sniffcell 0.9.3

SniffCell annotates structural variants using long-read methylation evidence and ctDMR signals.

SniffCell

SniffCell is a Python toolkit for annotating somatic structural variants (SVs) with cell-type origin using long-read DNA methylation. It integrates cell-type-specific differentially methylated regions (ctDMRs) derived from a reference methylation atlas with per-read methylation measurements from nanopore or PacBio long-read BAMs to assign each SV — or every read in a sample — to a cell population.

---

Why SniffCell?

Somatic SVs identified from bulk long-read sequencing are a mixture of events from different cell types. Without knowing the cell of origin, it is difficult to interpret their functional significance or estimate their true variant allele fraction within a specific compartment. SniffCell solves this by reading the epigenetic "fingerprint" imprinted on each DNA molecule and matching it against a reference atlas of cell-type-specific methylation patterns.

Core capabilities:

• ctDMR discovery — Mine a reference methylation atlas to find genomic regions with distinct methylation in each cell population
• Read-level deconvolution — Assign every read in a BAM to a cell type using ctDMR methylation signals, with no single-cell data required
• SV annotation — Link cell-type identity to SV-supporting reads and produce a per-SV cell-of-origin call
• Discovery pipeline — Run a full multi-stage SV / tandem-repeat / SNV calling workflow on cell-type-split BAMs produced by deconvolution
• Interactive reporting — Filter high-confidence SVs and generate an HTML review report with clickable per-SV figures and IGV screenshots

---

Overview

The typical workflow has three main stages:

Atlas (NPY + index + metadata)
        │
        ▼
  sniffcell find         ← Call cell-type-specific DMRs (ctDMRs)
        │
        ▼
  sniffcell anno         ← Extract methylation from BAM, classify reads, assign SVs
        │
        ▼
  sniffcell report       ← Filter high-confidence calls, build HTML review report
        │
        ├── sniffcell viz        ← Per-SV methylation figure (PNG / PDF)
        ├── sniffcell igvviz     ← IGV batch screenshots
        └── sniffcell dmsv       ← Differential methylation test near each SV

For multi-group analyses (e.g., comparing SVs enriched in one cell compartment vs. another):

  sniffcell deconv       ← Deconvolve all reads; split BAM by cell type
        │
        ▼
  sniffcell discover     ← Call SVs / TRs / SNVs independently per group
        │
        ▼
  sniffcell anno         ← Annotate harmonized variants

---

Quick Start

1. Install

pip install sniffcell

For the full environment including bioinformatics tools (Sniffles, bcftools, samtools, Truvari …):

micromamba env create -f environment.yml
micromamba activate sniffcell
pip install sniffcell

See Installation in the wiki for Docker instructions, optional extras, and manual tool setup.

2. Call ctDMRs from the reference atlas

sniffcell find \
  -n atlas/all_celltypes_blocks.npy \
  -i atlas/all_celltypes_blocks.index.gz \
  -cf atlas/index_to_major_celltypes.json \
  -m atlas/all_celltypes.txt \
  -ck pbmc \
  -o pbmc_ctdmr.tsv

3. Annotate SVs with cell-type evidence

sniffcell anno \
  -i sample.bam \
  -v sample.vcf.gz \
  -r ref.fa \
  -b pbmc_ctdmr.tsv \
  -o anno_out \
  -t 8

4. Build the review report

sniffcell report --anno_output anno_out

Open anno_out/report/index.html in a browser to review filtered high-confidence SVs with per-SV methylation evidence.

---

Commands at a Glance

Command	What it does
sniffcell find	Mine a reference atlas to call cell-type-specific DMRs (ctDMRs)
sniffcell anno	Extract read-level methylation from a BAM and assign each SV a cell-type code
sniffcell svanno	Re-score SV assignments from a saved read table without re-processing the BAM
sniffcell deconv	Assign every read in a BAM to a cell type; optionally split into per-group BAMs
sniffcell discover	Multi-stage SV / tandem-repeat / SNV pipeline on cell-type-split BAMs
sniffcell viz	Render a per-SV methylation figure (PNG or PDF)
sniffcell igvviz	Produce IGV batch-mode screenshots for a single SV
sniffcell report	Filter high-confidence SVs and build an interactive HTML review report
sniffcell dmsv	Test for differential methylation between SV-supporting and non-supporting reads

---

Input Requirements

Input	Format	Used by
Long-read alignment	BAM with MM/ML base-modification tags	anno, deconv, dmsv, viz
Structural variants	VCF / VCF.GZ or harmonized TSV from discover	anno, dmsv, viz, report
Reference genome	FASTA + index	anno, deconv, dmsv, viz
ctDMR table	TSV from sniffcell find	anno, deconv, viz
Methylation atlas	NumPy matrix + CpG index + metadata	find

---

Key Outputs

After a complete find → anno → report run, the outputs include:

pbmc_ctdmr.tsv                      ← Cell-type-specific DMRs (input to anno)
anno_out/
  reads_classification.tsv          ← Per-read × ctDMR methylation and cell-type assignment
  sv_assignment.tsv                 ← Per-SV cell-type code and quality metrics
  sv_assignment_readable.tsv        ← Human-readable version with expanded cell-type labels
  anno_run_manifest.json            ← Full run manifest (paths, parameters, versions)
  report/
    index.html                      ← Interactive HTML review report
    high_confidence_sv.tsv          ← Filtered high-confidence SVs
    figures/                        ← Per-SV methylation panels (when --with_figures)

---

Deconvolution and Discovery

For samples where you want to compare SVs across cell populations:

# Optional: plan regional inputs before running targeted deconvolution/discovery
sniffcell regions \
  -b pbmc_ctdmr.tsv \
  --regions loci_of_interest.bed \
  -o regional_plan \
  --regions-ctdmrs 10

This writes regional_plan/subset_regions.bed for downstream subsetting plus ctdmr_subset.tsv, ctdmr_region_summary.tsv, and ctdmr_selected_summary.tsv describing the ctDMRs selected from the sniffcell find output.

# Step 1: Deconvolve reads and split into cell-type-specific BAMs
sniffcell deconv \
  -i sample.bam \
  -r ref.fa \
  -b pbmc_ctdmr.tsv \
  -o deconv_out \
  --split_bam_groups "lymph=t_cell,b_cell,nk_cell;myeloid=monocyte" \
  -t 8

# Step 2: Call SVs, tandem repeats, and SNVs on each group independently
sniffcell discover tools run \
  --deconv-dir deconv_out \
  --reference ref.fa \
  --tr-bed atlas/adotto.v2.trgt.bed \
  --sex female \
  --stages sv,tdb \
  --threads 16

# Step 3: Annotate the harmonized variants
sniffcell anno \
  -i sample.bam \
  -v deconv_out/discover/harmonized_variants.tsv \
  -r ref.fa \
  -b pbmc_ctdmr.tsv \
  -o anno_out

Before running discover, validate your environment:

sniffcell-check-discover --stages all

---

Visualizing Individual SVs

# Minimal — loads all inputs automatically from the anno manifest
sniffcell viz --anno_output anno_out -s sniffles.SV123

# With table exports
sniffcell viz --anno_output anno_out -s sniffles.SV123 --export_tables

# IGV batch screenshot
sniffcell igvviz --anno_output anno_out -s sniffles.SV123

---

Documentation

Full documentation lives in the GitHub Wiki:

Page	Contents
Installation	PyPI, conda, Docker, manual tool setup, verification
End-to-End Workflow	Step-by-step walkthrough from atlas to HTML report
Find Workflow	ctDMR discovery internals and parameter guide
Methods	Technical methods for deconv, discover, and anno
Test Examples	Practical validation and QA queries

---

Citation

If you use SniffCell in your research, please cite:

SniffCell: cell-type annotation of somatic structural variants using long-read methylation Yilei Fu et al. (manuscript in preparation)

---

License

MIT License — see LICENSE for details.

Developed at Baylor College of Medicine by Yilei Fu.