drug-discovery-prompts/upstream/mims-harvard-ToolUniverse/skills/tooluniverse-variant-analysis/references/sv_cnv_analysis.md

title, task, lineage_type, upstream_source, upstream_sha, imported_at, prompt_class, upstream_changes, author, validated

title	task	lineage_type	upstream_source	upstream_sha	imported_at	prompt_class	upstream_changes	author	validated
Structural Variant and CNV Analysis Guide		import	https://github.com/mims-harvard/ToolUniverse/blob/e2520a96/skills/tooluniverse-variant-analysis/references/sv_cnv_analysis.md	e2520a96	2026-06-26	prompt	accepted	upstream	false

Structural Variant and CNV Analysis Guide

Complete workflow for interpreting structural variants (deletions, duplications, inversions, translocations) and copy number variants using ToolUniverse.

When to Use This Workflow

• VCF contains SV calls (from Manta, Delly, LUMPY, etc.)
• BED file with CNV regions
• Coordinate-based SV queries (chr17:43044295-43070295)
• Gene-based SV queries ("What SVs affect BRCA1?")
• Clinical interpretation of CNV pathogenicity

SV Analysis Workflow

Input: SV VCF or coordinates
    |
    v
Step 1: Identify Structural Variants
    |-- Detect SVTYPE (DEL, DUP, INV, BND, CNV)
    |-- Extract coordinates (chrom, start, end)
    |-- Determine size
    |
    v
Step 2: Annotate with Population Frequencies
    |-- Query gnomAD SV by gene or region
    |-- Get allele frequency (AF), allele count (AC), allele number (AN)
    |-- Interpret frequency (common vs rare)
    |
    v
Step 3: Query Known SVs from DGVa/dbVar
    |-- Use Ensembl to find reported SVs
    |-- Get clinical significance
    |
    v
Step 4: Identify Affected Genes
    |-- From gnomAD response
    |-- From VCF annotation
    |-- From coordinate overlap
    |
    v
Step 5: Query ClinGen Dosage Sensitivity
    |-- For each affected gene
    |-- Get haploinsufficiency (HI) score
    |-- Get triplosensitivity (TS) score
    |
    v
Step 6: Classify SV Pathogenicity
    |-- Combine evidence (dosage + frequency + literature)
    |-- Apply ACMG/ClinGen CNV guidelines
    |-- Classify: Pathogenic / Likely Pathogenic / VUS / Benign
    |
    v
Step 7: Generate SV Clinical Report
    |-- Summary (type, coordinates, size, frequency)
    |-- Affected genes (with dosage scores)
    |-- Clinical interpretation
    |-- Recommendations

Step 1: Identify Structural Variants

From VCF

SV callers (Manta, Delly, LUMPY) annotate SVs in VCF INFO field:

VCF example:

chr17  43044295  DEL_chr17_1  N  <DEL>  100  PASS  SVTYPE=DEL;END=43070295;SVLEN=-26000

Detect SVs:

• Check for SVTYPE in INFO (DEL, DUP, INV, BND, CNV)
• Check for large indels (|len(REF) - len(ALT)| > 50bp)
• Extract END coordinate
• Calculate size from SVLEN or (END - POS)

From Coordinates

User provides coordinates: "chr17:43044295-43070295"

Parse format:

import re

coord_str = "chr17:43044295-43070295"
match = re.match(r'chr(\w+):(\d+)-(\d+)', coord_str)
chrom = match.group(1)  # "17"
start = int(match.group(2))  # 43044295
end = int(match.group(3))    # 43070295
size = end - start           # 26000 bp

Step 2: Query gnomAD SV Population Frequencies

Query by Gene Symbol

When to use: When you know which gene is affected

from tooluniverse import ToolUniverse

tu = ToolUniverse()
tu.load_tools()

result = tu.execute_tool("gnomad_get_sv_by_gene", gene_symbol="BRCA1")

Response structure:

{
  "data": {
    "structural_variants": [
      {
        "variant_id": "DEL_chr17_24e4872b",
        "chrom": "17",
        "pos": 43044295,
        "end": 43070295,
        "sv_type": "DEL",
        "size": 26000,
        "af": 0.000008,
        "ac": 3,
        "an": 375000,
        "filters": ["PASS"],
        "genes": ["BRCA1"]
      }
    ]
  }
}

Query by Genomic Region

When to use: When you have coordinates but not gene

result = tu.execute_tool(
    "gnomad_get_sv_by_region",
    chrom="17",
    start=43044295,
    end=43070295
)

Query Specific SV Details

When to use: When you have a gnomAD SV ID

result = tu.execute_tool("gnomad_get_sv_detail", sv_id="DEL_chr17_24e4872b")

Response includes:

• Quality metrics (QUAL, GQ)
• Filter status
• Detailed population frequencies

Frequency Interpretation

gnomAD AF	Interpretation	Clinical Impact
AF > 0.01 (1%)	Common benign variant	Likely benign
AF 0.001-0.01	Low-frequency variant	Uncertain, context-dependent
AF < 0.001 (0.1%)	Rare variant	Potentially pathogenic if affects critical gene
Not in gnomAD	Very rare or novel	Unknown, requires additional evidence

Step 3: Query Known SVs from DGVa/dbVar

Use Ensembl to find previously reported SVs:

Query by Region

result = tu.execute_tool(
    "ensembl_get_structural_variants",
    chrom="17",
    start=43044295,
    end=43070295,
    species="human"
)

Response structure:

{
  "data": [
    {
      "id": "esv3647486",
      "var_class": "copy_number_variation",
      "clinical_significance": ["pathogenic", "likely_pathogenic"],
      "evidence": ["1000 Genomes", "dbVar"],
      "study": "estd199",
      "sample_count": 4
    }
  ]
}

Get SV Details

result = tu.execute_tool(
    "ensembl_get_sv_detail",
    sv_id="esv3647486",
    species="human"
)

Step 4: Identify Affected Genes

From gnomAD Response

gnomAD SV response includes genes field:

result = tu.execute_tool("gnomad_get_sv_by_region", chrom="17", start=43044295, end=43070295)
sv = result['data']['structural_variants'][0]
affected_genes = sv['genes']  # ["BRCA1"]

From VCF Annotation

Some SV callers annotate genes in INFO:

SVTYPE=DEL;END=43070295;GENES=BRCA1,NBR2

From Coordinate Overlap

If genes not provided, query by coordinates to find overlapping genes.

Step 5: Query ClinGen Dosage Sensitivity

For each affected gene, determine if it's dosage-sensitive.

Query by Gene Symbol

result = tu.execute_tool("ClinGen_dosage_by_gene", gene_symbol="BRCA1")

Response structure:

{
  "data": {
    "gene_symbol": "BRCA1",
    "haploinsufficiency_score": 3,
    "triplosensitivity_score": 0,
    "dosage_sensitivity_disease": "Breast-ovarian cancer, familial 1",
    "haploinsufficiency_description": "Sufficient evidence for dosage pathogenicity",
    "triplosensitivity_description": "No evidence available",
    "pli_score": 0.00,
    "loeuf": 0.46,
    "acmg_secondary_findings": true
  }
}

Query by Region

Find all dosage-sensitive genes in a region:

result = tu.execute_tool(
    "ClinGen_dosage_region_search",
    chromosome="17",
    start=43044295,
    end=43070295
)

Response structure:

{
  "data": {
    "genes": [
      {
        "gene_symbol": "BRCA1",
        "haploinsufficiency_score": 3,
        "triplosensitivity_score": 0
      }
    ]
  }
}

Dosage Score Interpretation

Score	HI Meaning	TS Meaning	Clinical Impact
3	Sufficient evidence	Sufficient evidence	HIGH - Established dosage sensitivity
2	Some evidence	Some evidence	MODERATE - Emerging evidence
1	Little evidence	Little evidence	LOW - Minimal evidence
0	No evidence	No evidence	MINIMAL - No known dosage effect
30	Gene associated with AD condition	-	HIGH (alternative scoring)
40	Dosage sensitivity unlikely	Dosage sensitivity unlikely	MINIMAL

Step 6: Classify SV Pathogenicity

Combine evidence to classify using ACMG/ClinGen CNV guidelines:

Pathogenic Criteria

Report as disease-causing if:

• Deletion + HI score = 3 (sufficient evidence for haploinsufficiency)
• Duplication + TS score = 3 (sufficient evidence for triplosensitivity)
• gnomAD AF < 0.0001 (very rare)
• Gene is in ACMG Secondary Findings list
• Known pathogenic SV from ClinVar/dbVar with same breakpoints

Likely Pathogenic Criteria

• Deletion + HI score = 2 (some evidence)
• Duplication + TS score = 2 (some evidence)
• gnomAD AF < 0.001
• Overlaps known pathogenic region but different breakpoints
• Disrupts critical functional domain

Uncertain Significance (VUS) Criteria

• HI/TS score = 0 or 1 (no/little evidence)
• gnomAD AF 0.001-0.01 (conflicting frequency data)
• Affects gene but unclear clinical relevance
• Novel SV not in databases

Likely Benign Criteria

• gnomAD AF > 0.01 (common in population)
• HI/TS score = 40 (dosage sensitivity unlikely)
• Does not affect any coding genes
• Matches known benign SV

Benign Criteria

• gnomAD AF > 0.05 (very common)
• No overlap with any genes or regulatory regions
• Established benign variant in ClinVar

Step 7: Generate SV Clinical Report

Report Template

# Structural Variant Clinical Report

## Variant Summary
- **Type**: Deletion (DEL)
- **Location**: chr17:43,044,295-43,070,295 (hg38)
- **Size**: 26 kb
- **Gene(s) Affected**: BRCA1 (complete deletion)

## Population Frequency
- **gnomAD AF**: 0.000008 (8 in 1 million)
- **Allele Count**: 3 / 375,000
- **Filter**: PASS
- **Interpretation**: Very rare variant

## ClinGen Dosage Sensitivity
- **Gene**: BRCA1
- **Haploinsufficiency Score**: 3 (Sufficient evidence)
- **Triplosensitivity Score**: 0 (No evidence)
- **Disease**: Breast-ovarian cancer, familial 1 (OMIM #604370)
- **ACMG Secondary Findings**: Yes

## Clinical Significance
**Classification**: PATHOGENIC

**Evidence**:
1. Complete deletion of BRCA1 gene
2. ClinGen HI score = 3 (established haploinsufficiency)
3. BRCA1 is ACMG Secondary Findings gene
4. Very rare in population (AF < 0.0001)
5. Known mechanism for hereditary breast/ovarian cancer

**ACMG/ClinGen Criteria Met**:
- Loss-of-function mechanism established (PVS1-equivalent)
- Haploinsufficiency score = 3 (dosage-sensitive gene)
- Absence/rarity in population databases

## Clinical Recommendations
1. **Genetic Counseling**: Strongly recommended
2. **Cancer Risk Management**: Enhanced surveillance per NCCN guidelines
   - Annual breast MRI starting age 25-30
   - Consider risk-reducing mastectomy
   - Consider risk-reducing salpingo-oophorectomy age 35-40
3. **Family Testing**: Cascade testing for first-degree relatives
4. **Confirmation**: MLPA or chromosomal microarray

Complete Example Workflow

Example 1: BRCA1 Deletion

Input: User asks "What is the clinical significance of a deletion affecting BRCA1?"

from tooluniverse import ToolUniverse

tu = ToolUniverse()
tu.load_tools()

# Step 1: Query gnomAD SVs by gene
gnomad_result = tu.execute_tool("gnomad_get_sv_by_gene", gene_symbol="BRCA1")

# Extract first deletion
svs = gnomad_result['data']['structural_variants']
deletions = [sv for sv in svs if sv['sv_type'] == 'DEL']

if deletions:
    sv = deletions[0]
    print(f"Found deletion: {sv['variant_id']}")
    print(f"  Location: chr{sv['chrom']}:{sv['pos']}-{sv['end']}")
    print(f"  Size: {sv['size']} bp")
    print(f"  AF: {sv['af']}")

    # Step 2: Query ClinGen dosage sensitivity
    clingen_result = tu.execute_tool("ClinGen_dosage_by_gene", gene_symbol="BRCA1")
    dosage = clingen_result['data']

    print(f"\nClinGen Dosage Sensitivity:")
    print(f"  HI Score: {dosage['haploinsufficiency_score']}")
    print(f"  TS Score: {dosage['triplosensitivity_score']}")
    print(f"  Disease: {dosage['dosage_sensitivity_disease']}")
    print(f"  ACMG Secondary Findings: {dosage['acmg_secondary_findings']}")

    # Step 3: Classify pathogenicity
    if sv['af'] < 0.0001 and dosage['haploinsufficiency_score'] == 3:
        classification = "PATHOGENIC"
    elif sv['af'] < 0.001 and dosage['haploinsufficiency_score'] >= 2:
        classification = "LIKELY PATHOGENIC"
    elif sv['af'] > 0.01:
        classification = "LIKELY BENIGN"
    else:
        classification = "UNCERTAIN SIGNIFICANCE"

    print(f"\nClassification: {classification}")

Example 2: Regional CNV Analysis

Input: User provides coordinates "chr17:43044295-43070295"

# Query region for SVs
gnomad_result = tu.execute_tool(
    "gnomad_get_sv_by_region",
    chrom="17",
    start=43044295,
    end=43070295
)

# Query region for dosage-sensitive genes
clingen_result = tu.execute_tool(
    "ClinGen_dosage_region_search",
    chromosome="17",
    start=43044295,
    end=43070295
)

# Analyze overlap
svs = gnomad_result['data']['structural_variants']
dosage_genes = clingen_result['data']['genes']

print(f"Found {len(svs)} SVs in region")
print(f"Found {len(dosage_genes)} dosage-sensitive genes")

for gene in dosage_genes:
    if gene['haploinsufficiency_score'] >= 2:
        print(f"\n{gene['gene_symbol']} is haploinsufficient (HI={gene['haploinsufficiency_score']})")
        print("  Deletions affecting this gene are likely pathogenic")

Example 3: Batch SV Interpretation

Input: VCF with multiple SV calls

# Parse SV VCF (pseudocode - actual implementation in python_implementation.py)
def interpret_sv_batch(sv_vcf_path):
    results = []

    # Parse SV VCF
    svs = parse_sv_vcf(sv_vcf_path)

    for sv in svs:
        # Get population frequency
        gnomad = tu.execute_tool(
            "gnomad_get_sv_by_region",
            chrom=sv.chrom,
            start=sv.start,
            end=sv.end
        )

        # Identify affected genes
        genes = sv.genes  # From VCF annotation or coordinate overlap

        # Query dosage sensitivity for each gene
        dosage_results = []
        for gene in genes:
            clingen = tu.execute_tool("ClinGen_dosage_by_gene", gene_symbol=gene)
            dosage_results.append(clingen['data'])

        # Classify pathogenicity
        classification = classify_sv_pathogenicity(
            sv_type=sv.type,
            gnomad_af=gnomad['data']['structural_variants'][0]['af'] if gnomad['data']['structural_variants'] else None,
            dosage_scores=dosage_results
        )

        results.append({
            'sv_id': sv.id,
            'classification': classification,
            'genes': genes,
            'gnomad_af': gnomad['data']['structural_variants'][0]['af'] if gnomad['data']['structural_variants'] else None,
            'dosage_sensitive_genes': [
                g['gene_symbol'] for g in dosage_results
                if g['haploinsufficiency_score'] >= 2 or g['triplosensitivity_score'] >= 2
            ]
        })

    return results

SV Type-Specific Interpretation

Deletions

Check: Haploinsufficiency (HI) scores

• HI = 3: Likely pathogenic (if rare)
• HI = 2: Possibly pathogenic (if very rare)
• HI = 0: Likely benign (unless affects multiple genes)

Duplications

Check: Triplosensitivity (TS) scores

• TS = 3: Likely pathogenic (if rare)
• TS = 2: Possibly pathogenic (if very rare)
• TS = 0: Likely benign

Inversions

Check: Gene disruption

• If breaks gene: Likely loss of function
• If no gene disruption: May affect regulatory elements
• Rare inversions affecting known disease genes: Potentially pathogenic

Translocations

Check: Fusion genes, gene disruption

• Known oncogenic fusions (e.g., BCR-ABL): Pathogenic
• Disrupts tumor suppressor: Potentially pathogenic
• No gene involvement: Likely benign

SV Size Considerations

Size	Impact	Considerations
< 1 kb	Usually low	May affect exons or regulatory elements
1-50 kb	Variable	Check if affects complete genes
50 kb - 1 Mb	Moderate-High	Likely affects multiple genes
> 1 Mb	High	Affects many genes, genomic instability

Key Considerations

Recurrent CNV Regions

Some regions have well-established clinical significance:

• 22q11.2 deletion (DiGeorge syndrome)
• 15q11-q13 deletion (Prader-Willi / Angelman)
• 7q11.23 deletion (Williams syndrome)
• 17p11.2 deletion (Smith-Magenis syndrome)

For these regions, check established guidelines.

De Novo vs Inherited

• De novo SVs: More likely pathogenic
• Inherited from healthy parent: Less likely pathogenic (unless incomplete penetrance)

Breakpoint Precision

• SV callers report confidence intervals (CIPOS, CIEND)
• Precise breakpoints: Higher confidence in gene impact
• Imprecise breakpoints: May need confirmation

Limitations

1. Breakpoint resolution: SV callers have limited precision
2. Complex SVs: May require case-by-case interpretation
3. Regulatory regions: Impact hard to predict
4. Incomplete penetrance: Pathogenic SVs may not always cause disease
5. Database coverage: Not all SVs are in gnomAD or DGVa

Best Practices

1. Always check both population frequency AND dosage sensitivity
2. Consider gene function in disease context
3. Check for known pathogenic CNVs in ClinVar/OMIM
4. Recommend confirmatory testing (MLPA, microarray)
5. Provide genetic counseling for pathogenic findings