GENOMICS-CDDBRG/SeqPure-BLAST

Bulamarwa Metagenomics Tool 2: Statistics and Reporting

Second tool in a two tool metagenomics workflow. Takes a species abundance table and sample metadata, typically species_relative_abundance.tsv (or species_estimated_reads.tsv for count based Chao1) and samplesheet_resolved.csv produced by Gulumba Metagenomics Tool 1. Abundance tables from other pipelines, such as MetaPhlAn or HUMAnN, are also accepted in either counts or relative abundance form.

What it does

• —Alpha diversity: observed richness, Shannon, Simpson, and Chao1 when a validated integer count matrix is supplied.
• —Beta diversity: Bray-Curtis, Jaccard, or Aitchison distance with PCoA.
• —PERMANOVA and PERMDISP, both computed by direct label permutation rather than a parametric approximation.
• —Exploratory CLR differential abundance with Benjamini-Hochberg FDR correction, explicitly labelled as exploratory rather than a covariate-adjusted ANCOM-BC2 or MaAsLin2 substitute.
• —Optional cross-validated Random Forest classification, with out-of-fold performance metrics, held-out permutation importance, and post-hoc SHAP explanation.
• —A publication_package.zip with a tables/ folder, a figures/ folder at 300 dpi with editable SVGs, a methods.txt you can adapt for a manuscript, and a README.txt describing every file.

Deploying this Space

1. 1.Create a new Space on Hugging Face, select the Docker SDK.
2. 2.Upload app.py, requirements.txt, and Dockerfile from this package (this README.md can go too, since its front matter configures the Space).
3. 3.Hugging Face builds the image and starts the container automatically. No external database is required for this tool.

Local test before deploying

pip install -r requirements.txt
python app.py

The app listens on port 7860 by default, matching app_port above.

Notes

• —All computation happens inside the container. No abundance data is sent anywhere outside the Space.
• —Machine learning results are only meaningful with enough samples per group; the tool will skip that step rather than report an unreliable score if the dataset is too small.