Annotation Pipeline

When you upload a VCF file to a case, Purna runs a multi-stage bioinformatics pipeline that annotates every variant with clinical classifications, population frequencies, functional predictions, and database cross-references. This page explains each stage in detail.

​

Supported inputs

File formats accepted:

• .vcf (uncompressed)
• .vcf.gz (gzip or bgzip compressed)
• .zip archives containing VCF files

Variant limits:

• Up to 1 million variants per case (suitable for Whole Exome Sequencing and targeted panels)
• For Whole Genome Sequencing files exceeding this limit, contact our enterprise team at contact@purna.ai

​

Pipeline overview

Each stage is described in detail below.

---

​

Stage 1: Validation and assembly detection

Once your VCF is uploaded, the pipeline validates the file format and automatically detects the genome assembly. Assembly detection uses multiple signals, in priority order:

1. VCF header fields (##genomeBuild, ##reference, ##assembly)
2. Contig lengths in the header (chromosome 1 length differs between GRCh37 and GRCh38)
3. Chromosome naming conventions (chr prefix patterns)

After detection, the pipeline selects the matching reference genome, VEP cache, and annotation databases for all downstream stages.

​

Stage 2: Variant annotation with Ensembl VEP

Every variant is annotated using Ensembl Variant Effect Predictor (VEP) release 115 with RefSeq transcripts.

​

Transcript selection

Purna selects the most clinically relevant transcript per gene using a strict priority order: This ensures that the transcript shown for each variant is the one most relevant for clinical interpretation.

​

Annotations added by VEP

HGVS nomenclature:

• Coding DNA change (c. notation, e.g., NM_007294.4:c.5266dupC)
• Protein change (p. notation, e.g., NP_009225.1:p.(Gln1756ProfsTer74))
• Genomic coordinate (g. notation)

Consequence and functional impact:

• Sequence Ontology consequence terms (e.g., missense_variant, frameshift_variant, splice_donor_variant)
• Consequence severity ranking (used for gene-level prioritization)
• Exon and intron positions (e.g., exon 10/23)
• Affected codons and amino acid changes

Transcript and gene metadata:

• RefSeq transcript ID (NM_) and protein ID (NP_)
• Gene symbol and NCBI Gene ID
• Biotype classification
• Canonical and MANE Select flags

In silico predictions: ClinVar annotations:

• Clinical significance (Pathogenic, Likely Pathogenic, Uncertain Significance, Likely Benign, Benign)
• Review status (criteria provided, reviewed by expert panel, practice guideline, etc.)
• Associated conditions and phenotypes
• ClinVar accession ID (VCV)

dbSNP identifiers:

• rsID from colocated variants (e.g., rs80357906)

​

Stage 3: Population frequency lookup (gnomAD)

Each variant is queried against the Genome Aggregation Database (gnomAD) for population allele frequencies.

​

Frequency fields per variant

Global frequencies:

• Allele frequency (AF), allele count (AC), allele number (AN)
• Homozygote count

Population-specific frequencies: Maximum population frequency:

• The highest AF across all populations is computed and stored alongside the name of that population group — useful for ACMG BA1 and BS1 criteria evaluation.

​

Stage 4: ACMG/AMP classification

Every variant receives an automated ACMG/AMP pathogenicity classification using a rule-based classifier that evaluates standard criteria against the annotated evidence.

​

Criteria evaluated

The classifier assesses 28 ACMG/AMP criteria across pathogenic and benign evidence categories: Pathogenic criteria: Benign criteria:

​

Classification output

Each variant is assigned one of five classes based on the combination of active criteria: For each variant, the specific criteria that contributed to the classification are stored along with their supporting evidence. You can review these in the ACMG tab of the variant detail panel.

​

Stage 5: Merge and transform

All annotation sources are merged into a single unified variant record:

• VEP annotations (consequence, HGVS, predictions, ClinVar, dbSNP)
• gnomAD population frequencies (global + per-population)
• ACMG classification and criteria evidence
• Sample-level genotype data extracted from the VCF (GT, DP, GQ, AD, VAF)

Genotype and zygosity derivation: Multi-gene handling: When a variant affects multiple genes, the most clinically severe gene annotation is designated as the primary record. Alternate transcript annotations are preserved and available in the variant detail panel.

​

Stage 6: Index and store

The fully annotated variants are indexed for fast querying and stored for long-term access.

• Search index — Each variant is indexed with full-text search on gene names and structured filtering on all annotation fields. This powers the variant search, filtering, and analytics features throughout the platform.
• Data warehouse — Variant records are stored in a columnar data warehouse for aggregations, exports, and cross-case analysis.

Once indexing completes, your case status updates to Ready and all variants are available for analysis in the case tabs, chat queries, and code execution.

---

​

Data available per variant

After the pipeline completes, each variant in your case contains the following fields:

​

Reference databases

Purna’s annotation pipeline draws from the following databases. All databases are updated regularly to ensure your analyses reflect the latest available evidence.

​

Processing time

Pipeline duration depends on the number of variants and the genome build: You will receive a notification when your case is ready for analysis. You can continue working in other cases or chat sessions while the pipeline runs.

​

Enterprise pipeline options

Enterprise customers have access to extended pipeline capabilities:

• FastQ processing — Upload raw sequencing reads (FastQ files) for alignment and variant calling on Nvidia Parabricks GPU infrastructure
• Custom databases — Integrate your own licensed databases such as HGMD Professional or VarSome Pro using your API keys
• Custom pipeline configurations — Request modifications to annotation sources, filtering criteria, or output formats tailored to your workflow

Contact contact@purna.ai to discuss enterprise options.