Installation

Shorkie’s model code lives in two pinned forks — baskerville-yeast and westminster — which ship as submodules of the shorkie-paper repository. That repo also provides an installable helper package (shorkie), the release catalogue, the examples, and the figure notebooks.

Requirements

Component

Requirement

OS

Linux

Python

3.9 (pinned by environment.yml; the conda env is the supported path)

TensorFlow

~2.15

CPU

Enough for inference, variant scoring, and reproducing the figures from released data

GPU

Needed only for training / fine-tuning and a few GPU-marked figure panels. Those also need tensorrt==8.6.1 and a CUDA-enabled TensorFlow build.

Note

Everything on this site that is labelled CPU was verified on CPU from a clean download — no GPU and no cluster access required.

Install

git clone --recurse-submodules https://github.com/calico/shorkie-paper.git
cd shorkie-paper

conda env create -f environment.yml
conda activate yeast_ml

# model code first, then the helper package
pip install -e external/baskerville-yeast -e external/westminster -e .

If you cloned without --recurse-submodules:

git submodule update --init

Configure paths

Every filesystem path in the repository resolves through config/paths.yaml — there are no hardcoded machine paths in the pipeline code.

cp config/paths.example.yaml config/paths.yaml

The defaults already point at the layout data/download.sh creates, so for normal use you only need to set release_root (where downloads land). Resolution order is $SHORKIE_CONFIGconfig/paths.yamlconfig/paths.example.yaml.

from shorkie import config
config.path("models.shorkie_finetuned")   # -> Path to the 8-fold ensemble
config.path("genome.fasta")               # -> Path to the R64 FASTA

Containers

containers/ ships a Dockerfile and an apptainer.def that build the environment and both submodules, for a scheduler-free run.

Verify the install

bash scripts/00_setup/verify_install.sh   # imports shorkie, resolves keys, runs pytest
pytest -q                                 # release-integrity + smoke tests

Next: Quick start.