Phred Quality Score

A Phred quality score (Q) is the universal logarithmic metric assigned to each individual nucleotide base call generated by automated DNA sequencing platforms, quantifying the probability (P_e) that the base caller made an incorrect substitution, insertion, or deletion error at that position. Invented in 1998 by Brent Ewing, LaDeana Hillier, Michael C. Wendl, and Phil Green at the University of Washington during the Human Genome Project, the Phred algorithm replaced subjective human inspection of four-color fluorescence chromatograms with calibrated per-base error probabilities that weighted automated de novo sequence assembly in the Phred–Phrap–Consed software suite.

Formally, the Phred quality score Q is defined as negative ten times the base-10 logarithm of the estimated base-calling error probability P_e:

Q = −10 log₁₀(P_e) ⟺ P_e = 10^−*Q*/10

Because the scale is logarithmic (analogous to the decibel scale in acoustics), every 10-unit increase in Q corresponds to a 10-fold reduction in base-calling error probability: Q10 denotes a 1-in-10 error probability (90.0% accuracy), Q20 denotes a 1-in-100 error probability (99.0% accuracy), Q30 denotes a 1-in-1,000 error probability (99.9% accuracy, the standard benchmark for clinical massively parallel sequencing), Q40 denotes a 1-in-10,000 error probability (99.99% accuracy, the 1997 Human Genome Project "Bermuda Standard" for finished reference sequence), and Q50 denotes a 1-in-100,000 error probability (99.999% accuracy).

Four-Parameter Trace Analysis in Automated Sanger Sequencing

When Brent Ewing and Phil Green developed the Phred base caller for slab-gel and capillary Sanger sequencing in 1998, existing base callers output only a four-letter string (A, C, G, T, or N) without distinguishing clean, well-resolved fluorescence peaks from noisy, overlapping peaks near the end of a read. Phred introduced a two-stage Fourier analysis and dynamic-programming trace parser that calculated four quantitative signal-quality parameters within a sliding window of 3 to 7 bases around each called peak:

1. Peak Spacing Ratio: The ratio of the largest peak-to-peak interval to the smallest peak-to-peak interval across a 7-base window (ideal value = 1.0 for uniformly migrating dideoxy fragments).
2. Uncalled-to-Called Amplitude Ratio (7-base window): The ratio of the maximum fluorescence intensity of the largest uncalled secondary background peak (noise or dye-blob artifact) to the smallest called peak across 7 contiguous bases.
3. Uncalled-to-Called Amplitude Ratio (3-base window): The local signal-to-noise amplitude ratio across the 3-base neighborhood (i − 1, i, i + 1).
4. Peak Resolution / Valley Depth: The distance in bases to the nearest unresolved peak pair that lacked a local minimum between adjacent nucleotides.

Using a calibration training set of hundreds of millions of sequenced bases aligned against finished reference cosmids from Caenorhabditis elegans and human chromosomes, Phred partitioned the four-dimensional parameter space into discrete lookup bins, measured the empirical error rate P_obs within each bin, and assigned the calibrated integer score Q = round(−10 log₁₀ P_obs).

Quantitative Conversion Table of Phred Scores, Error Probabilities, and ASCII Encodings

The Four-Line FASTQ Format and Single-Byte ASCII Compression

In early Sanger sequencing, Phred quality scores were stored in separate .qual text files as space-delimited multi-digit integers (e.g., 40 40 38 35 22 14), which doubled file management overhead and required 3 bytes per base. Around 2000, Jim Mullikin at the Wellcome Trust Sanger Institute (later formalized in 2010 by Peter J. A. Cock, Christopher J. Fields, Naohisa Goto, Michael L. Heuer, and Peter M. Rice) invented the FASTQ file format, which embeds both the nucleotide sequence and its per-base Phred quality scores into a four-line plain-text record per read:

• Line 1: @ followed by the instrument/flow-cell read identifier and optional description.
• Line 2: The raw nucleotide sequence letters (A, C, G, T, N).
• Line 3: + (optionally followed by a repeat of the read identifier).
• Line 4: A string of single printable ASCII characters of the same length as Line 2, encoding one Phred quality score per byte.

Because ASCII values 0 through 31 are non-printing control characters (such as newline, tab, and carriage return) and ASCII 32 is the space character ( ), the universal Sanger / Phred+33 encoding standard (adopted by Illumina CASAVA 1.8+, PacBio, Oxford Nanopore, and NCBI SRA) shifts each integer quality score Q ∈ [0, 93] by adding 33:

ASCIIByte = Q + 33 ⟺ Q = ASCIIByte − 33

Thus, Q = 0 maps to ASCII 33 (!), Q = 10 maps to ASCII 43 (+), Q = 20 maps to ASCII 53 (5), Q = 30 maps to ASCII 63 (?), and Q = 40 maps to ASCII 73 (I). Historically, early Solexa and Illumina Pipeline 1.3–1.7 instruments (2006–2011) used an incompatible Phred+64 offset (where Q = 0 mapped to ASCII 64, @), which bioinformatics quality-control tools (FastQC, fastp, Trimmomatic) detect and convert to Phred+33. To reduce cloud storage costs on NovaSeq X and NextSeq 2000 sequencers, Illumina introduced Q-score binning, quantizing integer Phred scores into 3 or 4 discrete bins (e.g., Q2, Q12, Q24, Q37 or Q40), which improves lossless CRAM/gzip compression ratios by >50% without degrading variant-calling accuracy.

Base Quality Score Recalibration (BQSR) and Averaging Phred Scores Correctly

Raw Phred scores emitted by a sequencing instrument's real-time analysis (RTA) software exhibit systematic calibration drift depending on sequencing cycle number (errors rise toward cycle 150 as phasing accumulates in Illumina sequencing-by-synthesis), flow-cell tile geometry, and preceding dinucleotide sequence context (such as GG or homopolymer runs). In the GATK pipeline for whole genome sequencing and exome sequencing, Base Quality Score Recalibration (BQSR) aligns reads to the reference genome, masks known polymorphic sites from dbSNP and gnomAD, counts empirical mismatch rates across every combination of (Reported Q, Machine Cycle, Dinucleotide Context), and outputs recalibrated Phred scores that match empirical error probabilities.

Finally, a frequent mathematical pitfall in sequence analysis is computing the mean Phred score of a read or amplicon by taking the direct arithmetic mean of the Q values, (1/L) ∑ Q_i. Because Q is logarithmic, arithmetic averaging of Q computes the geometric mean of the error probabilities, which underestimates the expected number of errors in a read containing even one or two low-quality bases. To compute the mean quality score (Q_mean) across a read of length L, each base's Phred score Q_i must first be converted back to its linear error probability P_e,i = 10^−*Q*_i/10, averaged in probability space, and transformed back to the Phred scale:

Q_mean = −10 log₁₀ [ (1 / L) ∑_*i*=1^L 10^−*Q*_i/10 ]

For example, in a 2-base sequence with one Q40 base (P_e = 0.0001) and one Q10 base (P_e = 0.1), the naive arithmetic average of Q is (40 + 10)/2 = Q25 (implying P_e = 0.00316), whereas the actual mean error probability is (0.0001 + 0.1)/2 = 0.05005, which corresponds to Q13.0—a 16-fold difference in error rate.

Why did the Human Genome Project establish Q40 (1 error in 10,000 bases) as the "Bermuda Standard" for finished genomic sequence?

At the Second International Strategy Meeting on Human Genome Sequencing in Bermuda (1997), consortium leaders set the quality standard for "finished" human reference sequence at 99.99% consensus accuracy (≤1 error per 10,000 bases, or Phred Q40). Because natural human heterozygosity averages ~1 single-nucleotide polymorphism (SNP) per 1,000 to 1,200 base pairs (0.08%–0.1% divergence between maternal and paternal chromosomes), a consensus error rate of Q20 (1% error) or Q30 (0.1% error) would generate as many false-positive sequencing errors as biological polymorphisms. Setting the finished threshold at Q40 (0.01% error) ensured that genetic variants outnumbered reference assembly errors by at least 10-to-1.

What is the difference between a Base Quality Score (BQ) in a FASTQ file and a Mapping Quality Score (MAPQ) in a BAM/CRAM file?

Although both metrics use the Phred logarithmic formula Q = −10 log₁₀(P), they measure two independent sources of uncertainty. Base Quality (BQ) is assigned by the sequencing instrument to each individual nucleotide letter inside a read and measures the biochemical/optical probability that the base caller misidentified that single nucleotide. Mapping Quality (MAPQ)—introduced by Heng Li in MAQ and BWA—is a single integer assigned to the read alignment in a SAM/BAM/CRAM file, measuring the probability that the aligner placed the read at the wrong genomic coordinate (e.g., a read with Q40 base qualities that maps equally well to two identical segmental duplications receives MAPQ = 0 because its true origin is 50% uncertain).

Why do Illumina sequencing-by-synthesis Phred scores systematically decline from 5′ to 3′ toward the end of a 150-bp or 300-bp read?

In Illumina cyclic reversible termination chemistry, each sequencing cycle requires the ~1,000 identical DNA strands within a flow-cell cluster to incorporate a single 3′-O-azidomethyl blocked dNTP and then undergo chemical deblocking by TCEP for the next cycle. In practice, ~0.1% to 0.2% of strands fail to incorporate or deblock in each cycle (phasing, falling one base behind the cluster) or incorporate an unblocked trace contaminant (pre-phasing, jumping one base ahead). By cycle 125 to 150 (and cycles 200–300 in Read 2), cumulative out-of-phase fluorescence lowers the signal-to-noise ratio of the dominant base call, causing the Phred score to drop from Q35–Q40 at cycle 1 down to Q20–Q28 at the 3′ terminus.

1. Ewing, B., Hillier, L., Wendl, M. C., & Green, P. (1998). "Base-Calling of Automated Sequencer Traces Using Phred. I. Accuracy Assessment." Genome Research, 8(3), 175–185. https://doi.org/10.1101/gr.8.3.175
2. Ewing, B., & Green, P. (1998). "Base-Calling of Automated Sequencer Traces Using Phred. II. Error Probabilities." Genome Research, 8(3), 186–194. https://doi.org/10.1101/gr.8.3.186
3. Cock, P. J. A., Fields, C. J., Goto, N., Heuer, M. L., & Rice, P. M. (2010). "The Sanger FASTQ File Format for Sequences with Quality Scores, and the Solexa/Illumina FASTQ Variants." Nucleic Acids Research, 38(6), 1767–1771. https://doi.org/10.1093/nar/gkp1137
4. DePristo, M. A., Banks, E., Poplin, R., Garimella, K. V., Maguire, J. R., Hartl, C., Philippakis, A. A., del Angel, G., Rivas, M. A., Hanna, M., et al. (2011). "A Framework for Variation Discovery and Genotyping Using Next-Generation DNA Sequencing Data." Nature Genetics, 43(5), 491–498. https://doi.org/10.1038/ng.806