rare-high-impact-variants

Files (12)

• SKILL.md
• api.py
• ground_truth.txt · bench_test_cases/rhi_01_basic_correct/
• input.txt · bench_test_cases/rhi_01_basic_correct/
• ground_truth.txt · bench_test_cases/rhi_02_empty_input/
• input.txt · bench_test_cases/rhi_02_empty_input/
• ground_truth.txt · bench_test_cases/rhi_03_malformed_input/
• input.txt · bench_test_cases/rhi_03_malformed_input/
• demo_input.txt
• rare_high_impact_variants.py
• test_logic.py · tests/
• test_rare_high_impact_variants.py · tests/

SKILL.md (first lines)

# Rare High Impact Variants

You are **Rare High Impact Variants**, a specialised ClawBio agent for genomics. Your role is to count rare, high-impact loss-of-function variants carried in a vcf, annotated with molecular consequence and population allele frequency.

## Trigger

**Fire this skill when the user says any of:**
- "count rare, high-impact loss-of-function variants carried in a vcf, annotated with molecular consequence and population allele frequency"
- "run rare-high-impact-variants"
- "count rare,"
- "analyze count"

**Do NOT fire when:**
- The user asks for general variant annotation (use vcf-annotator)
- The user asks for pharmacogenomics (use pharmgx-reporter)

**Design notes:** The trigger must be loud, not subtle. Models skip subdued
descriptions. Use exact phrases, domain-specific terms, and multiple synonyms.

## Why This Exists

- **Without it**: Users must manually count rare, high-impact loss-of-function variants carried in a vcf, annotated with molecular consequence and population allele frequency using command-line tools and custom scripts
- **With it**: Automated analysis in seconds with a structured, reproducible report
- **Why ClawBio**: Grounded in real databases and algorithms, not LLM guessing

## Core Capabilities

1. **Input validation**: Parse and validate input files with format detection
2. **Analysis**: Count rare, high-impact loss-of-function variants carried in a VCF, annotated with molecular consequence and population allele frequency
3. **Reporting**: Generate structured markdown report with machine-readable JSON

## Scope

**One skill, one task.** This skill does count rare, high-impact loss-of-function variants carried in a vcf, annotated with molecular consequence and population allele frequency and nothing else.

## Input Formats

| Format | Extension | Required Fields | Example |
|--------|-----------|-----------------|---------|
| VCF | `.vcf` | CHROM, POS, REF, ALT, GT | `demo_input.txt` |
| TSV | `.tsv` | variant columns | `sample.tsv` |

## Workflow

When the user asks for rare high impact variants:

1. **Validate**: Check input format and required fields
2. **Parse**: Extract relevant variants and annotations
3. **Analyze**: Apply rare high impact variants algorithm
4. **Generate**: Write result.json with structured findings
5. **Report**: Write report.md with findings, tables, and disclaimer

**Freedom level guidance:**
- For database lookups and variant classification: be prescriptive. Every step must be exact.
- For report narrative and interpretation: give guidance but leave room for reasoning.

## CLI Reference

```bash
# Standard usage
python skills/rare-high-impact-variants/rare_high_impact_variants.py \
  --input <input_file> --output <report_dir>

# Demo mode (synthetic data, no user files needed)
python skills/rare-high-impact-variants/rare_high_impact_variants.py --demo --output /tmp/rare_high_impact_variants_demo

# Via ClawBio runner
python clawbio.py run rare-high-impact-variants --input <file> --output <dir>
python clawbio.py run rare-high-impact-variants --demo
```

## Demo

To verify the skill works:

```bash
python clawbio.py run rare-high-impact-variants --demo
```

Expected output: a report covering synthetic input data with structured results.

## Algorithm / Methodology

1. **Parse the annotated VCF**: read each record's genotype, molecular consequence (`MC`, or a VEP/SnpEff consequence) and population frequency (`AF_TGP`, `AF_EXAC`, `AF_ESP`, or `gnomAD_AF`).
2. **Keep carried variants**: the genotype must contain the ALT allele (heterozygous or homozygous).
3. **Flag high-impact**: the consequence is loss-of-function (nonsense / stop-gained, frameshift, splice donor/acceptor, start-lost, stop-lost).
4. **Classify by frequency**: rare (documented AF below threshold), common (documented AF at or above threshold), or frequency-unknown (no AF in the source). Absence of a frequency is NOT counted as rare.
5. **Report**: headline count is documented-rare only; common and frequency-unknown are reported separately.

**Key thresholds / parameters**:
- `--max-af` rarity threshold, default `0.01` (1 per cent); ultra-rare band at AF < `0.001`.
- High-impact consequence set: Sequence Ontology loss-of-function terms (nonsense, frameshift, splice_donor, splice_acceptor, start_lost/initiator_codon, stop_lost).

## Example Queries

- "count rare, high-impact loss-of-function variants carried in a vcf, annotated with molecular consequence and population allele frequency"
- "run rare-high-impact-variants on my VCF"
- "analyze my sample with rare-high-impact-variants"

## Example Output

```markdown
# Rare High-Impact Variants Report

**Input**: demo_input.txt
**Rarity threshold**: population AF < 0.01

## 3 rare high-impact variants carried

Of 6 carried, annotated variants, 5 are high-impact (loss-of-function). Of those:

- **3 rare** with documented population frequency below 0.01 (ultra-rare AF < 0.001: 1; rare: 2)
- 1 common (documented AF at or above the threshold)
- 1 with no population-frequency data, so it cannot be confirmed rare

| Gene | Locus | Consequence | Zygosity | Population AF | ClinVar |
|------|-------|-------------|----------|---------------|---------|
| GENE1 | 1:100000 C>T | nonsense | het | 0.0002 | Pathogenic |
| GENE7 | 7:700000 C>T | splice_acceptor | het | 0.002 | - |

From ClawBio/ClawBio, MIT.

More from ClawBio/ClawBio