bio-variant-calling-clinical-interpretation

What this skill does

## Version Compatibility

Reference examples tested with: Entrez Direct 21.0+, bcftools 1.19+, cyvcf2 0.30+

Before using code patterns, verify installed versions match. If versions differ:
- Python: `pip show <package>` then `help(module.function)` to check signatures
- CLI: `<tool> --version` then `<tool> --help` to confirm flags

If code throws ImportError, AttributeError, or TypeError, introspect the installed
package and adapt the example to match the actual API rather than retrying.

# Clinical Variant Interpretation

Prioritize and interpret variants for clinical significance using databases and ACMG/AMP guidelines.

## Interpretation Framework

```
Annotated VCF
    │
    ├── Database Lookup
    │   ├── ClinVar (clinical assertions)
    │   ├── OMIM (disease associations)
    │   └── gnomAD (population frequency)
    │
    ├── Computational Predictions
    │   ├── SIFT, PolyPhen-2
    │   ├── CADD, REVEL
    │   └── SpliceAI
    │
    ├── ACMG Classification
    │   └── Pathogenic -> Likely Pathogenic -> VUS -> Likely Benign -> Benign
    │
    └── Prioritized Variant List
```

## ClinVar Annotation

**Goal:** Annotate variants with ClinVar clinical significance and filter by pathogenicity.

**Approach:** Download the ClinVar VCF, add CLNSIG/CLNDN/CLNREVSTAT fields with bcftools annotate, then filter by significance level.

**"Find pathogenic variants in my VCF"** -> Cross-reference variants against ClinVar clinical assertions and extract those classified as pathogenic or likely pathogenic.

### Download ClinVar

```bash
wget https://ftp.ncbi.nlm.nih.gov/pub/clinvar/vcf_GRCh38/clinvar.vcf.gz
wget https://ftp.ncbi.nlm.nih.gov/pub/clinvar/vcf_GRCh38/clinvar.vcf.gz.tbi
```

### Annotate with bcftools

```bash
bcftools annotate \
    -a clinvar.vcf.gz \
    -c INFO/CLNSIG,INFO/CLNDN,INFO/CLNREVSTAT \
    input.vcf.gz -Oz -o with_clinvar.vcf.gz
```

### Filter Pathogenic Variants

```bash
# Pathogenic or Likely pathogenic
bcftools view -i 'INFO/CLNSIG~"Pathogenic" || INFO/CLNSIG~"Likely_pathogenic"' \
    with_clinvar.vcf.gz -Oz -o pathogenic.vcf.gz

# Exclude benign
bcftools view -e 'INFO/CLNSIG~"Benign" || INFO/CLNSIG~"Likely_benign"' \
    with_clinvar.vcf.gz -Oz -o not_benign.vcf.gz
```

## ClinVar Significance Levels

| CLNSIG | Meaning | Action |
|--------|---------|--------|
| Pathogenic | Disease-causing | Report |
| Likely_pathogenic | Probably disease-causing | Report with caveat |
| Uncertain_significance | VUS | May report, needs follow-up |
| Likely_benign | Probably not disease-causing | Usually exclude |
| Benign | Not disease-causing | Exclude |
| Conflicting | Multiple interpretations | Manual review |

## ClinVar Review Status

| CLNREVSTAT | Stars | Meaning |
|------------|-------|---------|
| practice_guideline | 4 | Expert panel reviewed |
| reviewed_by_expert_panel | 3 | ClinGen expert reviewed |
| criteria_provided,_multiple_submitters | 2 | Consistent assertions |
| criteria_provided,_single_submitter | 1 | One submitter with criteria |
| no_assertion_criteria | 0 | No criteria provided |

```bash
# Filter for high-confidence assertions (2+ stars)
bcftools view -i 'INFO/CLNREVSTAT~"multiple_submitters" || \
    INFO/CLNREVSTAT~"expert_panel" || \
    INFO/CLNREVSTAT~"practice_guideline"' \
    with_clinvar.vcf.gz -Oz -o high_confidence.vcf.gz
```

## InterVar (ACMG Classification)

**Goal:** Classify variants according to ACMG/AMP guidelines using automated criteria evaluation.

**Approach:** Convert VCF to ANNOVAR format, run InterVar to evaluate 28 ACMG criteria, and output five-tier classification.

Automated ACMG/AMP variant classification.

### Installation

```bash
git clone https://github.com/WGLab/InterVar.git
cd InterVar
# Download databases per documentation
```

### Run InterVar

```bash
python Intervar.py \
    -i input.avinput \
    -o output \
    -b hg38 \
    -d humandb/ \
    --input_type=AVinput
```

### From VCF

```bash
# Convert VCF to ANNOVAR format
convert2annovar.pl -format vcf4 input.vcf > input.avinput

# Run InterVar
python Intervar.py -i input.avinput -o intervar_results -b hg38
```

## ACMG/AMP Criteria

### Pathogenic Criteria

| Code | Type | Description |
|------|------|-------------|
| PVS1 | Very Strong | Null variant in gene where LOF is disease mechanism |
| PS1-4 | Strong | Same AA change, functional studies, etc. |
| PM1-6 | Moderate | Hot spot, absent from controls, etc. |
| PP1-5 | Supporting | Co-segregation, computational evidence |

### Benign Criteria

| Code | Type | Description |
|------|------|-------------|
| BA1 | Stand-alone | AF >5% in gnomAD |
| BS1-4 | Strong | AF greater than expected, functional studies |
| BP1-7 | Supporting | Missense in gene with truncating mechanism |

### Classification Rules

These are guidelines, not absolute rules -- expert review is always required.

**Pathogenic:** (PVS1 AND >=1 PS) OR (>=2 PS) OR (1 PS AND >=3 PM) OR (>=2 PM AND >=2 PP) OR (1 PM AND >=4 PP)

**Likely Pathogenic:** (1 PVS1 AND 1 PM) OR (1 PS AND 1-2 PM) OR (1 PS AND >=2 PP) OR (>=3 PM) OR (2 PM AND >=2 PP) OR (1 PM AND >=4 PP)

**VUS:** Does not meet criteria for pathogenic, likely pathogenic, benign, or likely benign. This is not a classification of uncertainty about pathogenicity per se, but rather insufficient evidence to classify in either direction.

**Likely Benign:** (1 BS AND 1 BP) OR (>=2 BP)

**Benign:** (BA1 alone) OR (>=2 BS)

### Annotation Concordance and PVS1

Different annotation tools (VEP vs. SnpEff) disagree on loss-of-function classification in 33-44% of cases. This directly impacts PVS1 -- the strongest single evidence criterion. For clinical-grade interpretation, annotate with both VEP and SnpEff and flag discrepant LOF calls for manual review. A variant called LOF by only one tool should not receive PVS1 without further investigation.

## Population Frequency Filtering

**Goal:** Restrict to rare variants that could be disease-causing.

**Approach:** Filter by gnomAD allele frequency threshold appropriate for the disease model (dominant vs. recessive). Use filtering allele frequency (FAF) where available, as it accounts for sampling error and is more appropriate than raw AF for clinical filtering.

| Disease Model | AF Threshold | Rationale |
|---------------|-------------|-----------|
| Dominant | < 0.0001 (1/10,000) | Penetrant dominant variants cannot be common |
| Recessive | < 0.01 (1/100) | Carriers can be relatively common |
| BA1 (stand-alone benign) | > 0.05 | Too common to cause rare disease |

Population-specific frequencies must be checked, not just global AF. A variant rare globally may be common in one population (e.g., sickle cell HbS in West African populations). The gnomAD filtering allele frequency (FAF) field adjusts for population substructure and sampling uncertainty and is preferred over raw AF for clinical filtering.

```bash
# Dominant disease model (AF < 0.0001)
bcftools view -i 'INFO/gnomAD_AF<0.0001 || INFO/gnomAD_AF="."' \
    input.vcf.gz -Oz -o dominant_rare.vcf.gz

# Recessive disease model (AF < 0.01)
bcftools view -i 'INFO/gnomAD_AF<0.01 || INFO/gnomAD_AF="."' \
    input.vcf.gz -Oz -o recessive_rare.vcf.gz

# When FAF annotation is available (preferred for clinical use)
bcftools view -i 'INFO/gnomAD_FAF<0.0001 || INFO/gnomAD_FAF="."' \
    input.vcf.gz -Oz -o faf_filtered.vcf.gz
```

## Pathogenicity Score Filtering

**Goal:** Prioritize variants using computational pathogenicity predictors.

**Approach:** Filter by CADD PHRED score (deleteriousness) and REVEL score (missense pathogenicity), alone or in combination with ClinVar. No single predictor is sufficient for clinical classification; ACMG PP3/BP4 criteria allow computational evidence as supporting, not strong, evidence.

### Predictor Interpretation

| Predictor | Threshold | Notes |
|-----------|-----------|-------|
| CADD PHRED | >= 20 (top 1%), >= 30 (top 0.1%) | Low specificity (~12%); use for prioritization, not binary classification |
| REVEL | > 0.5 (bal