This document provides detailed QC checkpoints and expected metrics for each stage of variant annotation. --- ## 1. VCF Validation (Step 1) ### Critical Checkpoints ✅ **Format validation** - VCF passes format validation with no parsing errors - Conforms to VCF 4.2 specification - All required fields present (CHROM, POS, ID, REF, ALT, QUAL, FILTER, INFO) ✅ **Reference genome matching** - All contigs in VCF header match reference genome - Contig names consistent (e.g., "chr1" vs "1") - Variant positions within contig boundaries - No off-by-one coordinate errors ✅ **Variant integrity** - No duplicate variant positions - REF alleles match reference genome - ALT alleles are valid nucleotide sequences - Multiallelic sites properly formatted ✅ **Sample data quality** - FILTER fields consistent across variants - Genotype calls complete (minimal missing data) - FORMAT fields properly defined in header - Phase information valid if present ### Expected Metrics | Metric | Whole Genome | Whole Exome | Notes | | --- | --- | --- | --- | | **Ti/Tv ratio** | 2.0 - 2.1 | 2.8 - 3.0 | Transition/transversion ratio | | **Het/Hom ratio** | 1.5 - 2.0 | 1.5 - 2.0 | Heterozygous/homozygous ratio | | **Average QUAL** | > 30 | > 30 | Variant quality score | | **PASS rate** | > 90% | > 95% | % variants passing filters | | **Indel rate** | 10-15% | 5-10% | % indels vs SNVs | ### Red Flags ❌ **Poor variant quality:** - Ti/Tv < 1.5 (suggests low-quality variant calls) - Very high indel rate (> 20% of variants) - Mean QUAL < 20 - High proportion of multiallelic sites (> 10%) ❌ **Reference mismatch:** - Many REF alleles don't match reference genome - Variants outside contig boundaries - Inconsistent chromosome naming ❌ **Data integrity issues:** - Large number of missing genotypes (> 10%) - Inconsistent ploidy - FILTER fields not defined in header - Sample names duplicated or missing ### Common VCF Issues and Fixes | Issue | Cause | Solution | | --- | --- | --- | | "Invalid VCF header" | Malformed header lines | Fix ##INFO, ##FORMAT, ##contig definitions | | "Chromosome not found" | Chr naming mismatch (chr1 vs 1) | Use bcftools annotate to rename contigs | | "Duplicate positions" | Same variant listed multiple times | Deduplicate with bcftools norm -d | | "REF doesn't match genome" | Wrong reference genome | Re-call variants with correct reference | | "Missing genotype data" | Incomplete variant calling | Filter variants with high missingness | --- ## 2. Annotation Completeness (Steps 3-4) ### Critical Checkpoints ✅ **Annotation success** - 100% of variants successfully annotated - No skipped variants in output - VEP/SNPEff completed without errors - Output VCF/file size reasonable (not empty) ✅ **Consequence assignment** - Every variant has at least one consequence - Consequence terms valid (from Sequence Ontology) - IMPACT assigned (HIGH/MODERATE/LOW/MODIFIER) - Transcript information present for coding variants ✅ **Gene annotation** - Gene symbols present for coding/regulatory variants - Ensembl Gene IDs assigned - Transcript IDs correspond to correct genome build - Canonical transcripts marked (if requested) ✅ **Supplementary annotations** - Population frequencies present (if plugin enabled) - Pathogenicity scores populated (for missense variants) - Clinical annotations loaded (ClinVar, COSMIC) - HGVS notation generated correctly ### Expected Metrics | Metric | Whole Genome | Whole Exome | Notes | | --- | --- | --- | --- | | **Annotation rate** | 100% | 100% | All variants annotated | | **Coding variants** | 1-2% | 40-60% | % coding consequences | | **Gene assignment** | 30-50% | > 95% | % variants with gene symbol | | **HIGH impact** | 0.01-0.05% | 0.1-0.5% | Stop-gain, frameshift, etc. | | **MODERATE impact** | 0.5-1% | 10-20% | Missense, inframe indels | | **ClinVar matches** | 0.01-0.1% | 0.1-1% | Known clinical variants | ### Red Flags ❌ **Annotation failure:** - Large number of variants with no consequence assigned - No gene symbols for exome variants (> 50% intergenic) - All consequences are "intergenic\_variant" in exome - Empty or truncated output file ❌ **Genome build mismatch:** - Most exome variants annotated as intergenic (should be coding) - Very few gene hits in targeted sequencing - ClinVar matches absent for common clinical panels - Check: VCF genome build vs annotation database build ❌ **Plugin/database issues:** - All pathogenicity scores missing (SIFT, PolyPhen) - No population frequencies (gnomAD, 1000G) - ClinVar annotations absent - Check: Plugins installed and databases downloaded ### VEP-Specific Checks **Cache vs API mode:** - Cache mode: 100-1000x faster, requires 15-20 GB disk - API mode: Very slow for large VCFs, no disk space needed - Recommendation: Always use cache for > 1000 variants **Plugin verification:** ``` # Check installed VEP plugins vep --list_plugins # Expected plugins for clinical analysis: # - CADD # - dbNSFP # - REVEL # - SpliceAI # - Mastermind ``` **Common VEP errors:** ``` # "Cache directory not found" # Solution: Download cache vep_install -a cf -s homo_sapiens -y GRCh38 # "Plugin failed: CADD" # Solution: Download CADD scores cd $VEP_CACHE_DIR wget https://krishna.gs.washington.edu/download/CADD/v1.6/GRCh38/whole_genome_SNVs.tsv.gz ``` ### SNPEff-Specific Checks **Database verification:** ``` # List available databases snpEff databases | grep -i GRCh38 # Expected output for human: # GRCh38.105 (Ensembl release 105) # GRCh38.p13 (RefSeq annotation) ``` **Annotation field validation:** - ANN field present in INFO column - ANN subfields: Allele|Annotation|Impact|Gene\_Name|... - LOF field present (if loss-of-function plugin enabled) - NMD field present (if nonsense-mediated decay enabled) **Common SNPEff errors:** ``` # "Genome 'GRCh38' not found" # Solution: Download SNPEff database snpEff download -v GRCh38.105 # "Cannot read FASTA file" # Solution: SNPEff doesn't need FASTA (only database) # Remove -ref option if present ``` --- ## 3. Filtering Results (Step 5) ### Critical Checkpoints ✅ **Appropriate stringency** - Filtering criteria match use case - Not over-filtering (losing true positives) - Not under-filtering (too many variants) - Balance sensitivity vs specificity ✅ **Consequence filtering** - Impact thresholds appropriate - Critical consequence types included - Synonymous variants excluded (or included if needed) ✅ **Frequency filtering** - Population frequency threshold appropriate for disease model - Rare disease: AF < 0.01 (dominant) or < 0.05 (recessive) - Common disease: More permissive thresholds - Missing frequencies handled correctly ✅ **Quality filtering** - Variant quality (QUAL) threshold applied - Depth of coverage (DP) considered - Allele balance checked for heterozygous variants - PASS filter applied if appropriate ### Expected Metrics After Filtering | Filter Level | Whole Exome | Whole Genome | Clinical Sample | | --- | --- | --- | --- | | **All variants** | 20,000-40,000 | 3-5 million | Variable | | **Coding variants** | 8,000-15,000 | 50,000-100,000 | 100-5,000 | | **HIGH/MODERATE impact** | 5,000-10,000 | 30,000-60,000 | 50-2,000 | | **Rare (AF < 0.01)** | 500-2,000 | 10,000-30,000 | 20-500 | | **Rare + HIGH impact** | 10-50 | 100-500 | 1-20 | | **Prioritized pathogenic** | 1-20 | 10-100 | 0-10 | ### Decision Point: Adjust Filtering Thresholds **Too many variants (> 1000 after filtering):** - ✅ Decrease allele frequency threshold (0.01 → 0.001) - ✅ Increase pathogenicity score threshold (CADD > 25, REVEL > 0.7) - ✅ Require multiple pathogenicity predictions to agree - ✅ Filter to HIGH impact only (exclude MODERATE) - ✅ Include only ClinVar Pathogenic/Likely pathogenic **Too few variants (< 5 after filtering):** - ✅ Increase allele frequency threshold (0.01 → 0.05) - ✅ Decrease pathogenicity score threshold (CADD > 15, REVEL > 0.3) - ✅ Include MODERATE impact variants - ✅ Include variants of uncertain significance (VUS) - ✅ Check if variants were lost in earlier QC steps **No candidate variants:** - ⚠️ Review original VCF quality (Ti/Tv ratio, variant counts) - ⚠️ Verify genome build matches (VCF vs annotation database) - ⚠️ Check if disease gene is in targeted region (for panels) - ⚠️ Consider expanding to VUS or lower impact variants - ⚠️ Review inheritance model and filtering assumptions ### Use Case-Specific Filtering Strategies **1. Clinical Diagnostics (Rare Mendelian Disease)** ``` Stringent filtering: - Impact: HIGH or MODERATE only - Frequency: gnomAD AF < 0.01 (dominant) or < 0.05 (recessive) - Pathogenicity: CADD > 20 OR REVEL > 0.5 OR ClinVar P/LP - Quality: QUAL > 30, DP > 10, AB 0.3-0.7 (het) - Consequences: Loss-of-function, missense, splice site Expected output: 1-20 candidate variants per patient ``` **2. Cancer Genomics (Somatic Variants)** ``` Cancer-specific filtering: - Impact: HIGH, MODERATE, or known cancer hotspot - Frequency: Not present in germline (gnomAD AF < 0.001) - Databases: COSMIC mutations, OncoKB variants - Quality: Tumor VAF > 0.05, normal VAF < 0.02 - Genes: Cancer gene census, driver genes Expected output: 1-10 driver mutations per tumor ``` **3. Population Genetics (Research)** ``` Permissive filtering: - Impact: All impact levels - Frequency: AF < 0.05 or no frequency filter - Quality: QUAL > 20, DP > 5 - Consequences: All coding variants Expected output: 500-5,000 variants per sample ``` **4. Pharmacogenomics** ``` PGx-specific filtering: - Genes: PGx gene list (CYP2D6, CYP2C19, TPMT, etc.) - Impact: All impacts (including synonymous if functional) - Frequency: No frequency filter (common variants important) - Databases: PharmGKB clinical annotations - Include: Star alleles, structural variants Expected output: 10-100 PGx variants per sample ``` --- ## 4. Gene-Level Summary (Step 6) ### Critical Checkpoints ✅ **Gene aggregation** - Variants correctly grouped by gene - Multiple transcripts handled appropriately - Gene symbols standardized (HGNC for human) - Ensembl IDs match genome build ✅ **Variant counting** - Total variants per gene accurate - Counts by impact level (HIGH/MODERATE/LOW) - Counts by consequence type - No double-counting of variants ✅ **Score aggregation** - Maximum or mean scores calculated correctly - Missing scores handled appropriately - Scores correspond to correct variants ### Expected Metrics | Metric | Clinical Panel | Whole Exome | Whole Genome | | --- | --- | --- | --- | | **Genes with variants** | 50-200 | 8,000-12,000 | 15,000-20,000 | | **Genes with HIGH impact** | 5-20 | 50-200 | 100-500 | | **Genes with 2+ variants** | 10-50 | 1,000-3,000 | 3,000-8,000 | | **Mean variants per gene** | 1-3 | 1-2 | 2-5 | ### Red Flags ❌ One gene has > 100 variants: Possible alignment artifact or repeat region ❌ No genes with > 1 variant: Over-filtering or small sample size ❌ Many genes with only synonymous variants: Consider filtering these out --- ## 5. Variant Prioritization (Step 7) ### Critical Checkpoints ✅ **Prioritization criteria** - Scoring system appropriate for use case - Weights reflect clinical/research priorities - ACMG criteria applied correctly (for clinical) - ClinVar annotations weighted highly ✅ **Pathogenicity classification** - ACMG/AMP categories assigned - Evidence criteria documented - Conflicting predictions handled - Uncertain significance not over-interpreted ### Expected Metrics | Priority Level | Expected Count | Interpretation | | --- | --- | --- | | **Pathogenic** | 0-5 per sample | Known pathogenic, report immediately | | **Likely Pathogenic** | 1-10 per sample | Strong evidence, clinical follow-up | | **VUS** | 5-50 per sample | Uncertain, monitor or functional studies | | **Likely Benign** | Many | Can filter out for clinical reports | | **Benign** | Many | Can filter out for clinical reports | ### Decision Point: Clinical Reporting **Report immediately:** - ✅ ClinVar Pathogenic in disease-relevant gene - ✅ Known disease-causing variant (HGMD, ClinVar) - ✅ Loss-of-function in haploinsufficient gene - ✅ ACMG secondary findings (incidental findings) **Report with caution:** - ⚠️ VUS in disease-relevant gene with strong predictions - ⚠️ Novel missense with CADD > 30, REVEL > 0.9 - ⚠️ Splicing variant near canonical splice site - ⚠️ De novo variant in patient-parent trio **Do not report:** - ❌ Common variants (AF > 0.05) unless pharmacogenomic - ❌ Synonymous variants (unless functional evidence) - ❌ Benign or Likely benign ClinVar classification - ❌ Variants in genes unrelated to phenotype --- ## 6. Visualization QC (Step 8) ### Expected Visualizations **1. Consequence Distribution** - Most variants should be intronic or intergenic (genome) - Most variants should be coding (exome) - Missense variants most common coding consequence - Stop-gain and frameshift rare (< 1%) **2. Impact by Chromosome** - Impact distribution similar across chromosomes - No chromosome with disproportionate HIGH impact (may indicate QC issue) - Sex chromosomes (X, Y) may differ in coverage/calling **3. Pathogenicity Score Distributions** - CADD scores: Most variants < 20, few > 30 - REVEL scores: Bimodal distribution (benign vs deleterious) - PolyPhen: Categories clearly separated **4. Allele Frequency** - Most variants rare (AF < 0.01) - Log-scale distribution shows full spectrum - Very rare variants (AF < 0.0001) most enriched **5. Gene Burden** - Some genes naturally have more variants (large genes) - Outliers may indicate artifacts or true hotspots - Compare to known mutation-tolerant/intolerant genes ### Red Flags in Visualizations ❌ **Uniform consequences:** If all variants same consequence, annotation may have failed ❌ **Single chromosome outlier:** Possible batch effect or alignment issue ❌ **No rare variants:** Frequency filtering may be too stringent ❌ **All high-scoring:** Pathogenicity filtering may be too permissive --- ## Summary: QC Checklist Use this checklist at each workflow stage: ### ✅ Step 1: VCF Validation - [ ] VCF passes format validation - [ ] Ti/Tv ratio in expected range (2.0-2.1 WGS, 2.8-3.0 WES) - [ ] Het/Hom ratio 1.5-2.0 - [ ] No duplicate variants - [ ] Average QUAL > 30 ### ✅ Step 3-4: Annotation - [ ] 100% variants annotated - [ ] Coding variants percentage appropriate (1-2% WGS, 40-60% WES) - [ ] Gene symbols present for coding variants - [ ] Pathogenicity scores populated (if plugins enabled) - [ ] No genome build mismatch ### ✅ Step 5: Filtering - [ ] Filtered variant count reasonable (10-100 for clinical, 500-5000 for research) - [ ] HIGH impact variants: 10-50 (exome), 100-500 (genome) - [ ] Rare high-impact variants: 1-20 (clinical) - [ ] Filtering thresholds documented ### ✅ Step 6-7: Gene Summary & Prioritization - [ ] Gene counts reasonable (50-200 panel, 8000-12000 exome) - [ ] Prioritized variants: 1-20 for clinical reporting - [ ] ACMG classifications appropriate - [ ] Pathogenic/Likely pathogenic variants manually reviewed ### ✅ Step 8: Visualizations - [ ] Consequence distribution matches sequencing type - [ ] No chromosomal outliers - [ ] Allele frequency distribution shows rare variants - [ ] Pathogenicity scores show expected bimodal distribution --- **For additional troubleshooting, see:** troubleshooting\_guide.md