scMarkerGene: Ten Publicly Available scRNA-seq Datasets and Supporting Code

This study developed a deep learning tool, scMarkerGene, based on PyTorch (Python = 3.10) for the identification of cell type–specific marker genes in single-cell data. All parallel comparison methods and models used in the analysis were obtained from publicly available open-source repositories (GitHub or Zenodo), ensuring the reproducibility and scientific rigor of the results. For benchmarking, we collected a total of 10 publicly available single-cell RNA sequencing (scRNA-seq) datasets spanning multiple species, tissues, and technologies. The datasets include arabidopsis leaf, arabidopsis root, drosophila heart, drosophila intestine, mouse myeloid, mouse cortex, human oral cavity, human carotid, human bone marrow and human COVID-19 lung samples. Sequencing technologies used include 10X Genomics, MARS-seq, SMART-Seq2, and single-nucleus RNA-seq. Cell numbers range from 2,018 to 116,314, genes from 3,451 to 53,678, and the number of annotated cell types ranges from 5 to 59.