Related Experiment Video
Updated: Jun 4, 2025
A Data-Driven Approach to Quantifying Immune States in Sepsis
Published on: February 7, 2025
PAGE-based transfer learning from single-cell to bulk sequencing enhances model generalization for sepsis diagnosis
Nana Jin1,2, Chuanchuan Nan3, Wanyang Li1
1Guangdong Provincial Clinical Research Center for Geriatrics; Shenzhen Clinical Research Center for Geriatrics, Shenzhen People's Hospital, the Second Clinical Medical College of Jinan University, the First Affiliated Hospital of Southern University of Science and Technology, 1017 Dongmen Rd N, Luohu District, Shenzhen 518020, China.
A new method, scPAGE2, improves sepsis diagnosis by fusing single-cell and bulk transcriptome data. This approach enhances model generalizability and diagnostic accuracy, outperforming existing methods.
Area of Science:
- Genomics
- Computational Biology
- Immunology
Background:
- Sepsis diagnosis relies on host transcriptional patterns, but generalization is limited.
- Current methods struggle to integrate diverse transcriptomic data effectively.
Purpose of the Study:
- To develop an improved sepsis diagnostic model using transfer learning.
- To enhance data fusion between single-cell and bulk transcriptome data.
Main Methods:
- Introduced scPAGE2, an updated version of single-cell Pair-wise Analysis of Gene Expression (scPAGE).
- Utilized transfer learning to pretrain a model on single-cell data and retrain on bulk transcriptome data.
- Validated the model across seven datasets, three transcriptome platforms, and fluorescence-activated cell sorting (FACS).
Main Results:
- scPAGE2 demonstrated robust performance across platforms with an average AUROC of 0.947 and AUPRC of 0.987.
- Identified specific gene expression patterns in monocytes (JAM3-PIK3AP1, IRF6-HP) and other cells (ARG1-CCR7) associated with sepsis.
- Confirmed elevated IRF6-HP in sepsis monocytes via scRNA-seq and FACS.
Conclusions:
- scPAGE2 offers an effective and robust strategy for constructing sepsis diagnostic models.
- The method improves classification model generalizability for various clinical prediction scenarios.
- Demonstrated superior performance in classifying acute myeloid leukemia as well.
