trace-crispr 0.6.1

TRACE: Triple-aligner Read Analysis for CRISPR Editing

TRACE

Triple-aligner Read Analysis for CRISPR Editing

Robust quantification of CRISPR editing outcomes from amplicon sequencing data using consensus alignment across BWA-MEM, BBMap, and minimap2.

Quick Start

# Install
pip install trace-crispr

# Run on a single sample
trace run \
  --reference amplicon.fasta \
  --hdr-template donor.fasta \
  --guide GCTGAAGCACTGCACGCCGT \
  --r1 sample_R1.fastq.gz \
  --r2 sample_R2.fastq.gz \
  --output results/

# Run on multiple samples
trace run \
  --reference amplicon.fasta \
  --hdr-template donor.fasta \
  --guide GCTGAAGCACTGCACGCCGT \
  --sample-key samples.tsv \
  --output results/ \
  --threads 16

Output

Editing Summary

TRACE produces a comprehensive per-sample summary table with editing outcomes:

sample     total_reads  HDR_COMPLETE_%  HDR_PARTIAL_%  NHEJ_INDEL_%  MMEJ_INDEL_%  WT_%
──────────────────────────────────────────────────────────────────────────────────────────
sample_01  125,432      42.1            3.1            8.2           4.1           38.1
sample_02  98,765       48.7            3.4            5.6           3.1           35.8
sample_03  112,890      45.2            3.7            9.8           5.4           31.2

Classification Categories

TRACE classifies reads into 12 comprehensive categories:

HDR Categories (successful donor integration):

Outcome	Description
HDR_COMPLETE	All donor-encoded edits present, no other modifications
HDR_PARTIAL	Subset of donor SNVs integrated, no other modifications
HDR_PLUS_NHEJ_INDEL	Donor edits + classical NHEJ indel at cut site (0-2bp microhomology)
HDR_PLUS_MMEJ_INDEL	Donor edits + MMEJ indel at cut site (>2bp microhomology)
HDR_PLUS_OTHER	Donor edits + modifications NOT at cut site

Non-HDR Repair Outcomes:

Outcome	Description
DONOR_CAPTURE	Donor edits present + extra donor sequence duplicated at site (over-integration)
NHEJ_INDEL	Classical NHEJ indel at cut site (0-2bp microhomology at deletion boundaries)
MMEJ_INDEL	Microhomology-mediated indel (>2bp microhomology at deletion boundaries)

Other Outcomes:

Outcome	Description
WT	Wild-type / unedited
NON_DONOR_SNV	SNVs not matching donor template, no indels
UNCLASSIFIED	Does not fit other categories
UNMAPPED	Read did not align to reference

Note: LARGE_DELETION (≥50bp) is a flag applied to reads within the above categories, not a separate category.

NHEJ vs MMEJ Classification

TRACE distinguishes between two types of error-prone repair based on deletion microhomology:

• NHEJ (Non-Homologous End Joining): Deletions with 0-2bp of microhomology at the deletion boundaries. This is classical error-prone repair.
• MMEJ (Microhomology-Mediated End Joining): Deletions with >2bp of microhomology. This alternative repair pathway uses short homologous sequences to rejoin the break.

This distinction is biologically significant as MMEJ and NHEJ involve different repair proteins and have different mutational signatures.

Output Columns

The main output file (per_sample_editing_outcomes_all_methods.tsv) contains:

Column	Description
sample	Sample identifier
total_reads	Total reads processed
aligned_reads	Reads that aligned to reference
classifiable_reads	Reads that could be classified
duplicate_rate	Fraction of reads removed as duplicates
HDR_COMPLETE_%	Percentage with complete HDR
HDR_PARTIAL_%	Percentage with partial HDR
HDR_PLUS_NHEJ_%	HDR + classical NHEJ indel
HDR_PLUS_MMEJ_%	HDR + microhomology-mediated indel
HDR_PLUS_OTHER_%	HDR + other modifications
HDR_total_%	Sum of all HDR categories
DONOR_CAPTURE_%	Donor over-integration
NHEJ_INDEL_%	Classical NHEJ only
MMEJ_INDEL_%	MMEJ only
NHEJ_MMEJ_total_%	Combined NHEJ + MMEJ rate
WT_%	Wild-type percentage
NON_DONOR_SNV_%	Non-donor SNVs
UNCLASSIFIED_%	Unclassified reads
Edited_total_%	All edited reads (excludes WT)

Pooled Summary (Technical Replicates)

For experiments with technical replicates, use trace aggregate to generate pooled statistics:

bio_sample  quality_flag  n_reps  hdr_total_pct_mean  hdr_total_pct_sem  nhej_mmej_total_pct_mean
────────────────────────────────────────────────────────────────────────────────────────────────────
gene_A      good          3       45.2                1.3                12.1
gene_B      good          3       52.8                0.9                8.4
gene_C      good          3       38.1                2.1                18.7

The quality_flag column indicates:

• good: All replicates passed QC
• one_low_read_removed: One replicate had low reads and was excluded
• all_low_reads: All replicates had low read counts (use with caution)

How It Works

Triple-Aligner Consensus

TRACE uses three independent aligners to maximize accuracy:

                    ┌─────────────┐
     Raw Reads ────►│   BWA-MEM   │────┐
                    └─────────────┘    │
                    ┌─────────────┐    │     ┌───────────────┐     ┌────────────┐
     Raw Reads ────►│   BBMap     │────┼────►│   Consensus   │────►│  Classify  │
                    └─────────────┘    │     └───────────────┘     └────────────┘
                    ┌─────────────┐    │
     Raw Reads ────►│  minimap2   │────┘
                    └─────────────┘

Each aligner has different strengths—BWA-MEM for accuracy, BBMap for gapped alignments, minimap2 for speed. TRACE takes the consensus to reduce aligner-specific artifacts.

Edit Distance Classification

TRACE classifies reads by measuring edit distance to both reference and donor sequences:

1. Align read to reference amplicon
2. Extract mismatches in core edit region (±30bp from cut site)
3. Check if each mismatch brings sequence closer to donor template
4. Classify based on SNV pattern and presence of indels

This handles complex cases where aligners represent clustered SNVs as indels.

Automatic Detection

TRACE auto-detects library characteristics:

Detection	Method
Library type	TruSeq (fixed primers) vs Tn5 (tagmented) based on read start clustering
UMI presence	Sequence diversity analysis at read starts
Read overlap	Determines if R1/R2 can be merged
PCR duplicates	UMI-based (TruSeq) or position-based (Tn5) deduplication

Installation

pip (Python package only)

pip install trace-crispr

conda (includes aligners)

conda install -c bioconda -c conda-forge trace-crispr

This installs BWA, BBMap, minimap2, and samtools automatically.

Development

git clone https://github.com/k-roy/TRACE.git
cd TRACE
pip install -e ".[dev]"

Usage

Check Locus Configuration

Before running, verify TRACE correctly detects your editing design:

trace info \
  --reference amplicon.fasta \
  --hdr-template donor.fasta \
  --guide GCTGAAGCACTGCACGCCGT

Output:

============================================================
TRACE Analysis Configuration
============================================================

Reference sequence: 231 bp
HDR template: 127 bp
  - Template aligns at position 53 in reference

Donor template analysis:
  - Left homology arm: positions 53-124 (72 bp)
  - Right homology arm: positions 128-179 (52 bp)

  Edits detected (2 total):
    * Position 125: T -> A (substitution)
    * Position 127: G -> A (substitution)

Guide analysis:
  - Guide: GCTGAAGCACTGCACGCCGT
  - Target: positions 105-124 (+ strand)
  - PAM: TGG at positions 125-127
  - Cleavage site: position 122

Sample Key Format

Create a TSV file with sample information:

sample_id    r1_path                  r2_path                  condition
sample_01    /path/to/S1_R1.fastq.gz  /path/to/S1_R2.fastq.gz  treatment
sample_02    /path/to/S2_R1.fastq.gz  /path/to/S2_R2.fastq.gz  treatment
sample_03    /path/to/S3_R1.fastq.gz  /path/to/S3_R2.fastq.gz  control

Per-Sample Sequences

For experiments with different guides/donors per sample:

sample_id  r1_path      r2_path      reference      guide                   hdr_template
sample_01  S1_R1.fq.gz  S1_R2.fq.gz  locus1.fasta   AGAGAAACACACTGTACTCCGT  donor1.fasta
sample_02  S2_R1.fq.gz  S2_R2.fq.gz  locus2.fasta   TTGGTTACAACTCTGACCCA    donor2.fasta

trace run --sample-key manifest.tsv --output results/

Multi-Template Analysis (Barcode Screening)

For experiments with multiple possible HDR templates:

trace multi-template \
  --reference amplicon.fasta \
  --hdr-templates all_barcodes.fasta \
  --guide GAGTCCGAGCAGAAGAAGAA \
  --sample-key samples.tsv \
  --output results/

Cas12a Support

trace run \
  --reference amplicon.fasta \
  --hdr-template donor.fasta \
  --guide GCTGAAGCACTGCACGCCGTAA \
  --nuclease cas12a \
  --sample-key samples.tsv \
  --output results/

Nuclease	PAM	Cleavage
Cas9	NGG (3' of guide)	Blunt, 3bp from PAM
Cas12a	TTTN (5' of guide)	Staggered, 18-23bp from PAM

Snakemake Workflow

For large-scale processing:

snakemake --configfile config.yaml --cores 24

See workflow/README.md for details.

CLI Commands Reference

TRACE provides the following commands:

Command	Description
trace run	Main analysis pipeline for single or multiple samples
trace info	Display locus configuration (reference, donor, guide alignment)
trace classify	Re-classify reads from an existing BAM file
trace classify-batch	Batch re-classification on multiple BAM files
trace multi-template	Analyze samples with multiple possible HDR templates
trace aggregate	Aggregate results across technical replicates
trace extract-kmers	Extract k-mers for contamination filtering
trace generate-manifest	Generate a sample manifest from a directory of FASTQs
trace generate-templates	Generate multi-template reference from barcode list
trace init	Create a template configuration file

Use trace <command> --help for detailed options.

Example: Aggregating Replicates

# After running trace on all samples
trace aggregate \
  --input results/per_sample_editing_outcomes_all_methods.tsv \
  --sample-key samples.tsv \
  --group-by condition \
  --output results/aggregated_by_condition.tsv

Analysis Module

Compare editing outcomes across conditions with statistical testing:

from trace_crispr.analysis import (
    compare_metric_by_condition,
    results_to_dataframe,
    plot_condition_comparison,
)

# Compare HDR rates
comparisons = compare_metric_by_condition(
    results, samples,
    condition_col='treatment',
    metric='dedup_hdr_pct',
    base_condition='control'
)

# View statistics
print(comparisons.to_dataframe())

     condition        metric  mean   std   p_value  significance
0  treatment_A  dedup_hdr_pct  25.4  2.63   0.0003          ***
1  treatment_B  dedup_hdr_pct  12.1  1.89   0.4521           ns

Visualization

fig = plot_condition_comparison(
    stats, comparisons,
    base_condition='control',
    title='HDR Rate by Treatment',
    ylabel='HDR Rate (%)'
)
fig.savefig('hdr_comparison.png', dpi=150)

Install visualization dependencies:

pip install trace-crispr[visualization]

Dependencies

Python (core): click, pysam, pandas, numpy, pyyaml, rapidfuzz, tqdm

Python (visualization): matplotlib, seaborn, scipy

External (via conda): bwa, bbmap, minimap2, samtools

Citation

If you use TRACE in your research, please cite:

Roy, K.R. et al. (2026). TRACE: Triple-aligner Read Analysis for CRISPR Editing. In preparation.

Author

Kevin R. Roy ([e-mail])

License

MIT