problematic in two cases.
First, if only closely related samples were multiplexed on the flow
cell, these algorithms would be unable to differentiate crosscontamination. For this reason, the best practice is for multiplexed
runs on an instrument to be composed of different species when
possible. Second, novel genomes or genomes without a closely
related publicly available counterpart in the Kraken2 database may
have their reads binned into more than one bin when they should
only be part of a single bin. An alternative approach not dependent
on the genomic databases of metagenomic identification programs
is to simply map the reads onto a genome derived from a verified
isolate of a closely related strain, typically performed using read
mapping software such as bowtie2 [39] or bwa-mem [40]. In this
case, detecting a substantial number of reads failing to map to the
closely related genome would indicate the presence of contaminating reads.
414
Marco A. Riojas et al.