Related Experiment Video
Updated: Aug 8, 2025
10:12
Droplet Barcoding-Based Single Cell Transcriptomics of Adult Mammalian Tissues
Published on: January 10, 2019
18.6K
A multi-use deep learning method for CITE-seq and single-cell RNA-seq data integration with cell surface protein
Justin Lakkis1, Amelia Schroeder1, Kenong Su1
1Department of Biostatistics, Epidemiology and Informatics, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA 19104, USA.
Nature Machine Intelligence
|March 6, 2023
Summary
sciPENN integrates CITE-seq and single-cell RNA-seq data, overcoming batch effects and partial protein panel overlap. This deep learning approach enhances data utilization for discovering cell heterogeneity.
Area of Science:
- Single-cell multi-omics
- Computational biology
- Biomedical research
Background:
- CITE-seq measures RNA and protein expression simultaneously in single cells, widely used in immune disorders, influenza, and COVID-19 research.
- Generating CITE-seq data is costly, and integrating multiple datasets presents computational challenges like batch effects and partially overlapping protein panels.
- Integrating CITE-seq with single-cell RNA-seq (scRNA-seq) is crucial for maximizing data utility and uncovering cell population heterogeneity.
Purpose of the Study:
- To present sciPENN, a novel deep learning approach for seamless CITE-seq and scRNA-seq data integration.
- To address computational challenges in multi-dataset integration, including batch effects and varying protein panels.
- To enable protein expression prediction and imputation, quantify uncertainty, and facilitate cell type label transfer.
Main Methods:
- Developed sciPENN, a versatile deep learning framework.
- Implemented functionalities for data integration, protein prediction/imputation, uncertainty quantification, and cell type label transfer.
- Conducted comprehensive evaluations across multiple datasets to assess performance against state-of-the-art methods.
Main Results:
- sciPENN effectively integrates CITE-seq and scRNA-seq datasets, mitigating batch effects and handling partial protein panel overlaps.
- The method demonstrates superior performance in protein expression prediction and imputation compared to existing approaches.
- Evaluations confirm sciPENN's capability in quantifying prediction/imputation uncertainty and enabling accurate cell type label transfer.
Conclusions:
- sciPENN offers a robust and efficient solution for integrating diverse single-cell omics data.
- The approach significantly enhances the ability to uncover cell population heterogeneity by maximizing data utilization.
- sciPENN outperforms current state-of-the-art methods, providing a valuable tool for biomedical research.
