archaic-introgression

Files (8)

• SKILL.md
• archaic_introgression.py
• demo_archaic.vcf · examples/
• demo_modern.vcf · examples/
• pipeline.py
• __init__.py · tests/
• test_archaic_introgression.py · tests/
• test_pipeline.py · tests/

SKILL.md (first lines)

# Archaic Introgression Detector

## Trigger

**Fire when:**
- User asks about Neanderthal or Denisovan DNA in modern humans
- User wants to detect archaic introgression segments
- User mentions IBDmix, Sprime, or hmmix methods
- User has modern + archaic VCF files and wants to compare them

**Do NOT fire when:**
- User asks about general ancestry or population structure (use claw-ancestry-pca)
- User asks about pharmacogenomics or clinical variants
- User wants admixture proportions without segment-level detail

## Why This Exists

Between 1-4% of non-African modern human genomes derive from archaic hominins
(Neanderthals, Denisovans). Identifying these segments matters for understanding
human evolution, disease susceptibility, and immune adaptation. Three complementary
methods exist (IBDmix, Sprime, hmmix), each with different strengths. This skill
wraps all three behind a unified interface with a pure-Python fallback when
external binaries are unavailable.

## Core Capabilities

1. **IBDmix**: Detect segments shared IBD between modern and archaic genomes using LOD score thresholds
2. **Sprime**: Identify introgressed haplotypes without requiring an archaic reference panel
3. **hmmix**: HMM-based detection of archaic ancestry tracts
4. **Pure-Python fallback**: LOD-score-based segment calling when IBDmix binary is not installed
5. **EIGENSTRAT support**: Read .ind/.snp/.geno files (text and binary packed formats)
6. **Summary statistics**: Per-individual introgression burden, segment count, mean length

## Scope

One skill, one task: detect and report archaic introgression segments. Does not
perform downstream functional annotation of introgressed variants (chain with
vcf-annotator for that).

## Input Formats

- **Modern genotypes**: VCF (uncompressed or .vcf.gz with .tbi index)
- **Archaic genotypes**: VCF (Neanderthal/Denisovan reference panel)
- **EIGENSTRAT**: .ind + .snp + .geno files (text or binary packed format)

## Workflow

1. Parse modern VCF to extract sample names and genotype matrix (numpy array)
2. Parse archaic VCF to extract reference genotypes at matching positions
3. Identify shared variant positions between modern and archaic panels
4. Run selected method (IBDmix / Sprime / hmmix) or pure-Python LOD fallback
5. Collect IntrogressionSegment results per sample
6. Compute per-individual summary statistics
7. Write JSON report and optional BED file

## CLI Reference

```bash
# Run with IBDmix on VCF inputs
python archaic_introgression.py \
  --input modern.vcf --archaic archaic.vcf \
  --method ibdmix --output /tmp/introgression

# Run demo with synthetic data
python archaic_introgression.py --demo --output /tmp/introgression_demo

# Filter to specific samples
python archaic_introgression.py \
  --input modern.vcf --archaic archaic.vcf \
  --samples SAMPLE01,SAMPLE02 --output /tmp/introgression

# Adjust LOD threshold
python archaic_introgression.py \
  --input modern.vcf --archaic archaic.vcf \
  --lod 5.0 --output /tmp/introgression
```

## Demo

```bash
python archaic_introgression.py --demo --output /tmp/introgression_demo
```

Runs on bundled `examples/demo_modern.vcf` (3 samples, 10 SNPs on chr22) and
`examples/demo_archaic.vcf` (1 Neanderthal sample, same positions).

## Example Queries

- "How much Neanderthal DNA do I have?"
- "Detect archaic introgression in my VCF"
- "Run IBDmix on my genotype data"
- "Show me introgressed segments from Denisova"
- "Compare my genome against the Vindija Neanderthal"

## Output Structure

```
output_dir/
  introgression_results.json   # Full results with segments and summary
  segments.bed                 # BED file of introgressed regions
  reproducibility/
    commands.sh                # Command that reproduces this run
    environment.yml            # Conda environment (python, numpy)
    checksums.sha256           # SHA-256 of outputs; verify with sha256sum -c
```

### JSON structure

```json
{
  "method": "ibdmix",
  "lod_threshold": 3.0,
  "num_samples": 3,
  "segments": [
    {
      "sample": "SAMPLE01",
      "chrom": "chr22",
      "start": 16050075,
      "end": 16051249,

From ClawBio/ClawBio, MIT.

More from ClawBio/ClawBio