FAVOR-GPT is the conversational layer on top of FAVOR. It is most useful when your question is multi-step: find the right entities, inspect a variant list, run an analysis, then turn the result into a table or chart without leaving the same session.
Variant list follow-up
Start from a batch-annotated variant list, rank variants, filter by annotations, and ask for summary analyses.
Graph exploration
Compare genes, diseases, drugs, and variants, then walk paths and neighbourhoods in the knowledge graph.
Statistical analysis
Run PCA, regression, clustering, correlations, hypothesis tests, and QC workflows when the question needs actual computation.
Reusable sessions
Keep one session open while you refine the question. Entities, variant lists, and prior results stay in context across turns.
Output generation
Turn tool outputs into tables, plots, and concise summaries without manually exporting intermediate results.
Open the agent
Navigate to /agent. Sessions persist context, state, and artifacts across turns.
Set your context
Select a variant list from a batch annotation job, or paste variant IDs directly.
Ask for a concrete operation
Request a comparison, ranking, plot, QC pass, or statistical test so the agent can choose the right tool path.
Name the variant list or entity
The agent is more reliable when you anchor the conversation on a concrete variant list, gene, disease, or variant set.
Ask for an operation
Use verbs like compare, rank, summarize, test, plot, or explain so the agent can choose the right tool sequence.
State the output shape
If you want a table, chart, shortlist, or QC summary, say that explicitly. It reduces back-and-forth.
Keep related work in one session
A single session works better than restarting, because the agent can build on previous variant list selections and findings.
Set variant list <id> and show the top 10 variants by CADD with consequence and ClinVar labels.
Give me a QC summary for this variant list and flag possible data-quality issues before analysis.
Compare BRCA1 and TP53 for shared disease neighbours and show the strongest links.
Find paths from BRCA1 to breast cancer and summarize the shortest evidence-backed routes.
Run PCA on the numeric annotations in this variant list and explain the first two components.
Create a variant list from these variants: rs123, rs456, rs789, then rank them by deleteriousness.
| Domain | Operations |
|---|---|
| Variant list | Query rows, filter and sort, group-by aggregation, derive sub-lists, composite scoring, variant prioritization, export |
| Analytics | Regression, PCA, clustering, correlation, hypothesis testing, GWAS QC |
| Graph | Explore neighbors, compare entities, find shortest paths, enrichment queries, structural pattern matching |
| Workspace | Pin entities, switch variant lists, and save memories across sessions |
Tips
Interpretation
Engineering deep-dive