Docker Tutorial¶

spMetaTME is distributed with a Docker environment to ensure that all dependencies (Python, R for domain clustering, PyTorch Geometric) are exactly as needed. This is the simplest way to get up and running if you have Docker and an NVIDIA GPU.

Prerequisites¶

Docker Installed: Follow instructions on Docker’s website.
NVIDIA Container Toolkit (Optional but recommended): For using GPUs inside Docker. See instructions here.

Building the Image¶

If you want to build the spmetatme image from source:

Clone the repository:

git clone https://github.com/SurajRepo/spmetatme.git
cd spmetatme

Build the Docker image:
```
docker build -t spmetatme:latest .
```

Basic Usage¶

The Docker image uses the spmetatme CLI as its entry point. You can run any subcommand directly.

Get Help¶

docker run --rm spmetatme --help

Inference Help¶

docker run --rm spmetatme infer --help

Working with Local Data¶

Since the container has its own isolated file system, you must mount your local directory to the container’s /app directory to analyze your files. Select –model-species human or mouse based on your data. and –metabolic-model based on the model you want to use.

Windows (PowerShell/CMD)¶

Use the %CD% variable to mount your current folder:

docker run --rm -it --gpus all -v "%CD%:/app" spmetatme infer --input data/my_sample.h5ad --model-species human --metabolic-model breast_cancer

Linux / Mac¶

Use the $(pwd) variable to mount your current folder:

docker run --rm -it --gpus all -v $(pwd):/app spmetatme infer --input data/my_sample.h5ad --model-species human --metabolic-model breast_cancer

Understanding the Flags:¶

--gpus all: Enables GPU acceleration inside the container (highly recommended).
-v "%CD%:/app": Binds your local folder to the container’s work directory. Data and results are shared.
-it: Enables interactive mode so you can see live logs.
--rm: Automatically removes the container after it stops.

Windows Helper: `run_docker.cmd`¶

For Windows users, we provide a pre-configured script (run_docker.cmd) in the repository to simplify these commands.

Example Inference:

run_docker.cmd infer --input data/CID4535.h5ad --model-species human --metabolic-model breast_cancer

Example Pretraining:

run_docker.cmd pretrain --training-list data/*.h5ad --metabolic-model breast_cancer --repo-id MyRepo/v1

Advanced: Accessing the Container Shell¶

If you want to explore the container manually:

docker run --rm -it --gpus all -v "%CD%:/app" --entrypoint bash spmetatme

Once inside, you can run spmetatme or standard Python commands manually.