Skip to content

installation.md

Latest commit

 

History

History
76 lines (60 loc) · 2.62 KB

installation.md

File metadata and controls

76 lines (60 loc) · 2.62 KB
Raw

Installation

Using Poetry (recommended)

git clone https://github.com/bigbio/hvantk
cd hvantk
poetry install
eval "$(poetry env activate)"

Using pip

git clone https://github.com/bigbio/hvantk
cd hvantk
pip install -e .

Optional features (extras)

The core install is lightweight; visualization, machine-learning, and a few provider-specific dependencies ship as optional extras. Install the ones you need:

# pip
pip install -e ".[viz,hgc]"

# poetry
poetry install --extras "viz hgc"      # or: poetry install --all-extras
Extra Pulls in Needed for
viz matplotlib, seaborn, plotly plots and HTML reports
interactive plotly interactive QC dashboards
ml scikit-learn, scipy ML-backed analyses
ancestry scikit-learn, matplotlib, seaborn, scipy hvantk ancestry-inference
psroc scikit-learn, matplotlib, plotly, scipy hvantk psroc
hgc matplotlib, seaborn hvantk hgc QC plots/reports
ptm cptac, sorted-nearest CPTAC PTM downloads
constraint tspex, matplotlib, seaborn, scipy hvantk ptm constraint
enrichex scipy, matplotlib, seaborn hvantk enrichex overlap / burden
cohort scipy hvantk cohort burden
duckdb duckdb DuckDB-backed catalog queries
expression scanpy, scipy hvantk expression summarize and markers

Three command paths need this extra:

  • hvantk expression summarize — via summarize_expression_ad
  • hvantk expression markers — scanpy's rank_genes_groups
  • hvantk ptm constraint --expression-source anndata --expression-metric mean — routes through summarize_expression_ad for the mean aggregation only. The default metric is median, which uses a direct numpy path and does not need scanpy.

hvantk expression describe and summarize-ucsc do not touch scanpy and work on a base install. Without the extra, these paths exit with an actionable message naming it, not a traceback.

expression is unavailable on Intel macOS. scanpy pulls numba → llvmlite, which ships no x86_64 macOS wheel from 0.47 onward and fails to build from source. The extra installs normally on Linux and Apple Silicon. This is why scanpy is an extra rather than a base dependency: as a base dep it made the whole package uninstallable on Intel Macs. On Intel, run those two commands on the cluster; the rest of the toolkit is unaffected.

scikit-learn is optional — install ml, ancestry, or psroc if you use those analyses.

Prerequisites

  • Python ≥ 3.10
  • Hail ≥ 0.2.137
  • Java 8 or 11 (required by Hail/Spark)