Installation¶

Requirements¶

Python 3.8 or newer
TensorFlow 2.13 (installed by the extras below) for anything that runs the model
A GPU is optional. Prediction is fast on CPU; contribution maps mutate every position of a 1,344 bp window, so a GPU makes them noticeably quicker.

Install¶

Pick the extras for what you actually need — the base install pulls in only numpy and pandas.

EverythingPrediction onlyContribution mapsVariant effectsMotif synergy

pip install "peagent-tool[all]"

pip install "peagent-tool[prediction]"

pip install "peagent-tool[attribution]"

pip install "peagent-tool[variant]"

pip install "peagent-tool[synergy]"

Extra	Adds	Needed for
`prediction`	`tensorflow-cpu`	`predict_celltypes`, everything below
`attribution`	`matplotlib`, `logomaker`	`compute_attribution`, `plot_attribution`
`variant`	`pysam`	`score_variants` from a genome FASTA
`synergy`	`scipy`	`MotifSynergyTester`
`hub`	`huggingface_hub`	automatic download of model weights
`all`	all of the above

GPU builds

The extras install tensorflow-cpu so that a CPU-only install stays small. For GPU, install TensorFlow yourself first (pip install tensorflow==2.13.1 inside a CUDA-enabled environment), then pip install --no-deps peagent-tool plus the remaining extras.

From source¶

git clone https://github.com/YAOJ-bioin/peagent_tool.git
cd peagent_tool
pip install -e ".[all,test]"
pytest

Verify¶

peagent-tool --help
peagent-tool assets --species maize

peagent-tool assets reports where the model weights and cell-group metadata will be read from, and whether they are already present. If they are not, see Models and data.

The package name is peagent-tool; the Python import is peagent_tool.