Skiver: Reference-free quality control of (meta)genomic sequencing datasets using (k, v)-mer sketches

Skiver is a tool that aims to perform quality control for a set of reads, estimating the sequencing error rates/types, without relying on the quality scores or the need for a reference genome. It works the best for metagenomic samples where at least one genome has high coverage (>20 $\times$).

Installation

We provide several ways of installing skiver CLI.

Conda install

conda install -c bioconda skiver

Download executable

Simply download the executable from the latest release, via the following

wget https://github.com/GZHoffie/skiver/releases/latest/download/skiver
chmod +x ./skiver
./skiver

Build from source

Alternatively, build skiver from the source code. Install rust, and build using

git clone https://github.com/GZHoffie/skiver.git
cd skiver

# If default rust install directory is ~/.cargo
cargo install --path . --root ~/.cargo

Quick start

Basic commands:

# sketch the read files + analyze
skiver sketch [sequence_file_1] [sequence_file_2] ... -o sequences.kvmer
skiver analyze sequences.kvmer -o output_prefix

# Alternatively, analyze the files directly
skiver analyze [sequence_file_1] [sequence_file_2] ... -o output_prefix

# If a reference genome is provided
# Add `--use-all` to disable the iterative outlier filter
skiver analyze [sequence_file_1] [sequence_file_2] ... -r [reference_file] --use-all -o output_prefix

The input sequence files can be represented using regex. Gzipped files are also accepted.

For the full set of available options, use the help function,

skiver sketch -h
skiver analyze -h

skiver App

We provide an all-in-one app for analyzing and visualizing the information in sequencing datasets. Please refer to this guide for detailed steps to run the app on your system.

Please see the latest release page for download links. Currently, Windows and Ubuntu are supported.

Interpreting the results

See this guide for a detailed documentation of what each output file contains.

Example

We provide scripts in ./scripts for easy visualization of skiver's output. Below is an example of analysis using B. Subtilis isolate reads from Loman Lab.

# Download the reads with SRA toolkit. If not installed, install with
# `conda install -c bioconda sra-tools`
# This read set takes ~3GB space, ~10 min to download
prefetch SRR7498042
fasterq-dump SRR7498042 # In some versions, need to run `fastq-dump SRR7498042` instead

# Create the (k,v)-mer sketch of the data in the example/ folder
mkdir -p example
skiver sketch SRR7498042.fastq -o example/SRR7498042.kvmer

# Run skiver analyze, with all the verbose output
skiver analyze example/SRR7498042.kvmer -o example/SRR7498042

# visualize the output in the figures/ folder
mkdir -p figures
python scripts/plot_all.py example/SRR7498042 -o figures/SRR7498042

This creates the verbose output in example/SRR7498042.*.csv, and visualize the output in figures/SRR7498042_*.png.

skiver analyze also generates six summary figures as vector PDFs using the Rust kuva package. It writes example/SRR7498042.plot_*.pdf files alongside the CSV reports.

Apart from plot_all.py, you can also use the individual scripts and adjust the parameters like the following.

• Visualizing hazard rate and survival rate estimates

  python scripts/plot_hazard_survival_rate.py example/SRR7498042.hazard_rate.csv example/SRR7498042.summary_error_rate.csv figures/SRR7498042_hazard_survival.png -t 1 -T 100 > survival_rate_estimates.csv

  This command outputs the plot hazard_survival_rate.png, along with the estimated survival rates in survival_rate_estimates.csv, with the range of t specified using -t and -T.

  

• Visualizing error spectrum

  python scripts/plot_spectrum.py example/SRR7498042.summary_error_spectrum.csv figures/SRR7498042_spectrum.png --normalize

  will plot the error spectrum in figures/spectrum.png. If --normalize is set, the error spectrum is normalized such that the frequencies sum to 1. Otherwise, they sum up to the estimated per-base error rate. The output image looks like this.

  

  On the top subplots, the error rates are calculated by accounting for both the forward and reverse complement of the reads. On the bottom, only the forward strand is included.

• Visualizing single base substitution (SBS) spectrum

  python scripts/plot_sbs96_spectrum.py example/SRR7498042.summary_error_spectrum.csv figures/SRR7498042_sbs96_spectrum.png

  will plot the SBS96 spectrum.

  

• Visualizing single base substitution (SBS) spectrum

  python scripts/plot_error_spectrum_dependence_on_t.py example/SRR7498042.summary_error_spectrum_dependence_on_v.csv figures/SRR7498042_error_spectrum_dep_t.png

  will plot how the composition of error rate change with t. If our assumption is valid, the composition should not vary too much across t.

  

• Visualizing coverage

  python scripts/plot_coverage.py example/SRR7498042.kvmer.csv example/SRR7498042.summary_error_rate.csv figures/SRR7498042_coverage.png

  will plot the estimated true coverage of the analyzed file. The true coverage is estimated by the multiplicities of the key from the sketched (k,v)-mers, divided by $\hat{S}(k)$.

  

• Quality score calibration

  python ./scripts/plot_qscore_calibration.py example/SRR7498042.summary_phred.csv figures/SRR7498042_qscore_calibration.png

  will plot the theoretical and empirical error rates of the Phred scores in log scale (if included the --log option), along with a histogram of the Phred scores.

  

• Dependence of error rate on position in the read

  python ./scripts/plot_read_position.py example/SRR7498042.summary_read_position.csv figures/SRR7498042_read_position.png

  will plot the estimated error rate at the positions close to the two ends of the read, which can help to decide how many bases to trim at each end.

  

• Dependence of error rate on GC-content of the read

  python ./scripts/plot_gc_content.py example/SRR7498042.summary_gc_content.csv figures/SRR7498042_gc_content.png

  will plot the empirical error rate observed in sketched (k, v)-mers with respect to the GC-content of the read where they come from.

  

Contribution

This is my first project in rust and this project is in early stages of development. All contributions, suggestions, and feature requests are welcomed!

I document some key design choices as well as known limitations of skiver in this documentation for future reference and directions for improvements.

Citation

Gu, Z., Sharma, P., Wong, L., & Nagarajan, N. (2026). Skiver: Reference-free quality control of metagenomic sequencing datasets using (k, v)-mer Sketches. bioRxiv, 2026-06.