BioFastq-A v2.3.0

[DilaDeniz]

Changelog

---

v2.3.0 2026-05-18

Summary

v2.3.0 is the biggest release since the initial Rust write. Every analysis that takes extra CPU is now offset by the new parallel pipeline — experimental benchmarks show v2.3.0 is faster than v2.2.0 on identical hardware despite computing significantly more statistics.

New Features

Per-Position Quality Boxplots

The quality chart now shows the full Q25/median/Q75 band per position in addition to the mean line. This makes it immediately obvious whether a position has a tight quality distribution or a fat tail of bad reads.

Per-Read GC Distribution

A histogram of per-read GC% across all reads. Bimodal distributions flag contamination; a sharp spike flags adapter dimer; a broad flat distribution flags PCR over-amplification.

Quality-vs-Length Scatter (Long-Read Mode)

For ONT/PacBio data (--long-read flag), a 2D scatter of read quality vs. read length is plotted instead of per-position quality. Automatically activated when median read length > 1 000 bp.

Long-Read / ONT Support (--long-read)

• N50, N90, read-length histogram, longest/shortest read
• Correct adapter detection for ONT consensus adapters
• Quality-vs-length scatter (see above)

Two-File Comparison Mode (--compare file2.fastq)

Run BioFastq-A on two FASTQ files and produce a unified comparison report. Useful for before/after trimming QC, sample-vs-control, or replicate agreement. Side-by-side charts highlight every metric that diverges between the two files.

Auto Parameter Suggestions

After analysis, BioFastq-A inspects the results and prints actionable trimming recommendations:

• Adapter content detected → --adapter-trim
• 3′ quality drop-off → --quality-trim 20
• Poly-G tail (NextSeq/NovaSeq) → --poly-g-trim
• Overrepresented sequences → --filter-overrep

Sliding-Window Quality Trimming (--window-size, --window-quality)

Trim reads as soon as a sliding window of N bases drops below a mean quality threshold. Gentler than hard 3′ trimming; preserves more of the read when only the last few bases degrade.

Poly-G / Poly-X Trimming (--poly-g-trim, --poly-x-trim)

Two-colour chemistry (NextSeq, NovaSeq) encodes signal loss as poly-G. BioFastq-A now detects and removes poly-G/X tails without any adapter file.

Performance

Three sources of overhead introduced by the new statistics were eliminated:

Change	Effect
Vec<Vec<u8>> → Vec<[u8;50]> for overrep sampling	Eliminates 200K heap allocs per file; rayon reduce becomes memcpy instead of HashMap merge
qual_hist_by_pos [u64;43] → [u32;43]	Halves working set from ~52 KB to ~26 KB; active region fits in L1 cache
Split combined quality loop into two passes	Allows LLVM to auto-vectorise the sequential pass; scatter pass runs independently

Result on 1M × 150 bp benchmark (8 threads):

Version	Throughput
v2.2.0 (main)	~295 MB/s
v2.3.0 (this release)	~308 MB/s

v2.3.0 computes more statistics and is still ~4% faster.

What BioFastq-A Does That Others Don't

Feature	BioFastq-A	FastQC	fastp	MultiQC
Single native binary (no JVM, no Python)	✓	✗	✓	✗
mmap zero-copy I/O pipeline	✓	✗	✗	✗
Per-position Q25/median/Q75 boxplot	✓	✓	✗	✗
Per-read GC histogram	✓	✓	partial	✗
Quality-vs-length scatter (long read)	✓	✗	✗	✗
Long-read / ONT mode (N50, N90)	✓	✗	✗	✗
Two-file comparison report	✓	✗	✗	✓
Auto trimming parameter suggestions	✓	✗	✗	✗
Sliding-window quality trimming	✓	✗	✓	✗
Poly-G / poly-X trimming	✓	✗	✓	✗
Built-in adapter library	✓	✓	✓	✗
Paired-end support	✓	✓	✓	✗
HyperLogLog deduplication estimate	✓	✗	✗	✗
JSON + HTML output	✓	✗	✓	✓
Bioconda package	✓	✓	✓	✓

Bug Fixes

• Fixed cut_tail edge case where reads shorter than the trim window produced an off-by-one in the output length calculation
• Fixed N90 threshold precision: was using integer division for the cumulative length threshold; now uses exact u64 arithmetic

Installation

# Bioconda
conda install -c bioconda biofastq-a

# Build from source
cargo install --path .

---

Have fun using!

by Dila Deniz

---

This discussion was created from the release BioFastq-A v2.3.0.