kira-qc

Project overview

kira-qc is a FastQC-compatible QC tool written in Rust. It generates FastQC-style text and HTML/LaTeX reports from FASTQ/FASTQ.GZ input with deterministic, bounded-memory processing and SIMD-accelerated hot paths. It is designed for short-read Illumina QC with a dedicated long-read mode for ONT/PacBio data.

Key features

• FastQC-compatible outputs: fastqc_data.txt, summary.txt, HTML, and ZIP bundle
• FASTQ and FASTQ.GZ support
• Streaming, bounded memory processing
• SIMD acceleration (AVX2 on x86_64, NEON on aarch64)
• Deterministic multithreaded execution
• Short-read (Illumina) and long-read (ONT/PacBio) modes

Supported QC modules

Module	Short-read mode	Long-read mode
Basic Statistics	Yes	Yes
Per base sequence quality	Yes	No
Per sequence quality scores	Yes	Yes
Per base sequence content	Yes	No
Per sequence GC content	Yes	Yes
Per base N content	Yes	No
Sequence Length Distribution	Yes	Yes (log-binned + N50/N90)
Sequence Duplication Levels	Yes	No
Overrepresented Sequences	Yes	No
Adapter Content	Yes	Yes (summary only)
K-mer Content (k=7)	Yes	No
Per sequence N content	No	Yes

Installation

Install from crates.io (Rust 1.91+ / Windows / Linux / macOS):

cargo install kira-qc

Or

From source (Rust 1.91+):

# build
cargo build --release

# run
./target/release/kira-qc --help

Notes:

• On x86_64, AVX2 is required at runtime.
• On aarch64 (e.g., Apple Silicon), NEON kernels are used.
• Optional build feature to disable K-mer Content (compile-time):
  • cargo build --release --features no-kmer

Usage

Basic FASTQ:

kira-qc run reads.fastq --out qc/

FASTQ.GZ:

kira-qc run reads.fastq.gz --out qc/

Short-read (default):

kira-qc run reads.fastq.gz --out qc/ --mode short

Long-read:

kira-qc run reads.fastq.gz --out qc/ --mode long

LaTeX export (supplement):

kira-qc run reads.fastq.gz --out qc/ --export-latex supplement

CLI options

Option	Description	Default
run	Run QC on a single FASTQ/FASTQ.GZ file	Required
--out <DIR>	Output directory	Required
--threads <N>	Number of worker threads	Logical CPU count
--sample-name <NAME>	Sample name (used in output folder/ZIP)	Input file stem
--phred-offset auto|33|64	Quality encoding detection or fixed offset	auto
--mode short|long	QC mode: short-read or long-read	short
--no-zip	Disable ZIP bundle creation	Off (ZIP enabled)
--export-latex summary|supplement	Generate LaTeX export	Disabled

Output description

Output directory structure:

<out>/<sample_name>_fastqc/
  fastqc_data.txt
  summary.txt
  fastqc_report.html
  latex/ (if --export-latex)
<out>/<sample_name>_fastqc.zip

• fastqc_data.txt: FastQC-style module sections and tabular data.
• summary.txt: One-line PASS/WARN/FAIL status per module.
• fastqc_report.html: Self-contained HTML report (no external assets).
• {sample_name}_fastqc.zip: ZIP bundle of the output directory (unless --no-zip).
• latex/: LaTeX export (optional), including SVG figures and kira_qc.tex.

LaTeX export for publications

kira-qc can generate an article-ready LaTeX report alongside the HTML output. Use --export-latex summary for a concise figure set or --export-latex supplement for a full module-by-module report. PDF figures are generated directly.

LaTeX artifacts are written under:

<out>/<sample_name>_fastqc/latex/

For Overleaf upload (Upload Project Overleaf feature), a ready-to-use ZIP is created:

<out>/<sample_name>_fastqc/latex/kira_qc_latex.zip

Determinism and reproducibility

kira-qc produces identical results regardless of thread count or scheduling. All aggregation is deterministic and merged by chunk index. This ensures that pipelines can rely on stable outputs for regression tests and reproducible QC.

Performance notes

• Streaming design with bounded memory for large inputs
• SIMD acceleration for base counting and quality processing
• Adapter matching includes a SIMD prefix prefilter before full pattern matching

Long-read mode notes

FastQC's per-base modules assume uniform read length and are not appropriate for ONT/PacBio data. Long-read mode disables those plots and instead focuses on:

• Log-binned length distribution with N50/N90
• Per-read quality scores
• Per-read GC content
• Per-read N content
• Adapter presence summary

FastQC compatibility notes

• Drop-in replacement for FastQC in short-read pipelines
• Output formats and module names follow FastQC conventions
• Some heuristics (e.g., duplication, k-mer content) are approximate by design

Limitations

• FASTQ/FASTQ.GZ input only (no BAM/CRAM)
• No GPU acceleration
• Long-read mode is not FastQC-compatible by design

License

MIT, See LICENSE.