phredsort is a command-line tool for sorting sequences in FASTQ files by their quality scores.

Usage

Basic usage:

# Read from `input.fastq.gz` and write to `output.fastq.gz`
phredsort -i input.fastq.gz -o output.fastq.gz

# Read from stdin and write to stdout (default when -i/-o not specified)
zcat input.fastq.gz | phredsort | less -S

# Explicit stdin/stdout (equivalent to above)
zcat input.fastq.gz | phredsort -i - -o - | less -S

Sort sequences using pre-computed maxEE scores in headers

phredsort headersort -i input.fasta -o output.fasta --metric maxee

Sort by avgphred scores with quality filtering

phredsort headersort -i input.fastq -o output.fastq --metric avgphred --minqual 20 --maxqual 40

Sort in ascending order (lower quality first)

phredsort headersort -i input.fa -o output.fa --metric meep --ascending

Examples of supported header formats:

• Space-separated: ">seq1 maxee=2.5 size=100"
• Semicolon-separated: ">seq1;maxee=2.5;size=100"

Installation

Download compiled binary (for Linux)

wget https://github.com/vmikk/phredsort/releases/download/1.4.0/phredsort
chmod +x phredsort
./phredsort --help

Build from source

git clone --depth 1 https://github.com/vmikk/phredsort
cd phredsort
go build -ldflags="-s -w" phredsort.go
./phredsort --help

Quality metrics

phredsort supports several metrics (--metric parameter) to assess sequence quality:

1. (Back-transformed) average Phred score (avgphred)

• Properly calculated mean quality score that accounts for the logarithmic nature of Phred scores
• Converts Phred scores to error probabilities, calculates their arithmetic mean, then converts back to Phred scale
• Formula: -10 * log10(mean(10^(-Q/10)))
• More accurate than simple arithmetic mean of Phred scores, which would overestimate quality

2. Maximum expected error (maxee) (as per Edgar & Flyvbjerg, 2014)

• Sum of error probabilities for all bases in a sequence
• Formula: sum(10^(-Q/10))
• Higher values indicate lower quality
• Depends on sequence length (longer sequences tend to have higher MaxEE)

3. Maximum expected error percentage (meep)

• MaxEE standardized by sequence length
• Represents expected number of errors per 100 bases
• Formula: (MaxEE * 100) / sequence_length
• Higher values indicate lower quality
• Allows fair comparison between sequences of different lengths

4. Low quality base count (lqcount)

• Number of bases below specified quality threshold
• Useful for binned quality scores (e.g., data from Illumina NovaSeq platform)
• Counts bases with Phred score < threshold (default: 15)
• Higher values indicate lower quality

5. Low quality base percentage (lqpercent)

• Percentage of bases below quality threshold
• Formula: (lqcount * 100) / sequence_length
• Higher values indicate lower quality
• Normalizes low-quality base count by sequence length