scBIOT documentation#
scBIOT (Single-Cell Biological Insights via Optimal Transport) unifies optimal-transport alignment for preprocessing and embedding single-cell RNA, ATAC, and multi-omic data. The library is designed for embeddings, reproducible benchmarking, and scalable inference across modalities.
Highlights#
Fast optimal transport with GPU.
A unified scBIOT framework that can embed RNA, ATAC, or multi-omics modalities.
Supports scRNA-seq, snATAC-seq, and paired and unpaired multi-omics.
Supports label transfer across disjoint datasets, such as scRNA-seq to Xenium, scRNA-seq to snATAC-seq.
Built-in preprocessing steps (iterative LSI, gene activity annotation from peaks, coembedding of PCA from multiomics).
Support both CPU and GPU.
Tutorials
- scBIOT
- Highlights
- Getting started
- 1. scRNA-seq
- 2. scRNA-seq in R (Seurat + reticulate)
- 3. snATAC-seq
- 4. Paired Multiomics
- 5. Unpaired Multiomics Integration
- Load
- Build reference UMAP if want to map embeddings to reference
- Preprocess
- Integrate
- Integrate disjoint datasets with label transfer (scRNA-seq reference → snATAC-seq gene activity)
- Prediction of cell types with supBIOT
- Project joint embedding back to reference embedding (optional)
- Evaluate
- Visualize
- Evaluate
- 6. Centroid-Level Optimal Transport
- 7. brain_1.3M_integration
- 8. Label transfer with supBIOT
- a. Supervised scBIOT (supBIOT)
- b. label transfer from scRNA-seq to Xenium
- Load Xenium spatial data
- Preprocess
- Load scRNA-seq reference
- Subset for a quick speed test
- Preprocess
- Build the reference UMAP
- Concatenate adata
- PCA
- Integrate disjoint datasets with label transfer (reference scRNA-seq –> Xenium)
- Evaluate
- Visualize
- Visualize (markers)
- Evaluate (DE)
- Visualize (correlation)
- c. Supervised Panc8 benchmarking
API reference