bqc

CBQ-native all-in-one quality control tool
The Hiller Lab at the Senckenberg Research Institute

binseq . usage . benchmark . us

---

removes adapter sequences, trims, deduplicates, extracts UMIs, and filters sequencing
reads stored in CBQ files. It works natively on CBQ: no FASTQ conversion,
no intermediate file, and every stage runs in one pass over the input.

---

Important

• 📄 CBQ-native: reads CBQ directly.
• ⚡~ 4-8x faster than the field on typical data (fastp, cutadapt, atropos, trimmomatic), with 📉 ~90% less memory.
• 🧬 Reference-free sniff adapters (curated 234-known adapters) and 👃 sniff strandeness for RNA against a Salmon index (given or produced on-the-fly through --transcriptome <PATH>).
• 🔄 Everything in one pass: 3' adapter removal (incl. --allow-indels), quality/positional/homopolymer trim, polyA/G tails, overlap error correction from the mate, internal-adapter split, UMI extraction, filtering — configurable.
• 🔒 Deterministic + safe: single static binary

---

Installation

Binary

cargo install --all-features bqc

Docker

docker pull ghcr.io/hillerlab/bqc:latest

Conda

conda install -c bioconda bqc

---

Benchmarks

See bench.md for full details.

• typical — synthetic, generated by bench/gen.py: 150 bp, A1/A2 adapters with per-base quality decay (phred33), i.e. a realistic "standard adapter-trimming" workload.
• real — the first 2M pairs (head -$((P_REAL*4)) lines) of real sequencing data SRR8997011 (~100 bp), which has no adapters to trim — so it measures raw throughput/overhead on genuine data rather than trimming effect.

dataset	tool	wall s	RSS MB	reads out	bases out (M)	mean len	pairs kept	adapter ppm
typical	bqc	1.59	176	1,957,469	251.7	128.6	1,957,469	0.0
typical	fastp	6.46	1235	1,957,469	251.3	128.4	1,957,469	0.0
typical	cutadapt	3.86	50	1,957,469	251.2	128.3	1,957,469	0.0
typical	atropos	16.65	60	1,957,469	251.2	128.3	1,957,469	0.0
typical	trimmomatic	12.95	1154	1,957,469	251.8	128.7	1,957,469	0.0
real	bqc	1.23	103	1,965,023	196.6	100.0	1,965,023	0.0
real	fastp	5.00	1240	1,828,671	180.9	98.3	1,828,671	0.0
real	cutadapt	3.08	51	1,964,259	196.2	99.7	1,964,259	0.0
real	atropos	15.43	62	1,964,259	196.2	99.7	1,964,259	0.0
real	trimmomatic	7.72	1149	1,921,288 / 1,863,229	188.1 / 181.4	97.6	1,827,052	0.0

UMI and deduplication (2M SE reads, 150 bp, 8 threads)

op	bqc wall s	bqc RSS MB	fastp wall s	fastp RSS MB
UMI (8 bp read1)	0.48	100	1.32	80
dedup 0%	1.05	612	3.21	4204
dedup 50%	2.09	1473	3.61	4192

bqc dedup is exact; fastp's Bloom filter over-deletes (~65–168 unique reads at 10–50% duplication) and pins 4.2 GB regardless of duplication, while bqc memory scales with candidate families (612 MB → 1.47 GB).