Tertiary Analysis¶

Tertiary analysis in AIVA provides a unified workspace where the data table and AI chat are displayed side by side. This layout is designed for focused, per-sample variant interpretation. You can filter, sort, and flag variants in the table while simultaneously querying AIVA for context, literature, and computational analysis.

Accessing Tertiary Analysis¶

1. Navigate to the Analysis section from the main navigation.
2. Select a sample from the sample list.
3. The tertiary analysis view opens with two panels:
4. Left panel: The Data Table displaying the sample's variant data.
5. Right panel: An AIVA Chat session scoped to the selected sample.

Layout and Workflow¶

The side-by-side layout enables a continuous interpretation workflow:

Data Table Panel¶

The left panel provides the full data table experience:

• All columns are available via the column chooser, including VCF fields, INFO subfields, and Small Variant Annotation / Structural Variant Annotation columns.
• Filtering lets you narrow the variant list using text, numeric, date, and multi-select filters.
• Sorting by any column to prioritize variants of interest.
• Variant flagging directly in the table to mark variants as Pathogenic, VUS, Benign, etc. See Variant Flagging.
• Threaded comments on individual variants for discussion and documentation. See Threaded Comments.
• ACMG classification accessible from the variant row for evidence-based assessment. See ACMG Classification.

AI Chat Panel¶

The right panel provides a chat session connected to the selected sample's data:

• AIVA can query the sample's data directly using the Genomic Data Query tool.
• Ask questions about specific variants visible in the table, or request summaries across the entire dataset.
• Use Variant Annotation for real-time ClinVar, gnomAD, and prediction score lookups.
• Search literature with the Biomedical Literature tool and the web for evidence.
• Generate plots and run statistical tests with the Code Interpreter.

Typical Tertiary Analysis Workflow¶

Step 1: Initial filtering¶

Apply filters to reduce the variant list to candidates of interest:

1. Filter FILTER column to PASS to exclude low-quality calls.
2. Filter by allele frequency (e.g., gnomAD AF < 0.01) to focus on rare variants.
3. Filter by consequence type (e.g., missense, frameshift, stop gained) to focus on functional variants.

Step 2: Review and flag¶

Scroll through the filtered variants and flag those requiring attention:

1. Review each variant's annotation columns (Gene, Consequence, SIFT, PolyPhen, ClinVar).
2. Flag variants using the appropriate category (Pathogenic, VUS, etc.).
3. Add comments to document your initial impressions.

Step 3: AI-assisted deep dive¶

For each flagged variant, use AIVA to gather additional evidence:

1. Ask AIVA to look up ClinVar and gnomAD data for the variant.
2. Request a literature search for the gene and associated conditions.
3. Query the knowledge graph for drug-target interactions if relevant.
4. Check for related clinical trials.

Step 4: Formal classification¶

For variants requiring a formal assessment, open the ACMG classifier and apply criteria based on the evidence gathered.

Step 5: Export¶

Export your flagged and classified variants with comments for inclusion in clinical reports or team review.

Tips¶

• Filter aggressively first: Reduce the variant list to a manageable size before starting detailed review. Most samples contain thousands of variants, but only a fraction are clinically relevant.
• Use AIVA for repetitive lookups: Instead of manually checking ClinVar or gnomAD for each variant, ask AIVA to batch-query or look up individual variants as you encounter them.
• Document as you go: Add comments and flags during review rather than after. This creates a real-time record of your interpretation process.