AIRI-Institute/genatator-ab-initio-segmentation-leaderboard
3
Gene-level Metric + Live Leaderboard
This Space now includes:
- Metric usage page (Evaluate API + interactive calculator).
- Live leaderboard page that computes scores from prediction files automatically.
Leaderboard pipeline
Leaderboard starts automatically when the app starts. Backend will:
- Clone/pull:
https://github.com/alexeyshmelev/genatator-leaderboard-predictions.git. - Enter
predictions/and process all*.gfffiles. - Compute gene-level metric for each file.
- Compute BUSCO metric for each file.
- Stream progress and partial results via
/api/leaderboard/status.
Required local assets
Put these files into leaderboard_required_files/:
chr20.gffHs_NC_060944.1.falineage/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:
conda create -n busco_env -c conda-forge -c bioconda python=3.12 busco==5.7.1BUSCO 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
docker build -t gene-level-metric .
docker run -p 7860:7860 gene-level-metricOpen http://localhost:7860.
