Multimodal Integration

Projects heterogeneous data types into a shared low-dimensional representation where cells profiled with different modalities can be directly compared.

• Enables joint clustering of cells from scRNA-seq and scATAC-seq experiments
• Uses canonical correlation analysis (CCA) or variational autoencoders to align modalities
• Preserves both shared and modality-specific biological variation

Predicts one data modality from another, enabling in silico inference of unmeasured molecular layers.

• Predicts chromatin accessibility from RNA expression and vice versa
• Leverages paired multimodal datasets (e.g., CITE-seq, 10x Multiome) as training anchors
• Enables regulatory inference without performing additional assays

Constructs a cell-specific weighted combination of modality affinities to define a single similarity graph.

• Learns the relative information content of each modality per cell
• Implemented in Seurat v4+ for CITE-seq and multiome data
• Outperforms simple concatenation when modalities have differing signal-to-noise ratios

Uses cells measured with multiple modalities as bridges to transfer annotations between unimodal datasets.

• Identifies mutual nearest neighbors across modalities in a reference
• Transfers cell-type labels from scRNA-seq references onto scATAC-seq queries
• Enables integrative analysis when fully paired data is scarce

Decomposes multi-modal data into a set of latent factors that capture shared and private sources of variation.

• Identifies factors driving coordinated changes across RNA, protein, and chromatin
• Handles missing data modalities through matrix factorization
• Reveals axes of biological variation invisible to single-modality analysis

Links distal regulatory elements to target genes by integrating chromatin accessibility and gene expression from the same cells.

• Constructs cis-regulatory networks using peak-to-gene linkage
• Identifies enhancer-promoter pairs active in specific cell types
• Provides mechanistic hypotheses for non-coding genetic variants

Cellular Indexing of Transcriptomes and Epitopes by Sequencing is a foundational multimodal assay that simultaneously profiles the whole transcriptome and a panel of surface proteins from the same single cell. It uses oligonucleotide-conjugated antibodies, where the barcode is detected by sequencing, bridging the gap between RNA expression and proteomic phenotype.

An unsupervised analysis framework in Seurat v4+ that learns cell-specific modality weights. Instead of simple concatenation, WNN integrates modalities like RNA and protein by assigning higher importance to the modality with more predictive power for each cell, enabling a unified latent space for clustering and visualization.

Single-cell Assay for Transposase-Accessible Chromatin profiles open chromatin regions to reveal the regulatory landscape of individual cells. Integrating scATAC-seq with scRNA-seq links transcription factor binding to gene expression output, enabling the construction of Gene Regulatory Networks (GRNs).

A deep generative model from scvi-tools designed for the end-to-end analysis of CITE-seq data. TotalVI jointly models RNA counts and protein background noise, producing a shared latent representation that corrects for batch effects and enables imputation of missing protein data based on correlated transcriptomic signatures.

Multi-Omics Factor Analysis v2 is a statistical framework that infers a low-dimensional representation of multi-modal data. It identifies latent factors that capture the principal sources of variation across data types, such as RNA, chromatin, and methylation, revealing coordinated biological drivers without requiring paired measurements on the same cells.

A computational strategy that identifies mutual nearest neighbors (MNNs) or 'anchors' between datasets. In multimodal contexts, these anchors represent cells in a shared biological state across different technologies, enabling the transfer of cell-type labels and the harmonization of disparate modalities into a common reference atlas.