Machine Learning on Single-Cell Data Workshop by Damin

Damin Kühn recently led a workshop on machine learning for omics data, organized in collaboration with the CCLS. To help bridge the two disciplines, the session began with introductory primers for both computational genomics and machine learning researchers. Working directly in Jupyter notebooks, participants then learned how to handle scRNA-seq data in the AnnData format and use it to build PyTorch dataloaders, implement models, and train them.

The goal of the workshop was twofold: helping genomics researchers move beyond standard toolboxes like Scanpy to build their own deep learning models, and introducing ML researchers to complex single-cell datasets to evaluate novel methods. The event concluded by demonstrating how to write custom dataloader classes and loss functions, using Supervised Optimal Transport as an advanced practical example of a method that conceptually links microscopic single cells to macroscopic patient conditions. All hands-on materials and tutorials are freely available online and can be run directly in the browser.