Integrated Rare-Disease Variant Annotation Pipeline

Small variants and CNVs on GRCh38

Generated: 2026-07-20T12:35:50

Teaching and research use only. The datasets are synthetic and the results are not clinical diagnoses.

Executive Summary

A reproducible, container-based workflow was developed for the annotation of small variants and copy-number variants in four synthetic rare-disease datasets: Achondroplasia, Klippel–Trénaunay syndrome, McCune–Albright syndrome and Pompe disease.

The small-variant branch uses bcftools, Ensembl VEP 115, SnpEff, ClinVar, SpliceAI and cached gnomAD fields. The CNV branch uses AnnotSV, ClassifyCNV and ISV-CNV with SHAP values.

All four pipelines completed successfully. Every final small-variant VCF retained all 20 input records. The GPT/manual synthetic baseline and pipeline outputs matched for both small variants and CNVs in all four diseases.

Automated Workflow

Small variants

Raw VCF
  → bcftools normalization
  → VEP
  → SnpEff
  → ClinVar
  → SpliceAI
  → cached gnomAD
  → final VCF.gz + index
  

CNVs

Raw CNV VCF
  → BED conversion
  → AnnotSV
  → ClassifyCNV
  → ISV-CNV probabilities and SHAP values
  

Tools and Containers

Container Purpose
core_tools.sifbcftools, samtools, bgzip, tabix and bedtools
vep.sifEnsembl VEP 115 annotation
snpeff.sifSnpEff consequence annotation
spliceai.sifSplice-effect prediction
annotsv.sifStructural-variant annotation and ranking
classifycnv.sifACMG-style CNV classification
isvcnv.sifMachine-learning CNV prioritization and SHAP values

Cross-Disease Results

Disease Gene ClinVar AnnotSV ClassifyCNV ISV probability SNV comparison CNV comparison
Achondroplasia FGFR3 Pathogenic 3 Uncertain significance 0.006764889229089022 MATCH MATCH
Klippel–Trénaunay syndrome PIK3CA Pathogenic 3 Uncertain significance 0.013497603125870228 MATCH MATCH
McCune–Albright syndrome GNAS Pathogenic/Likely pathogenic 5 Likely pathogenic 0.8791712522506714 MATCH MATCH
Pompe disease GAA Pathogenic 5 Likely pathogenic 0.3094403147697449 MATCH MATCH

Achondroplasia

Disease and inheritance

Main gene
FGFR3
Inheritance
Autosomal dominant

Main small variant

Variant
chr4:1804392 G>A; rs28931614; p.Gly380Arg
GPT/manual expectation
Pathogenic
Pipeline ClinVar
Pathogenic
Record comparison
MATCH
Records retained
20 of 20

Main CNV

CNV
chr4:1793293–1808867 DUP; genotype 0/1
Expected class
VUS
AnnotSV
3
ClassifyCNV
Uncertain significance
ISV probability
0.006764889229089022
Comparison
MATCH

Integrated educational interpretation

The activating FGFR3 small variant matches the established Achondroplasia mechanism more directly than the synthetic duplication.

Mechanism-fit summary: SNV: High; CNV: Low

Klippel–Trénaunay syndrome

Disease and inheritance

Main gene
PIK3CA
Inheritance
Postzygotic somatic mosaic

Main small variant

Variant
chr3:179234297 A>G; rs121913279; synthetic VAF 8%
GPT/manual expectation
Pathogenic
Pipeline ClinVar
Pathogenic
Record comparison
MATCH
Records retained
20 of 20

Main CNV

CNV
chr3:179148126–179240093 DUP; genotype 0/1
Expected class
VUS
AnnotSV
3
ClassifyCNV
Uncertain significance
ISV probability
0.013497603125870228
Comparison
MATCH

Integrated educational interpretation

The mosaic activating PIK3CA small variant matches the usual disease mechanism more directly than the synthetic duplication.

Mechanism-fit summary: SNV: High; CNV: Low

McCune–Albright syndrome

Disease and inheritance

Main gene
GNAS
Inheritance
Postzygotic somatic mosaic

Main small variant

Variant
chr20:58909366 G>A; rs121913495; synthetic VAF 12%
GPT/manual expectation
Pathogenic or likely pathogenic
Pipeline ClinVar
Pathogenic/Likely pathogenic
Record comparison
MATCH
Records retained
20 of 20

Main CNV

CNV
chr20:58839748–58911192 DEL; genotype 0/1
Expected class
Pathogenic or likely pathogenic
AnnotSV
5
ClassifyCNV
Likely pathogenic
ISV probability
0.8791712522506714
Comparison
MATCH

Integrated educational interpretation

The activating mosaic GNAS small variant matches McCune–Albright syndrome. The deletion may be pathogenic through dosage but is not the typical mechanism of this named disease.

Mechanism-fit summary: SNV: High; CNV: Low for named disease

Pompe disease

Disease and inheritance

Main gene
GAA
Inheritance
Autosomal recessive

Main small variant

Variant
chr17:80104542 T>G; rs386834236; homozygous
GPT/manual expectation
Pathogenic
Pipeline ClinVar
Pathogenic
Record comparison
MATCH
Records retained
20 of 20

Main CNV

CNV
chr17:80101581–80119881 DEL; genotype 1/1
Expected class
Pathogenic
AnnotSV
5
ClassifyCNV
Likely pathogenic
ISV probability
0.3094403147697449
Comparison
MATCH

Integrated educational interpretation

The homozygous small-variant and deletion designs both fit a recessive GAA loss-of-function testing model.

Mechanism-fit summary: SNV: High; CNV: High

GPT/Manual vs Pipeline Comparison

All four small-variant files contained 20 GPT/manual baseline records and 20 pipeline records. All variant identities matched.

The expected main-variant classifications agreed with the relevant pipeline fields. CNV expected classifications also agreed with AnnotSV and ClassifyCNV for every disease.

The GPT/manual files are transparent manually curated synthetic baselines. They are not presented as literal output from an independent clinical annotation engine.

Limitations

Conclusion

A complete two-branch GRCh38 annotation pipeline was successfully built, containerized and validated. The pipeline preserved all background variants and produced complementary small-variant and CNV results for all four diseases.

The final repository includes synthetic inputs, automated pipeline scripts, selected annotated outputs, GPT/manual baseline files, comparison reports and an integrated clinical-style summary.