CoolFace
Apppublic

AIRI-Institute/genatator-ab-initio-segmentation-leaderboard

sourceHugging Faceapache-2.0updated 3mo agoView on Hugging Face
3likes
App README

Gene-level Metric + Live Leaderboard

This Space now includes:

  1. 1.Metric usage page (Evaluate API + interactive calculator).
  2. 2.Live leaderboard page that computes scores from prediction files automatically.

Leaderboard pipeline

Leaderboard starts automatically when the app starts. Backend will:

  1. 1.Clone/pull: https://github.com/alexeyshmelev/genatator-leaderboard-predictions.git.
  2. 2.Enter predictions/ and process all *.gff files.
  3. 3.Compute gene-level metric for each file.
  4. 4.Compute BUSCO metric for each file.
  5. 5.Stream progress and partial results via /api/leaderboard/status.

Required local assets

Put these files into leaderboard_required_files/:

  • —chr20.gff
  • —Hs_NC_060944.1.fa
  • —lineage/mammalia_odb10/ (BUSCO offline lineage directory)

See leaderboard_required_files/README.md.

BUSCO installation

BUSCO is installed at container startup (runtime) to avoid OOM during image build on free tier.

Startup script installs Miniconda (if missing), then installs:

bash
conda create -n busco_env -c conda-forge -c bioconda python=3.12 busco==5.7.1

BUSCO is installed in a dedicated conda env (busco_env) with Python 3.12 to avoid solver conflicts, and is run in offline proteins mode for each model.

Local run

bash
docker build -t gene-level-metric .
docker run -p 7860:7860 gene-level-metric

Open http://localhost:7860.

AIRI-Institute/genatator-ab-initio-segmentation-leaderboard · CoolFace