Combining single-cell ATAC and RNA sequencing for supervised cell annotation.

Area of Science:

• Single-cell multiomics analysis
• Immunology
• Neuroscience

Background:

• Cell type annotation is crucial for single-cell analysis, enabling insights into cellular heterogeneity and function.
• Current methods primarily use single-cell RNA sequencing (scRNA-seq) data.
• While combining scRNA-seq and single-cell ATAC sequencing (scATAC-seq) improves unsupervised annotation, its impact on supervised methods is less explored.

Purpose of the Study:

• To investigate the utility of integrating scRNA-seq and scATAC-seq data for supervised cell type annotation.
• To assess the impact of multiomic data on annotation accuracy and confidence compared to unimodal data.

Main Methods:

• Utilized 10x Genomics multiome datasets with paired RNA and ATAC sequencing from human PBMCs and Alzheimer's Disease neuronal cells.
• Employed dimensionality reduction techniques (linear and nonlinear) and various classification models (Random Forest, SVM, Logistic Regression).
• Evaluated annotation performance using F1 scores and prediction confidence, particularly with scVI embeddings.

Main Results:

• Integration of scRNA-seq and scATAC-seq significantly improved supervised annotation and prediction confidence for human PBMC subtypes.
• Specific cell types, such as CD4 T effector memory cells, demonstrated the most substantial improvement in F1 scores.
• No comparable improvement was observed when annotating neuronal cells from Alzheimer's Disease patients.

Conclusions:

• Multiomic data integration offers a significant advantage for supervised cell type annotation in specific cell populations like PBMCs.
• The effectiveness of multiomic data may vary depending on the cell type and biological context.
• Future research should explore the application of multiomic supervised annotation across diverse cell types and disease states.