Tutorials

DIRAC provides end-to-end workflows on real datasets covering both vertical (multi-omics integration) and horizontal (cross-dataset label transfer) use cases. This section focuses on:

  • Glioblastoma — Spatial RNA + Protein (ADT) integration, joint embedding, clustering, and spatial visualization.

  • Alzheimer’s disease (AD) — Horizontal annotation of stereo-seq bin100 samples using DLPFC (10x Visium, normal) as a reference, with marker-driven training and confidence-based novel type discovery.

What you’ll learn

  • Preprocess per modality (normalize_total → log1p → scale; HVGs/PCA for RNA).

  • Build spatial graphs (k-NN; multi-batch for reference vs single-sample for target).

  • Train DIRAC: - integrate_app for multi-omics integration (Glioblastoma). - annotate_app for label transfer (AD).

  • Evaluate and visualize (spatial maps, optional UMAP; ARI / Accuracy / Precision / Recall / F1).

  • Use marker genes (from the reference) and confidence filtering (e.g., 0.9) to mark low-confidence predictions as "unassigned" when references are incomplete.

Notebooks

DIRAC on Human Glioblastoma — Vertical Tutorial
DIRAC on Alzheimer’s disease — Horizontal Tutorial
DIRAC × DBiT-seq Reproducibility Demo (version-pinned to 0.1.1.post1)