Tested source runtime

MAGeCK 0.5.9.5 was downloaded from the official SourceForge release and built on macOS ARM64 with Python 3.11, NumPy 1.26.4, SciPy 1.13.1, setuptools 69.5.1, and the system C++ compiler. count, paired/unpaired test (including compiled RRA), control normalization, sparse-normalization fallbacks, and a small mle smoke ran successfully on review date 2026-10-01. MAGeCK is not a PyPI dependency installed by this repository's isolated test environment; integration tests skip when the executable is absent. Put the complete source/conda environment's bin on PATH to run them.

The following is the tested installation pattern; choose an environment directory outside the skill.

uv venv --python 3.11 mageck-env
uv pip install --python mageck-env/bin/python numpy==1.26.4 scipy==1.13.1 setuptools==69.5.1
curl -fL 'https://sourceforge.net/projects/mageck/files/0.5/mageck-0.5.9.5.tar.gz/download' -o mageck.tar.gz
tar -xzf mageck.tar.gz
# This archive contains liulab-mageck-c491c3874dca, not a directory named mageck-0.5.9.5.
cd liulab-mageck-c491c3874dca
../mageck-env/bin/python setup.py install
cd ..
export PATH="$PWD/mageck-env/bin:$PATH"
mageck --version
mageck test --help

The upstream installation uses legacy setup.py and emits setuptools deprecation messages. Its source package compiles RRA and mageckGSEA. A wrapper-only installation without these binaries cannot perform gene ranking. Bioconda is another upstream-documented route, but was not exercised for this skill. R/LaTeX rendering, copy-number correction, and realistic multi-factor MLE fits are outside the tested integration surface. The MLE smoke used 30 genes, five guides each, two samples per condition and one permutation round; this does not establish FDR calibration.

Release and source review

On 2026-10-01 the official release directory listed 0.5.9.5 (2021-12-01). Its freshly downloaded archive matched SHA-256 b06a18036da63959cd7751911a46727aefe2fb1d8dd79d95043c3e3bdaf1d93a. The tested source tree is archive directory liulab-mageck-c491c3874dca. Runtime help and source (argsParser.py, mageckCount.py, mageckCountIO.py, mageckCountNorm.py, fileOps.py, crisprFunction.py, mleargparse.py, mledesignmat.py, mleinstanceio.py) resolve stale details in the wiki:

• The current test default is --remove-zero both, although the usage table still says none.
• Numeric sample names are unsafe: integer contrasts are treated as indices, and an all-numeric header may be read as counts. The helper requires nonnumeric names and an explicit header.
• --control-sgrna supplies the RRA null as well as an optional normalization reference. A control gene cannot mix control and non-control guide IDs, and at least two controls are required.
• Median and control-median normalization can fall back to totals; the helper extracts actual factors and fallback warnings from the execution log rather than reporting the request as fact.
• A named MLE design file controls count-column selection/order through its sample labels.

MAGeCK Home now points to the separate MAGeCK2 project. Its PyPI package is mageck2 (0.3.0 at review), while the mageck PyPI JSON endpoint returned 404. MAGeCK2 is not exercised by this skill; installing it does not verify the mageck/RRA runtime used here. No remote analysis API is used.