Alpaqa: Assembly-Level Profiling And Quality Assessment

Alpaqa analyzes base-level quality scores in bacterial genome assemblies generated from Oxford Nanopore sequencing data.

The tool masks noisy regions and quantifies the number of Low-Quality Bases (LQBs) per megabase, a metric we found to correlate with overall base-level assembly accuracy. In addition, Alpaqa performs binomial tests to identify 4-, 5-, and 6-mer motifs that are significantly associated with LQBs. These motifs often represent systematic, error-prone sequence contexts and may correspond to targets of DNA modification systems.

Alpaqa operates on polished assemblies generated with Dorado polish or Medaka2. By detecting systematic error signatures, Alpaqa allows users to flag assemblies that may appear complete but contain motif-specific inaccuracies. This is particularly valuable in applications requiring high base-level accuracy, such as outbreak investigation and transmission analysis.

An automated bacterial genome assembly pipeline for ONT data which includes Medaka2 polishing, quality assessment with alpaqa, and masking of low-quality bases is available at boap.

Installation

Via pip

Alpaqa can be installed directly from PyPI:

pip install alpaqa-bio

Via uv (recommended)

If you use uv, you can run it without installation:

uvx --from alpaqa-bio alpaqa --help

Or install it:

uv tool install alpaqa-bio

Generating input files for alpaqa

Alpaqa relies on fastq assembly files with phred quality scores generated with dorado polish or medaka2 (-q flag). The tool has been tested with data generated with SUP@v5.0, SUP@v5.2, HAC@v5.0, and HAC@v5.2 basecalling models. SUP data is strongly recommended.

Using medaka2

READS=reads.fastq.gz
ASSEMBLY=draft_assembly.fasta
THREADS=16

# medaka2
medaka_consensus -i $READS -d $ASSEMBLY -o polished_assembly -t $THREADS --bacteria -q

Using dorado polish

READS=reads.fastq.gz
ASSEMBLY=draft_assembly.fasta
THREADS=16

dorado aligner $ASSEMBLY $READS | samtools sort -@ $THREADS -o aligned.bam
samtools index aligned.bam
dorado polish aligned.bam $ASSEMBLY -t $THREADS -o polished_assembly --bacteria -q

Alpaqa usage

Once installed, run alpaqa by passing your FASTQ assembly files:

alpaqa -i *.fastq --threads 16

Options

usage: alpaqa -i assembly.fastq -o output.tsv --threads 16 [options]

ALPAQA v0.1.3

options:
  -h, --help            show this help message and exit
  -i INPUT [INPUT ...], --input INPUT [INPUT ...]
                        Input fastq assembly file(s).
  -o OUTPUT, --output OUTPUT
                        Output TSV filename. (default: alpaqa_report.tsv)
  --report              Generate detailed reports on kmer analysis.
  -t THREADS, --threads THREADS
                        Number of threads. (default: 1)
  -v, --version         show program's version number and exit

Advanced:
  --all-contigs         Analyze ALL contigs > min_contig_len (Default is longest contig only).
  --no-mask             Disable masking of LQB dense regions.
  --lqb-threshold LQB_THRESHOLD
                        Phred Q-score threshold (inclusive). (default: 5)
  --min-contig-len MIN_CONTIG_LEN
                        Minimum contig length to include. (default: 50000)
  --window-size WINDOW_SIZE
                        Window size (bp) for scanning LQB dense regions. (default: 5000)
  --density-multiplier DENSITY_MULTIPLIER
                        Multiplier for baseline density to trigger masking. (default: 5.0)
  --mask-floor MASK_FLOOR
                        Absolute minimum LQB density (0.001 = 0.1%). (default: 0.001)

Output

Alpaqa generates a tab-separated (TSV) report. Each row represents one input assembly file with the following fields:

Interpretation

For assemblies generated with SUP@v5.2 data, following thresholds were established:

• <5 LQBs/Mbp: Highly accurate. These assemblies typically yield identical or near-identical cgMLST profiles when compared to Illumina-polished references.
• 5 to 10 LQBs/Mbp: Potentially reliable but should be interpreted with caution. Masking low-quality bases with fastq2a.py (see below) is recommended for downstream cgMLST analyses.
• >10 LQBs/Mbp: Usually unsuitable for high-resolution genotyping due to excessive base-level errors.

Beyond systematic sequencing errors, elevated LQB counts may indicate sample contamination, insufficient sequencing depth, or low read quality, which can be further investigated using tools such as nanoq and CheckM.

Masking low-quality bases for downstream genotyping

The helper script fastq2a.py can be used to convert FASTQ assembly files into FASTA format while masking low-quality bases. Replacing bases below a user-defined Q-score threshold with "N" helps maintain robustness in downstream bacterial cgMLST analyses. Tools such as Ridom SeqSphere exclude alleles containing these ambiguous bases to prevent false distance calculations. However, the final count of remaining target loci should be monitored to ensure sufficient resolution for high-resolution typing. For assemblies with ~5 to 10 LQBs/Mpb, masking bases with qscores ≤10 provides a good balance between accuracy and genomic resolution.

Note: if you installed via pip/uv, you can run this helper script using:

python -m alpaqa.fastq2a -i assembly.fastq -o assembly.masked.fasta -q 10

License

GNU General Public License, version 3