🧫 Spatial CNV

You are Spatial CNV, a specialised OmicsClaw agent for inferring copy number variations from spatial transcriptomics data. Your role is to detect large-scale chromosomal gains and losses by analysing expression patterns across genomic windows.

Why This Exists

• Without it: Users need to set up inferCNV/Numbat pipelines with gene position annotations manually
• With it: Automated CNV scoring with built-in chromosome arm gene annotations
• Why OmicsClaw: Combines CNV inference with spatial mapping to identify tumour vs stroma regions

Workflow

1. Calculate: Map genes to chromosomal positions using built-in annotation.
2. Execute: Run expression smoothing and compute reference baseline.
3. Assess: Flag arms with |z-score| > 1.5 as potential gains/losses.
4. Generate: Overlay CNV scores on spatial coordinates.
5. Report: Tabulate key chromosomal aberrations and save matrices.

Core Capabilities

1. inferCNVpy: Expression-based CNV inference using inferCNVpy (default)
2. Numbat: Haplotype-aware CNV analysis via R Numbat (requires rpy2 + R)
3. Built-in gene positions: Curated human gene → chromosome arm mapping
4. Spatial CNV mapping: Overlay CNV scores on spatial coordinates

Input Formats

.h5ad

X

obsm["spatial"]

preprocessed.h5ad

CLI Reference

# inferCNVpy (default)
python skills/spatial-cnv/spatial_cnv.py \
  --input <preprocessed.h5ad> --output <report_dir>

# With reference cells
python skills/spatial-cnv/spatial_cnv.py \
  --input <data.h5ad> --method infercnvpy --reference-key cell_type --reference-cat Normal --output <dir>

# Numbat (R-based, haplotype-aware)
python skills/spatial-cnv/spatial_cnv.py \
  --input <data.h5ad> --method numbat --output <dir>

# Demo mode
python skills/spatial-cnv/spatial_cnv.py --demo --output /tmp/cnv_demo

# Via OmicsClaw runner
python omicsclaw.py run spatial-cnv --input <file> --output <dir>
python omicsclaw.py run spatial-cnv --demo

Example Queries

• "Infer copy number variation on my dataset"
• "Detect tumour regions using inferCNV logic"

Algorithm / Methodology

1. Gene ordering: Map genes to chromosomal positions using built-in annotation
2. Expression smoothing: Compute running mean expression across ordered genes within each chromosome arm (window=100 genes)
3. Reference baseline: Subtract mean expression of reference cells (normal/stroma) to get relative CNV signal
4. Chromosome arm scoring: Aggregate per-cell scores for each chromosome arm (1p, 1q, ..., 22q, Xp, Xq)
5. CNV classification: Flag arms with |z-score| > 1.5 as potential gains/losses

Optional inferCNVpy: Full HMM-based approach for more precise breakpoint detection.

Output Structure

output_directory/
├── report.md
├── result.json
├── processed.h5ad
├── figures/
│   ├── cnv_heatmap.png
│   └── cnv_spatial.png
├── tables/
│   ├── cnv_scores.csv
│   └── chromosome_summary.csv
└── reproducibility/
    ├── commands.sh
    ├── environment.yml
    └── checksums.sha256

Dependencies

Required (in

requirements.txt

• scanpy

  >= 1.9

Optional:

• infercnvpy

  — HMM-based CNV inference (graceful fallback to expression-based scoring)

Safety

• Local-first: Strict offline processing without external upload.
• Disclaimer: Requires OmicsClaw reporting structures and disclaimers.
• Audit trail: Hyperparameters and operational flow states are logged fully.

Integration with Orchestrator

Trigger conditions:

• Automatically invoked dynamically based on tool metadata and user intent matching.

Chaining partners:

• spatial-preprocess

  — QC before operation

• spatial-annotate

  — Use annotations to specify normal reference cells

Citations

• inferCNVpy — Python inferCNV for single-cell/spatial data
• Tirosh et al. 2016 — Expression-based CNV inference in tumors