arboreto — Claude Code skill: gene regulatory networks

What arboreto does

The skill gives the agent a recipe for candidate gene regulatory networks. Rows are observations, columns are genes, and the result is a ranked list of regulator-target pairs with an importance score. GRNBoost2 fits gradient boosting per target, GENIE3 fits random forests. A gene never predicts itself. These are predictive associations, so binding, activation or causality still needs separate evidence.

Everything runs in an isolated Python 3.11 environment: arboreto==0.1.6 with dask[complete]==2024.7.1, distributed==2024.7.1, numpy==1.26.4, pandas==2.2.3, scikit-learn==1.5.2, scipy==1.13.1. The pins are not decoration. The PyPI wheel builds an empty metadata graph that newer Dask rejects, and 0.1.6 calls .A on sparse targets, which SciPy 1.14 dropped. Set dask.config.set({"dataframe.query-planning": False}) before importing Arboreto, and restart a notebook kernel that already loaded the newer backend. No credentials required; network only for installation.

The bundled scripts/basic_grn_inference.py takes a TSV, rejects duplicate headers, non-numeric values and an empty overlap with the TF list, then writes a headerless file in TF, target, importance order. --limit caps links across the whole network, not per target. One worker is the default, which is the right place to start before scaling to available memory.