- Raw sequencing data are the primary outputs generated by DNA and RNA sequencing technologies. They contain information about the nucleotide sequences obtained from biological samples and, in many sequencing platforms, measurements or estimates of the confidence associated with each base. These data form the starting point for most bioinformatics workflows, including sequence quality control, read preprocessing, reference alignment, genome assembly, variant detection, gene expression analysis, metagenomics, and comparative genomics. Understanding raw sequencing data and their associated quality scores is essential for evaluating sequencing reliability and ensuring that downstream biological conclusions are supported by accurate measurements.
- The form of raw sequencing data depends on the sequencing technology. In conventional short-read sequencing, such as Illumina sequencing, the primary output commonly consists of base-called reads stored in FASTQ files. Each read contains a nucleotide sequence and a corresponding quality score for every base. In long-read technologies, including Oxford Nanopore and Pacific Biosciences sequencing, the original instrument output may also include electrical-current measurements or fluorescence-based observations. These measurements are processed by basecalling software to produce nucleotide sequences and quality estimates. Therefore, it is important to distinguish raw instrument signals from base-called sequencing reads. FASTQ files generally contain base-called sequence data rather than the original physical measurements produced by the sequencing instrument.
- In Illumina sequencing, DNA fragments are amplified into clusters on a flow cell and sequenced through cycles of fluorescently labelled nucleotide incorporation. Images collected during sequencing are processed by basecalling software to identify the nucleotide incorporated at each cycle. The resulting reads are assigned quality scores that estimate the probability of an incorrect base call. Modern sequencing systems perform much of this processing automatically and produce demultiplexed FASTQ files, often separated by sample and read direction. Paired-end sequencing generates two reads from opposite ends of the same DNA fragment, commonly represented in separate files such as sample_R1.fastq.gz and sample_R2.fastq.gz.
- Oxford Nanopore sequencing produces raw electrical-current measurements as DNA or RNA molecules pass through nanopores. These measurements are commonly stored in POD5 or, in older workflows, FAST5 files. Basecalling software, such as Dorado, analyses the signal and converts it into nucleotide sequences with associated quality scores. Depending on the workflow, the resulting data may be written in FASTQ or BAM format. Nanopore data can also contain information about modified bases, although the ability to identify particular modifications depends on the sequencing chemistry, basecalling model, and analysis method. PacBio sequencing records fluorescent signals associated with nucleotide incorporation in zero-mode waveguides. Its raw and intermediate data can be processed to generate subreads and consensus sequences, including circular consensus sequencing (CCS) reads, commonly called HiFi reads. These outputs can also include quality information that helps evaluate the confidence of individual bases.
- The FASTQ format is one of the most widely used formats for storing raw, base-called sequencing reads. It is a text-based format in which each read is normally represented by four lines. The first line begins with the @ symbol and contains the read identifier and optional metadata. The second line contains the nucleotide sequence. The third line begins with a + symbol and may repeat the read identifier. The fourth line contains quality characters corresponding position by position to the nucleotide sequence. Each quality character encodes a numerical quality score using a defined character-encoding scheme.
- A simplified FASTQ record is shown below:
@READ_001
ATGCTAGCTAGCTACGATCG
+
IIIIIIIIIIIIIIIIIIII
- In this example, the first line identifies the read, the second line contains its nucleotide sequence, and the third line separates the sequence from its quality information. The fourth line contains encoded quality characters. The character I represents a quality value under the commonly used Phred+33 encoding. This example is illustrative; actual FASTQ files may contain longer identifiers, additional metadata, different sequence lengths, and a mixture of quality characters.
- The quality score associated with a nucleotide is commonly expressed as a Phred quality score, represented by the symbol Q. It measures the estimated probability that a base call is incorrect. The Phred score is calculated using the negative logarithm of the probability of error:
[
Q = -10\log_{10}(P)
]
- Here, (P) represents the estimated probability that the nucleotide has been called incorrectly. A higher Phred score indicates a lower estimated probability of error and therefore greater confidence in the base call. Because the scale is logarithmic, an increase of 10 in the Phred score corresponds to a tenfold reduction in the estimated error probability.
- For example, a Phred score of 10 corresponds to an estimated error probability of 10%, while a score of 20 corresponds to 1%. A score of 30 corresponds to 0.1%, and a score of 40 corresponds to 0.01%. These values are estimates of base-calling accuracy, not guarantees that an individual nucleotide is correct. The quality model depends on the sequencing platform, chemistry, basecalling software, calibration, and characteristics of the sequencing data.
| Phred score | Estimated error probability | Estimated base-call accuracy |
| Q10 | 10% | 90% |
| Q20 | 1% | 99% |
| Q30 | 0.1% | 99.9% |
| Q40 | 0.01% | 99.99% |
- Quality scores are stored as ASCII characters in FASTQ files rather than as ordinary numerical values. In the widely used Phred+33 encoding, the numerical Phred score is obtained by subtracting 33 from the ASCII code of the quality character. For example, the character I has an ASCII code of 73, corresponding to a Phred score of 40 under Phred+33 encoding. Older sequencing datasets may use Phred+64 encoding, in which 64 is subtracted instead. The encoding must be interpreted correctly because using the wrong offset produces incorrect quality values. Modern sequencing workflows commonly use Phred+33, but the encoding should be verified when working with older or unusual datasets.
- The quality score of each base provides information about the reliability of an individual nucleotide call. However, sequencing quality is also assessed across complete reads, sequencing cycles, samples, and datasets. Per-base sequence quality examines the distribution of quality scores at each position across all reads. Per-sequence quality evaluates the overall quality distribution of individual reads. Per-base quality often declines toward the ends of short reads, particularly in some sequencing workflows, because signal quality can deteriorate over successive sequencing cycles. Low-quality terminal bases may increase the likelihood of mismatches and incorrect alignments if they are not handled appropriately.
- Other important quality measurements include nucleotide composition, GC content, read length distribution, ambiguous base frequency, sequence duplication, adapter contamination, and overrepresented sequences. Unexpected nucleotide composition may indicate contamination, biased library preparation, or unusual biological sequence composition. Abnormal GC distributions can arise from technical bias or from the biological characteristics of the sample. High duplication levels may reflect PCR amplification, low library complexity, or genuine high-abundance sequences. In RNA sequencing, highly expressed transcripts can naturally produce many reads from the same regions, so duplication should not automatically be interpreted as a sequencing failure.
- Ambiguous bases are sometimes represented by the letter N, indicating that the sequencing system could not confidently identify a nucleotide at a particular position. The presence of Ns can result from weak signals, overlapping signals, poor-quality measurements, or other technical difficulties. A high proportion of ambiguous bases can reduce the usefulness of reads for alignment, assembly, and variant detection. However, the significance of ambiguous bases depends on their frequency, location, and the intended analysis.
- Quality scores are also used to evaluate sequencing reads before downstream analysis. A common approach is to calculate the proportion of bases that meet a selected quality threshold, such as Q20 or Q30. The Q30 percentage represents the proportion of bases with an estimated error probability of 0.1% or lower. Average quality scores can provide a general overview of a dataset, but they may conceal important differences between individual reads or positions. Consequently, quality assessment should examine distributions and positional patterns rather than relying on a single average value.
- For paired-end sequencing, quality information should be evaluated separately for the forward and reverse read files. The two read directions may have different quality profiles because of differences in sequencing cycles, library characteristics, or platform-specific effects. Read pairs should also be checked for consistency in identifiers and sample assignment. If one read file contains records that do not correspond correctly to the other, paired-end alignment and downstream processing may be affected.
- Several bioinformatics tools are commonly used to inspect raw sequencing data and quality scores. FastQC provides graphical summaries of FASTQ quality, sequence composition, duplication, adapter content, and other metrics. MultiQC combines results from multiple quality-control reports into a single summary, making it particularly useful for large sequencing projects. fastp performs quality assessment and preprocessing, including adapter detection, trimming, filtering, and reporting. NanoPlot and NanoStat are designed for quality assessment of Oxford Nanopore sequencing data, while PacBio workflows can use SMRT Link and related tools to inspect read quality and sequencing performance. The choice of software should reflect the sequencing platform and data type.
- A simple example of running FastQC on a FASTQ file is:
fastqc sample_R1.fastq.gz
- For paired-end data, both files can be assessed together:
fastqc sample_R1.fastq.gz sample_R2.fastq.gz
- The resulting HTML report can be opened in a web browser to examine the quality metrics and identify possible problems. For a project containing many samples, MultiQC can combine the reports:
multiqc .
- The FASTQ format can also be examined programmatically using Python. Biopython provides functions for reading FASTQ records and accessing their sequences and quality scores. The following example reads a FASTQ file and 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, average_quality)
- The phred_quality annotation contains the numerical quality score for each nucleotide. The example assumes that the input is a standard FASTQ file whose quality encoding is correctly recognized by the parser. Compressed FASTQ files can also be read with suitable file handling or libraries that support gzip-compressed input.
- Quality scores are important in read preprocessing because low-quality bases can contribute to mismatches, false variant calls, incorrect sequence alignments, and fragmented genome assemblies. Trimming and filtering methods may remove low-quality terminal bases, discard reads below a selected quality threshold, or remove reads that are too short after trimming. Tools such as fastp, Cutadapt, and Trimmomatic support different preprocessing operations. However, aggressive trimming can remove useful biological information, reduce read length, and introduce bias. Quality thresholds should therefore be selected according to the sequencing platform, read characteristics, and downstream analytical objective.
- Quality scores also influence reference alignment and variant calling. During alignment, low-quality bases may be less reliable evidence for matching a read to a reference genome. Some aligners use base-quality information in their scoring or downstream processing, while variant callers use quality-related evidence to evaluate alternative alleles. In workflows using the Genome Analysis Toolkit (GATK), base quality score recalibration can be applied under suitable conditions to adjust systematic quality-score biases. Such recalibration depends on appropriate reference and known-variant resources and is not automatically necessary or appropriate for every organism or dataset.
- In genome assembly, sequencing quality affects the reliability of overlaps, consensus sequences, and contig construction. High-quality reads generally provide stronger evidence for resolving nucleotide positions, while errors can produce incorrect overlaps or misassemblies. Long-read sequencing introduces additional considerations because error profiles vary among platforms and basecalling models. PacBio HiFi reads generally provide highly accurate consensus sequences, whereas Oxford Nanopore read accuracy depends on chemistry, pore type, basecalling model, and analysis conditions. Quality scores should therefore be interpreted in the context of the technology that generated the reads.
- In RNA sequencing, quality scores help assess the reliability of reads used for transcript quantification, differential gene expression analysis, transcript assembly, and alternative splicing studies. Low-quality reads may map ambiguously or fail to align, potentially affecting gene-expression estimates. In metagenomic sequencing, sequencing errors can create artificial sequence variants, influence taxonomic classification, and complicate microbial genome assembly. In single-cell sequencing, read quality is considered alongside cell barcodes and unique molecular identifiers (UMIs), because errors in these regions can affect cell assignment and molecule counting.
- It is important to distinguish base-call quality from other measures of sequencing reliability. A high Phred score indicates that the basecaller estimates a low probability of an incorrect nucleotide call, but it does not prove that the read originated from the intended organism or genomic region. Contamination, sample swaps, index hopping, PCR bias, reference-genome errors, and systematic sequencing artifacts may remain even when base-call quality is high. Similarly, a high average quality score does not guarantee uniform coverage or the absence of systematic errors. Quality assessment must therefore combine base-level scores with sample-level checks, contamination screening, coverage analysis, and biological validation.
- Quality scores are also not directly interchangeable between sequencing platforms. Different technologies use different signal-generation processes, error models, and basecalling algorithms. A Q30 score from one platform should not automatically be assumed to represent exactly the same practical performance as a Q30 score from another platform without considering calibration and empirical validation. For this reason, platform-specific quality-control tools and benchmarking datasets are useful when comparing sequencing technologies or evaluating changes in chemistry and software.
- For reliable and reproducible bioinformatics analysis, the original sequencing files should be preserved before preprocessing. Researchers should record the sequencing platform, instrument and chemistry where available, basecalling software and model, file format, quality encoding, sample identifiers, read layout, library preparation method, and relevant sequencing-run metadata. The versions and parameters of quality-control and preprocessing tools should also be documented. Both raw and processed files should be retained when possible, together with quality-control reports and checksums that can be used to verify file integrity.
- Raw sequencing data and quality scores are fundamental to modern bioinformatics because they provide the nucleotide information and confidence estimates required for downstream analysis. FASTQ files are widely used to store base-called reads together with per-base quality values, while platform-specific raw signal formats preserve information generated directly by sequencing instruments. Phred quality scores provide a standardized logarithmic measure of estimated base-call error probability, but they must be interpreted alongside other quality metrics and the characteristics of the sequencing technology. Careful assessment, appropriate preprocessing, and complete documentation help researchers minimize technical errors and improve the accuracy, reliability, and reproducibility of genomic and other omics analyses.