Skip to content

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.

conda run -n agentic-spliceai python -m server.bio.app
# → http://localhost:8005/

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

  1. Getting started — launch, warm the caches, know what needs prebuilt data.
  2. Gene Browser — find a gene (including by protein name, e.g. TDP-43).
  3. 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.