Bioinformatics Lab UI — User Guide¶
The Bio Lab is the interactive front end for Agentic-SpliceAI: browse genes, run splice-site
prediction on demand, compare a meta model against its base model, and inspect novel-site candidates.
FastAPI + Jinja2 + Plotly on port 8005, backed by the same src/ library as everything else — so
what you see here is what the models actually produce, not a re-implementation.
The four pages¶
| Page | Question it answers |
|---|---|
Gene Browser (/) |
Which gene do I want, and what annotation am I looking at? |
Genome View (/genome/{gene}) |
Did the model get this gene's known sites right? |
Novel Site Explorer (/novel/{gene}) |
What's here that no annotation knows about? |
Metrics Dashboard (/metrics) |
How do the models compare across all held-out genes? |
Read Reading the numbers before quoting anything from these pages. It covers the two mistakes that are easy to make here: treating 0.5 as an operating point, and reading a score without knowing which annotation it was measured against.
Start here¶
- Getting started — launch, warm the caches, know what needs prebuilt data.
- Gene Browser — find a gene (including by protein name, e.g.
TDP-43). - Genome View — the main workspace, and the base-vs-meta comparison.
What is not here¶
- Training and evaluation live in the Meta-Layer MLOps workflow. The Lab consumes trained checkpoints; it never trains.
- The driver scripts that build the caches this UI serves are in
examples/UI_integration/. Those are development scripts, not user documentation. - Published results are in meta-layer results. Numbers shown in the UI are per-gene and exploratory; the results pages are the held-out, citable ones.