cross-modality-embedding-integration

cross-modality-embedding-integration

Summary

Integration of paired scATAC-seq chromatin accessibility and scRNA-seq gene expression data into a unified reduced-dimension embedding space using ArchR's multiome workflow. This skill enables joint analysis of epigenetic and transcriptomic modalities in the same cell population through sequential ingestion, alignment, and dimensionality reduction.

When to use

You have paired scATAC-seq peak matrices and scRNA-seq gene expression matrices from the same cells (multiome data) and need to perform joint clustering, visualization, or correlation analysis across both chromatin accessibility and gene expression in a single coordinate system. Apply this skill when single-modality analysis is insufficient and you require integrated interpretation of regulatory and expression signals.

When NOT to use

• Cells from scATAC-seq and scRNA-seq are not the same population or lack reliable alignment anchors
• Only single-modality data is available (scATAC-seq OR scRNA-seq, not both)
• Gene expression matrix is already embedded or summarized to a lower dimension incompatible with raw counts

Inputs

• scATAC-seq peak feature matrix (rows=peaks, columns=cells)
• scRNA-seq gene expression feature matrix (rows=genes, columns=cells)
• Cell metadata with consistent cell identifiers across modalities

Outputs

• ArchR project object with integrated gene expression data attached
• Unified reduced-dimension embedding (combined dimensions) integrating both modalities
• Joint iterative LSI components derived from peaks and gene expression

How to apply

Begin by loading the scATAC-seq peak matrix and cell metadata, then call importFeatureMatrix to register the feature matrix into an ArchR project object. Next, load the scRNA-seq gene expression matrix (as a standard feature matrix) and call addGeneExpressionMatrix to append gene expression data to the same project while aligning cells across both modalities. Execute addIterativeLSI on the combined project to compute latent semantic indexing jointly across accessibility peaks and gene expression signals. Finally, call addCombinedDims to generate a unified reduced-dimension embedding that integrates both scATAC-seq and scRNA-seq signal into a single coordinate space suitable for downstream analysis (clustering, trajectory inference, or correlation studies).