Related Experiment Video
Updated: May 17, 2026
11:26
Single-cell RNA-Seq of Defined Subsets of Retinal Ganglion Cells
Published on: May 22, 2017
scMarkerGene: an interpretable neural network framework for cell-type-specific marker gene discovery.
Jingkai Zhang1,2,3, Si Hoi Kou1,2,3, Jiulu Zhao1,2,3,4
1Center for Biomedical Digital Science, Guangzhou Institutes of Biomedicine and Health, Chinese Academy of Sciences, No. 190, Kaiyuan Avenue, Huangpu District, Guangzhou 510530, China.
Briefings in Bioinformatics
|May 15, 2026
Summary
scMarkerGene accurately identifies cell-type-specific marker genes in single-cell transcriptomics using an interpretable neural network. This framework improves upon existing methods by robustly capturing distinguishing features across diverse datasets.
Area of Science:
- Genomics
- Computational Biology
- Bioinformatics
Background:
- Accurate cell-type identification in single-cell transcriptomics is crucial for understanding cellular heterogeneity.
- Existing marker gene discovery methods often suffer from noise, bias, and may identify highly expressed genes rather than truly specific ones.
Purpose of the Study:
- To introduce scMarkerGene, an interpretable neural network framework for robust and accurate marker gene discovery in single-cell transcriptomics.
- To provide a quantitative measure of gene influence on cell-type discrimination.
Main Methods:
- Developed scMarkerGene, a neural network framework that generates a Contribution Score (CS) matrix.
- Implemented a downstream specificity filtering process to refine gene identification.
- Validated the framework on diverse single-cell RNA sequencing (scRNA-seq) datasets, spatial transcriptomics, and pseudotime data.
Main Results:
- scMarkerGene demonstrates robustness to noise, varying cell numbers, annotation resolutions, and sequencing technologies.
- The framework effectively identified cell-type-distinguishing genes, dynamic markers, and spatially resolved markers.
- scMarkerGene transforms neural network predictions into interpretable gene-level importance signals.
Conclusions:
- scMarkerGene offers an efficient, accurate, and interpretable solution for marker gene discovery in single-cell analysis.
- The framework shows broad potential for multi-omics data integration and interpretation.
- The approach successfully identified marker genes across various transcriptomic data types and species.
