Tutorials

The tutorial path is intentionally Scanpy-first and model-focused.

Start with Scanpy for the dataset object and downstream neighborhood analysis, then use scDLKit for training, evaluation, and model comparison.

Scanpy PBMC quickstart

Learn the main scDLKit workflow on pbmc3k_processed, then store the latent representation in adata.obsm for Scanpy analysis. This notebook ships with a quickstart profile and a longer full profile.

Scanpy PBMC quickstart

PBMC model comparison

Compare PCA, AutoEncoder, VAE, and Transformer AE baselines on the same PBMC workflow.

PBMC model comparison

PBMC classification

Run the classification baseline and inspect accuracy, macro F1, and a confusion matrix.

PBMC classification baseline

Synthetic smoke tutorial

Use the minimal synthetic notebook only when you want the smallest dependency path or a smoke run.

Synthetic smoke tutorial

Learning order

1. Scanpy PBMC quickstart

2. Re-run the PBMC quickstart in full mode when you want a longer baseline fit

3. PBMC model comparison

4. PBMC classification

5. Synthetic smoke tutorial