Read Quality Assessment

  • Read quality assessment is an essential step in bioinformatics that evaluates the accuracy, reliability, and overall suitability of sequencing reads for downstream analysis. Sequencing technologies generate large numbers of DNA or RNA reads, but not every read has the same quality. Some reads may contain incorrect nucleotide calls, adapter contamination, ambiguous bases, unusual nucleotide composition, or other technical artifacts. These problems can affect sequence alignment, genome assembly, variant detection, gene expression analysis, and metagenomic classification. Read quality assessment helps identify such problems before further processing and allows researchers to determine whether sequencing data are suitable for their intended biological investigation.
  • Sequencing read quality depends on several factors, including the sequencing platform, library preparation method, sample quality, sequencing chemistry, instrument performance, and basecalling software. In short-read sequencing, quality may decline toward the ends of reads as sequencing cycles progress. In long-read sequencing, quality can vary with the length of the molecule, signal characteristics, basecalling model, and sequencing chemistry. Other problems, such as contamination, PCR amplification bias, low library complexity, and uneven coverage, may affect the dataset even when individual base-call quality scores are high.
  • Read quality assessment generally examines both individual reads and the sequencing dataset as a whole. Individual-read assessment evaluates the nucleotide sequence, base-call quality, read length, ambiguous bases, and other characteristics of each read. Dataset-level assessment examines distributions and patterns across all reads, including per-base quality, GC content, sequence duplication, adapter contamination, and read-length variation. Combining these measurements provides a more complete picture of sequencing performance than relying on a single average quality score.
  • One of the most important measurements is the Phred quality score, which estimates the probability that a nucleotide has been called incorrectly. It is calculated using the equation:

[
Q = -10\log_{10}(P)
]

  • Here, (Q) represents the Phred quality score and (P) represents the estimated probability of an incorrect base call. A Q10 score corresponds to an estimated error probability of 10%, Q20 to 1%, Q30 to 0.1%, and Q40 to 0.01%. Higher quality scores indicate greater confidence in the base call. In FASTQ files, these scores are usually represented by ASCII characters, commonly using Phred+33 encoding in modern sequencing workflows. Quality assessment software converts these characters into numerical scores for analysis.
  • Per-base sequence quality measures the distribution of quality scores at each position across all reads. It is particularly useful for identifying sequencing-cycle effects and positional quality deterioration. In a typical FastQC report, the results are displayed as box plots showing the median, interquartile range, and other quality-score distributions for each base position. A gradual decline toward the end of reads may indicate that later sequencing cycles have lower base-call confidence. This pattern is commonly observed in some short-read sequencing datasets, although its severity varies with the platform, chemistry, and read length.
  • Per-sequence quality evaluates the distribution of average quality scores across individual reads. It helps identify whether a subset of reads has consistently poor quality. A dataset may have a high overall average quality while still containing a considerable number of low-quality reads. Examining the distribution of per-read quality scores can reveal such problems and help determine whether filtering is necessary. However, a low-quality read is not automatically biologically useless, and the decision to remove it should depend on the intended analysis.
  • Read length distribution is another important quality metric. It shows the lengths of sequencing reads across a dataset and can reveal unexpected short reads, inconsistent library preparation, excessive trimming, or platform-specific effects. In fixed-length short-read sequencing, most reads are expected to have similar lengths before preprocessing. After quality trimming, read lengths may vary because low-quality terminal bases or adapter sequences have been removed. In long-read sequencing, read lengths are naturally more variable, so the distribution should be interpreted in relation to the platform and library preparation method.
  • GC content represents the proportion of guanine and cytosine nucleotides in a sequence. It is commonly calculated as:

[
GC\ content = \frac{G+C}{A+T+G+C}\times 100
]

  • Here, A, T, G, and C represent the counts of the four standard DNA nucleotides. RNA sequence analysis generally uses uracil (U) instead of thymine (T). The GC content distribution across sequencing reads can help identify unusual sequence composition, contamination, or amplification bias. For example, a sample containing organisms with very different genomic GC contents may naturally show a broad or multimodal distribution. Therefore, an unusual GC profile should be investigated in its biological context rather than automatically classified as a sequencing failure.
  • Per-base sequence content examines the relative frequency of each nucleotide at every position in the reads. In an unbiased random library, nucleotide proportions may be expected to remain reasonably stable across positions, although biological sequence composition and library preparation can produce legitimate deviations. Strong positional biases may indicate adapter contamination, primer-related sequences, random-priming bias, restriction-site enrichment, or other technical effects. In RNA sequencing and targeted sequencing, nucleotide composition may naturally differ from the pattern expected in random genomic DNA libraries.
  • Sequence duplication measures how frequently identical or highly similar reads occur in a dataset. High duplication may indicate PCR amplification, low library complexity, over-sequencing of a limited number of molecules, or the presence of highly abundant biological sequences. In RNA sequencing, highly expressed transcripts can naturally generate many reads from the same regions. In targeted sequencing, enrichment can also produce high duplication levels. Consequently, duplication should not automatically be interpreted as a problem or removed without considering the experiment. Unique molecular identifiers (UMIs) can help distinguish independent original molecules from PCR duplicates in workflows that use them.
  • Adapter contamination occurs when sequencing reads contain artificial adapter sequences introduced during library preparation. This can happen when DNA fragments are shorter than the sequencing read length, allowing the sequencer to read through the insert and into the adapter. Adapter contamination can interfere with alignment, assembly, and other downstream analyses. Quality-control tools can identify overrepresented adapter sequences or compare reads against known adapter collections. Adapter removal is generally performed using tools such as Cutadapt, fastp, and Trimmomatic.
  • Overrepresented sequences are nucleotide sequences that occur more frequently than expected in a dataset. They may represent adapter sequences, primers, ribosomal RNA, highly expressed transcripts, contamination, or other abundant biological sequences. Their interpretation depends on the type of sequencing experiment. For example, ribosomal RNA reads may be expected in some RNA sequencing datasets if rRNA depletion was incomplete. In small RNA sequencing, particular sequence classes may naturally be abundant. Identifying overrepresented sequences helps researchers investigate possible contamination or technical bias without assuming that every highly represented sequence is an artifact.
  • Ambiguous nucleotide bases are commonly represented by the letter N in FASTQ files. They indicate that the basecaller could not confidently assign a specific nucleotide at that position. The percentage of ambiguous bases can be calculated by dividing the number of N bases by the total number of bases and multiplying by 100. A high proportion of ambiguous bases may indicate poor signal quality or technical problems. However, ambiguous bases can also occur in certain biological datasets, and their significance should be evaluated according to the sequencing method and research objective.
  • Read quality assessment also examines sequence complexity. Low-complexity reads contain repetitive or biased nucleotide patterns, such as long stretches of a single nucleotide or short motifs repeated many times. Such reads may align to multiple genomic locations or produce misleading similarity-search results. Low-complexity sequence detection is particularly important in genome analysis, metagenomics, and sequence similarity searching. Tools such as DUST and SEG are used in different contexts to identify low-complexity nucleotide and protein regions, respectively. Researchers should distinguish genuine biological repeats from technical artifacts before masking or removing sequences.
  • Contamination screening is another important part of read quality assessment. Sequencing libraries may contain unwanted DNA or RNA from host organisms, microorganisms, laboratory reagents, or other sources. Contamination can affect taxonomic classification, genome assembly, gene annotation, and variant analysis. Tools such as Kraken2, Centrifuge, and FastQ Screen can help identify possible sources of reads by comparing them with reference databases. The reliability of contamination screening depends on the completeness and quality of the reference database, the taxonomic resolution of the reads, and the presence of closely related organisms.
  • Quality assessment of paired-end sequencing data requires additional checks. Paired-end reads originate from opposite ends of the same DNA fragment and are commonly stored in two separate FASTQ files. The read identifiers should correspond between the forward and reverse files, and the records should remain in the correct order when paired-end tools are used. Researchers should also examine the quality profiles of both read directions because one direction may have lower quality than the other. Incorrect pairing, missing records, or mismatched sample files can cause errors in alignment and downstream analysis.
  • The methods used for read quality assessment differ between short-read and long-read sequencing. In Illumina sequencing, common quality-control measurements include per-base quality, per-sequence quality, GC content, adapter contamination, duplication, and read length. FastQC and fastp are widely used for these tasks. For Oxford Nanopore sequencing, read length, base-call quality, read yield, length distribution, and signal-related metrics are particularly important. NanoPlot, NanoStat, pycoQC, and platform-specific tools can provide useful summaries. PacBio sequencing workflows commonly assess read length, read quality, subread characteristics, and consensus-read quality using SMRT Link and related software. The metrics and thresholds should be selected according to the characteristics of the sequencing technology.
  • Several bioinformatics tools are available for read quality assessment. FastQC is one of the most widely used programs for evaluating FASTQ files. It produces graphical reports covering sequence quality, nucleotide composition, GC distribution, read length, duplication, adapter content, and overrepresented sequences. MultiQC combines results from FastQC and many other bioinformatics tools into a consolidated report, making it useful for comparing multiple samples and sequencing runs.
  • fastp is a fast and flexible tool that performs quality assessment as well as read preprocessing. It can identify adapter sequences, filter low-quality reads, trim low-quality bases, and generate HTML and JSON reports. Cutadapt is particularly useful for identifying and removing adapter and primer sequences. Trimmomatic provides trimming and filtering operations for short-read sequencing data. For long-read sequencing, NanoPlot and NanoStat provide summaries of read length and quality, while other platform-specific tools support more detailed performance assessment.
  • A simple example of running FastQC on a FASTQ file is shown below:

fastqc sample_R1.fastq.gz

  • For paired-end sequencing, both read files can be examined together:

fastqc sample_R1.fastq.gz sample_R2.fastq.gz

  • The generated HTML reports can be opened in a web browser. For a project containing several samples, MultiQC can combine the quality-control results:

multiqc .

  • The following command demonstrates quality assessment and preprocessing using fastp:

fastp \

    -i sample_R1.fastq.gz \

    -I sample_R2.fastq.gz \

    -o sample_R1_clean.fastq.gz \

    -O sample_R2_clean.fastq.gz \

    -h fastp_report.html \

    -j fastp_report.json

  • This command processes paired-end reads, performs automatic adapter detection where supported, applies the tool’s default filtering and trimming behaviour, and produces HTML and JSON reports. The output files should be examined to confirm that the processing is appropriate for the experiment. Default settings are not necessarily optimal for every sequencing dataset.
  • Read quality can also be assessed programmatically using Python. Biopython allows researchers to read FASTQ records and examine their nucleotide sequences and quality scores. The following example calculates the average Phred quality score for each read:

from Bio import SeqIO

fastq_file = “sample_R1.fastq”

for record in SeqIO.parse(fastq_file, “fastq”):

    qualities = record.letter_annotations[“phred_quality”]

    if qualities:

        average_quality = sum(qualities) / len(qualities)

        print(record.id, round(average_quality, 2))

  • A further analysis can calculate the proportion of bases meeting a selected quality threshold. The following example calculates the percentage of bases with a Phred score of at least Q30:

from Bio import SeqIO

fastq_file = “sample_R1.fastq”

total_bases = 0

high_quality_bases = 0

for record in SeqIO.parse(fastq_file, “fastq”):

    qualities = record.letter_annotations[“phred_quality”]

    total_bases += len(qualities)

    high_quality_bases += sum(q >= 30 for q in qualities)

if total_bases > 0:

    q30_percentage = (high_quality_bases / total_bases) * 100

    print(“Q30 bases (%):”, round(q30_percentage, 2))

else:

    print(“No bases found.”)

  • This calculation provides a dataset-level summary of the proportion of bases with estimated error probabilities of 0.1% or lower. It should be considered alongside per-base quality distributions and other quality metrics because a high Q30 percentage does not guarantee that every read is reliable.
  • Quality-control reports commonly classify results using warning and failure indicators. In FastQC, these indicators are intended to draw attention to unusual patterns, not to provide a universal pass-or-fail decision for every biological experiment. For example, a sample may receive a warning for GC distribution because its biological composition differs from the expected reference distribution. Similarly, high duplication may be normal in targeted sequencing or RNA sequencing. Researchers should interpret the results in the context of the sample type, library preparation, sequencing technology, and experimental design.
  • Read quality assessment is closely connected to preprocessing. If adapter contamination is detected, adapter trimming may be required. If terminal bases have consistently low quality, quality trimming may improve the reliability of downstream analysis. Reads containing too many ambiguous bases or very short sequences after trimming may be filtered, depending on the intended application. However, aggressive trimming can remove useful sequence information, reduce read length, and introduce bias. The aim should be to correct identifiable technical problems while preserving as much valid biological information as possible.
  • The appropriate quality requirements depend on the downstream analysis. For genome sequencing, read quality and coverage influence alignment, variant calling, and genome assembly. In RNA sequencing, read quality affects transcript mapping, gene quantification, and differential expression analysis. In metagenomics, quality problems can influence taxonomic classification, microbial genome reconstruction, and functional profiling. In targeted sequencing, coverage uniformity, amplicon representation, and primer-related artifacts may be particularly important. In single-cell sequencing, quality assessment must also consider cell barcodes, UMIs, read assignment, and the number of usable reads per cell.
  • Quality assessment should also include sample-level checks that cannot be established from base-call quality alone. These may include confirming sample identity, checking expected organism or tissue composition, evaluating contamination, examining coverage distribution, and comparing sequencing metrics across biological replicates. For human genomic data, contamination and sample identity checks can be particularly important before clinical or research interpretation. In population and metagenomic studies, sample composition and biological diversity must also be considered when interpreting unusual sequence patterns.
  • For reproducible bioinformatics analysis, researchers should preserve the original sequencing files and retain the quality-control reports generated before and after preprocessing. The sequencing platform, instrument, chemistry, basecalling software, library preparation method, sample identifiers, and relevant experimental metadata should be recorded. Software names, versions, command-line parameters, and reference databases should also be documented. File checksums can help verify data integrity, while workflow systems such as Snakemake and Nextflow can support consistent execution of quality-control and preprocessing steps across multiple samples.
  • Read quality assessment is a fundamental stage in sequencing data analysis because it identifies technical problems that could compromise biological interpretation. By evaluating Phred quality scores, read length, GC content, nucleotide composition, duplication, adapter contamination, ambiguous bases, sequence complexity, and contamination, researchers can develop a more complete understanding of sequencing data reliability. Tools such as FastQC, MultiQC, fastp, Cutadapt, NanoPlot, and Biopython support this process across different sequencing technologies. Careful interpretation of quality-control results, followed by appropriate preprocessing and documentation, helps ensure that downstream bioinformatics analyses are accurate, reliable, and reproducible.
Author:

Leave a Reply

Your email address will not be published. Required fields are marked *