bench3c 0.0.2

Synthetic Hi-C / Micro-C / 3C triplet FASTQ benchmark generator and .pairs recovery analyser.

bench3c

bench3c is a synthetic benchmark generator for 3C-derived sequencing workflows. It generates controlled Hi-C-like and Micro-C-like paired-end FASTQ reads containing known triplet structures, then evaluates whether a mapping / splitting / .pairs reconstruction pipeline recovers the expected fragments.

The benchmark is mainly designed to test preprocessing tools for chimeric or multiplex reads in Hi-C, Micro-C, Pore-C-like or split-read 3C workflows.

Model

The benchmark encodes a known triplet of genomic fragments:

R1: AAAAAAABBBBBBBBB
R2: CCCCCCCCCCCCCCCC

The read name stores the true genomic coordinates:

@chrA-startA-endA:chrB-startB-endB::chrC-startC-endC

This allows downstream analysis to compare the expected fragment lengths with the observed alignments recovered in .pairsam files.

Modes

bench3c has three main modes:

• --hic: generate Hi-C-like triplet reads from a digested genome.
• --microc: generate Micro-C-like triplet reads directly from a FASTA.
• --analyse: analyse a .pairsam.gz or .pairsam file and compare recovered alignments to the encoded truth.

Installation

From source:

git clone <repo-url>
cd <repo>
uv sync
uv run bench3c --help

Or with pip after packaging:

pip install bench3c
bench3c --help

Hi-C simulation

Generate Hi-C-like paired-end FASTQ reads:

bench3c --hic \
  --fasta genome.fa \
  --site GATC \
  --out bench/hic_sim \
  --number-reads-pairs 100000 \
  --read-len 150 \
  --min-piece 50 \
  --max-jump 300

If --fasta is not provided, bench3c can generate a random FASTA. If --digested is not provided, the genome is digested internally using --site.

Typical outputs:

bench/hic_sim_R1.fq
bench/hic_sim_R2.fq

Micro-C simulation

Generate Micro-C-like paired-end FASTQ reads:

bench3c --microc \
  --fasta genome.fa \
  --out bench/microc_sim \
  --number-reads-pairs 100000 \
  --read-len 150 \
  --min-piece 50

Add non-chimeric pairs:

bench3c --microc \
  --fasta genome.fa \
  --out bench/microc_mixed \
  --number-reads-pairs 100000 \
  --read-len 150 \
  --prop-nonchimeric 0.2

Analysis mode

Analyse a .pairsam.gz or .pairsam file after mapping and reconstruction:

bench3c --analyse \
  --pairs output.pairs.gz \
  --read-len 150 \
  --analyse-out-dir benchmark_results \
  --condition my_pipeline

The analyser expects read names following the truth-encoding format:

chrA-startA-endA:chrB-startB-endB::chrC-startC-endC

It also expects recoverable alignment information, typically through sam1 and sam2 columns in the .pairs / .pairsam file.

Analysis outputs

The analysis mode writes summary tables and plots, including:

<condition>_read_recovery.tsv
<condition>_problem_fragments.tsv
<condition>_problem_summary.tsv
<condition>_cut_summary.tsv
<condition>_histogram.pdf
<condition>_tolerance_curve.pdf

Depending on the current version, additional outputs may include chimeric-size summaries and chimeric-specific histograms.

Typical benchmark workflow

# 1. Generate synthetic reads
bench3c --microc \
  --fasta genome.fa \
  --out bench/microc \
  --number-reads-pairs 10000 \
  --read-len 150

# 2. Map the reads with your mapper
bwa mem -SP genome.fa bench/microc_R1.fq bench/microc_R2.fq > bench/microc.sam

# 3. Convert the mappings to .pairs with your pipeline

# 4. Analyse fragment recovery
bench3c --analyse \
  --pairs bench/output.pairs.gz \
  --read-len 150 \
  --analyse-out-dir bench/results \
  --condition my_pipeline

Interpretation

A perfect read is a read for which all expected fragments are recovered with no extra observed fragment.

Common failure classes:

• missing: an expected fragment was not recovered.
• too_short: the observed fragment is shorter than the truth.
• too_long: the observed fragment is longer than the truth.
• over_split: extra observed fragments were recovered.
• under_split_or_missing: one or more truth fragments were not recovered.

Limitations

bench3c is a controlled synthetic benchmark. It does not fully model all experimental biases of real Hi-C or Micro-C libraries, such as PCR duplicates, GC bias, mappability bias, restriction efficiency, ligation bias, base-quality degradation, optical duplicates, or complex multi-mapping.

It is intended to test whether a pipeline can recover known chimeric or multiplex structures under controlled conditions.

License

AGPL-3.0-or-later.