We use Phred quality score to define the basecalling quality of each base in the sequence.

A Phred quality score is a measure of confidence based on the estimated error rate and is calculated as -10 x log (Pe) where Pe is the estimated probability of error.

Phred quality scores are given in a negative log 10 scale, such that increasing q-scores will indicate greater confidence in a predicted base.

For example, an error of 1 in 100 will give a q-score of 20 and an error of 1 in 1000 will give a q-score of 30.

More information can be found in this technical document.

Per-base quality scores are stored together with the base sequence in FASTQ files output by the basecalling algorithms and are then encoded in the Sanger format using ASCII characters with values of 33 to 126 (up to 93 ASCII character values).

Note that Oxford Nanopore Technologies represents the average per-read accuracy as the modal accuracy from a sequencing run.

Phred quality scores are also used to assess consensus accuracy.

Consensus sequences are generated from the accumulation of reads from the same location, combining reads to produce a single high-quality “consensus” sequence at that region, and are most commonly used in the context of sequence assembly.

More information describing consensus accuracy can be found in this technical document and this accuracy page.