Getting Started

Tutorials for using single-cell data with PyTorch. These materials are aimed at two groups: machine learning practitioners new to genomics, and genomics researchers wanting to use deep learning.

Tutorials

Tutorial	Audience	Duration	Open in Colab	What it covers
Basics: Single-cell Data in AnnData format	ML practitioners new to single-cell data	30-45 min		What scRNA-seq data is and how it differs from typical ML datasets
Basics: Machine Learning with PyTorch	Genomics researchers new to ML	45-60 min		PyTorch basics explained using genomics examples
Single-cell data in PyTorch	All participants	45-60 min		Use AnnData’s AnnLoader for per-cell classification and regression
Patient-level learning	Participants interested in sample-level prediction	30-45 min		Model patient-level cell populations with PyTorch datasets and ggml-ot

Suggested workshop flow: Choose the primer for the domain that is less familiar to you. In a mixed workshop, participants can work through the two primers in parallel, then regroup for Single-cell data in PyTorch. Continue with Patient-level learning as an optional advanced tutorial.

Local installation

Google Colab requires no local setup. To run the tutorials locally on macOS or Linux:

git clone https://github.com/DaminK/AnnData_Pytorch_Tutorials.git
cd AnnData_Pytorch_Tutorials
python -m venv .venv && .venv/bin/python -m pip install -r requirements.txt
.venv/bin/python -m jupyter lab

The third line creates the virtual environment and installs the dependencies into it in one shell command. Activation is not required. On Windows, replace .venv/bin/python with .venv\Scripts\python.exe.

What you’ll learn

These tutorials show you how to:

• Load scRNA-seq data with Scanpy and inspect AnnData objects

• Build supervised models with PyTorch

• Handle sparse, high-dimensional biological data

• Batch cells with AnnData’s AnnLoader and model patient-level data with ggml-ot

• Evaluate classification and regression workflows

The tutorials include notes explaining genomics concepts for ML practitioners and ML concepts for genomics researchers.

Acknowledgments

These tutorials use Scanpy, AnnData, PyTorch, and ggml-ot. Documentation is built with Jupyter Book.

Additional Resources

Useful links:

• Single-cell best practices

• Scanpy tutorials

• PyTorch tutorials

• AnnLoader documentation - PyTorch batches from AnnData

• ggml-ot - Patient-level distribution learning

License

This project is available under the MIT License.