GenKI (Gene Knock-out Inference)

A VGAE (Variational Graph Auto-Encoder) based model to learn perturbation using scRNA-seq data.
New! Data has been added.
Paper

Choose the environment file for your platform:

macOS Apple Silicon (arm64):

conda env create -f genki_macos_arm64.yaml
conda activate genki
python -c "import torch; print(torch.backends.mps.is_available())"

Linux x86_64, NVIDIA GPU (CUDA 12.1):

conda env create -f genki_linux_gpu.yaml
conda activate genki
python -c "import torch; print(torch.cuda.is_available())"

Linux x86_64 or macOS Intel, CPU-only:

conda env create -f genki_cpu.yaml
conda activate genki

pip install git+https://github.com/yjgeno/GenKI.git

or install it manually from source:

git clone https://github.com/yjgeno/GenKI.git
cd GenKI
pip install .

Virtual KO experiment:
https://github.com/yjgeno/GenKI/blob/master/notebook/notebook.ipynb

Environment

• Replaced the single environment.yml (Python 3.9.6, PyTorch 1.11) with three platform-specific conda environment files:
  • genki_macos_arm64.yaml — macOS Apple Silicon (M1/M2/M3/M4); uses default PyPI wheels which include MPS acceleration. Sets KMP_DUPLICATE_LIB_OK=TRUE via conda env variables to prevent the PyTorch/OpenMP library conflict on macOS.
  • genki_linux_gpu.yaml — Linux x86_64 with NVIDIA GPU; installs PyTorch 2.4+ from the CUDA 12.1 wheel index.
  • genki_cpu.yaml — CPU-only for Linux x86_64 or macOS Intel; installs PyTorch 2.4+ from the PyTorch CPU wheel index.
• All environments now use Python 3.12, PyTorch ≥ 2.4, scanpy ≥ 1.12, anndata ≥ 0.10, and numba ≥ 0.61.

Package (setup.py)

• Updated all pinned/stale dependency versions to ranges compatible with Python 3.12:
  • anndata==0.8.0 → >=0.10
  • matplotlib~=3.5.1 → >=3.8
  • pandas~=1.4.2 → >=2.0
  • scanpy==1.9.1 → >=1.12
  • scipy~=1.8.0 → >=1.12
  • statsmodels~=0.13.2 → >=0.14
  • ray>=1.11.0 → >=2.30
  • tqdm~=4.64.0 → >=4.66
  • numpy>=1.21.6 → >=1.24,<2.4 (numba upper-bound)
• pip install . now completes successfully on Python 3.12.