Run SpliceAI with tensorflow-gpu on WSL2

spliceai_on_wsl2.md

SpliceAI with GPU Acceleration on Windows 11 (WSL2)

WSL2 is a Windows feature that allows near-native execution of ELF binaries like Unix bash and Gnome apps. With GPU acceleration support, it turns a gaming PC into a powerful bioinformatics workstation without needing to dual-boot.

This guide details how to install WSL2 on a fresh Windows 11 (25H2) system and use Pixi, a high-performance package manager, to install SpliceAI and its GPU dependencies.

1. Install Ubuntu 24.04 on WSL2

1. Install WSL2:

   • Press Windows Key + R, type cmd, and press Enter to open Terminal.
   • Type the following and press Enter. This command enables Windows 11 virtualization features and downloads the latest Linux kernel optimized for WSL2.

     wsl --install

   • Click Yes if prompted by User Account Control.
   • Reboot your computer. You can use the command below in Terminal to reboot immediately.

     shutdown /r /t 0

2. Install Ubuntu:

   • Press Windows Key + R, type cmd, and press Enter to open Terminal.
   • Install Ubuntu 24.04 as the default distribution.

     wsl --install -d Ubuntu-24.04

   • Create a username and password when prompted (these do not need to match your Windows credentials).

3. Verify GPU Support:

   • In the Ubuntu terminal, verify that your NVIDIA GPU is accessible.

     nvidia-smi -L

   • You should see your GPU listed. If not, ensure you have the latest Windows NVIDIA drivers installed. You do not need to install Linux drivers.

2. Install Pixi

We will use pixi instead of Conda/Mamba for faster, lock-file-based dependency management.

1. Run the Installer:

   curl -fsSL https://pixi.sh/install.sh | bash

2. Update Shell:

   source ~/.bashrc

3. Create Project and Install Dependencies

1. Initialize a New Project: Create a directory for your analysis and initialize it.

   mkdir -p src/spliceai
   cd src/spliceai
   pixi init

2. Configure Channels: Add the necessary channels (bioconda for SpliceAI, conda-forge for dependencies):

   pixi project channel add conda-forge
   pixi project channel add bioconda

3. Install SpliceAI and GPU Dependencies: Conda-forge's tensorflow package usually includes GPU support on Linux, but explicitly adding cudatoolkit and cudnn ensures the necessary libraries are present.

   pixi add spliceai spliceai-wrapper tensorflow cudatoolkit cudnn

4. Verify the Setup: Run a Python one-liner inside the pixi environment to confirm TensorFlow can see the GPU.

   pixi run python -c 'import tensorflow as tf; print(tf.config.list_physical_devices("GPU"))'

   You should see output indicating a GPU device is found.

4. Running SpliceAI

You can now test SpliceAI using pixi run within your project directory.

pixi run spliceai --help