Joint probabilistic modeling of single-cell multi-omic data with totalVI.

Area of Science:

• Single-cell multi-omics
• Immunology
• Computational biology

Background:

• Cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq) pairs RNA and surface protein measurements.
• Integrating these distinct data types into a unified cell state representation is challenging due to technical variations.

Purpose of the Study:

• To develop a computational framework for joint analysis of CITE-seq data.
• To probabilistically model both biological and technical factors in CITE-seq measurements.

Main Methods:

• Introduced Total Variational Inference (totalVI), a probabilistic framework for end-to-end CITE-seq data analysis.
• Applied totalVI to profile murine immune cells using CITE-seq, measuring over 100 surface proteins.

Main Results:

• Demonstrated totalVI's capability for dimensionality reduction and dataset integration with varying protein panels.
• Showcased totalVI's effectiveness in estimating molecule correlations and performing differential expression testing.

Conclusions:

• totalVI offers a cohesive solution for analyzing paired RNA and protein data in single cells.
• The framework facilitates a deeper understanding of cell states by integrating transcriptional and proteomic information.