bio-single-cell-multimodal-integration

Multi-modal single-cell integration uses weighted nearest-neighbor graphs to jointly analyze RNA with protein or chromatin data. This approach builds a unified embedding to improve cell-state discovery across modalities like CITE-seq and Multiome.

Per-modality preprocessing and normalization are required before multi-modal integration: standard normalization for RNA, CLR transformation for protein (ADT) data, and TF-IDF for ATAC chromatin data to ensure accurate modality alignment.

Yes, multi-modal single-cell integration supports spatial multimodal datasets alongside CITE-seq, Multiome, and SHARE-seq. It aligns spatial modalities with RNA data to infer joint cell states and generate integrated visualizations.

After building a unified embedding, you can compare modality weights to track modality contributions and perform cross-modality marker discovery. This helps identify cell types and states defined by both RNA and protein or chromatin signals.

The best way to analyze CITE-seq data is loading the RNA and ADT data, applying per-modality normalization, and running a joint multimodal integration workflow. This builds a unified embedding for joint clustering and cell type identification.

Weighted nearest-neighbor graphs are used when you need to jointly analyze multi-modal single-cell data, such as RNA combined with protein or chromatin. They enable coherent cell-state discovery by integrating signals across distinct modalities into one graph.