TOGGLE Unveils Functional Heterogeneity and Epigenetic Memory in Single Cells

  1. Junpeng Chen
  2. Tianda Sun
  3. Tian Song
  4. Zhouweiyu Chen
  5. Haowei Xu
  6. Zixuan Guo
  7. Eric Jiang
  8. Yibing Nong
  9. Tao Yuan
  10. Charles C. Dai
  11. Yexing Yan
  12. Jinwen Ge
  13. Haihui Wu
  14. Tong Yang
  15. Shanshan Wang
  16. Zixiang Su
  17. Park Tian
  18. Xiao Yang
  19. Ahmed Abdelbsset-Ismail
  20. You Li
  21. Changping Li
  22. Richa A. Singhal
  23. Kailin Yang
  24. Lu Cai
  25. Zixin Wu
  26. Alex P. Carll
3 evaluations Published on
Read the full article Related papers
This article on Sciety

Abstract

Cells that appear transcriptionally identical can maintain vastly different functions or fates—an enduring blind spot in single-cell transcriptomics. Here we introduce TOGGLE, a self-supervised graph diffusion framework that delineates fine-grained functional heterogeneity within phenotypically stable cell populations. By combining deep diffusion learning with reinforcement-guided clustering and a BERT-inspired masking strategy, TOGGLE reconstructs hidden trajectories of cell fate without prior labels or temporal information. Applied across multiple single-cell RNA-seq datasets, TOGGLE achieves up to 90 % accuracy in unsupervised fate prediction, surpassing existing trajectory algorithms such as WOT and Cospar. It distinguishes ferroptotic, apoptotic, and intermediate neuronal states in ischemic stroke, with predictions experimentally validated in animal models. In neural stem cells TOGGLE reveals epigenetic memory of metabolic activity, linking local DNA demethylation and chromatin accessibility to mitochondrial RNA expression. We further introduce the Graph Diffusion Functional Map, which isolates subtle RNA functional groupings otherwise obscured by conventional dimensionality reduction. TOGGLE thus establishes a generalizable framework for mapping functional identity and epigenetic dynamics at single-cell resolution, providing new insights into cellular memory, regeneration, and disease mechanisms.

Biographical Note

Chen’s guidance comes from Dr. Alex P. Carll. Another mentor is Dr. Lu Cai, one of the contributors to the standardization of autophagy detection methods, whose research on criteria for identifying diabetes induced apoptosis has been included in numerous clinical guidelines. Professor Jinwen Ge is a State Council Special Allowance Expert of China. He has been involved in formulating medical standards for integrated traditional Chinese and Western medicine in China and has participated in developing retrieval/indexing norms for journals in the field of integrated traditional Chinese and Western medicine.

Code and Data availability

The raw data for the hematopoiesis dataset can be accessed at the Gene Expression Omnibus database with accession number GSE140802, the reprogramming dataset with accession number GSE99915, the e-cigarette aerosol inhalation on mouse cardiac tissue with accession number GSE183614, the rat myocardial infarction dataset with accession number GSE253768, the ischemic stroke dataset with accession number GSE232429, and the adult NSC lineage dataset with accession number GSE209656 and GSE211786.

All codes and guide page & analyzed data are full open available at: <ext-link xmlns:xlink="http://www.w3.org/1999/xlink" ext-link-type="uri" xlink:href="https://github.com/FullBlackWolf/TOGGLE?tab=readme-ov-file">https://github.com/FullBlackWolf/TOGGLE?tab=readme-ov-file</ext-link> <ext-link xmlns:xlink="http://www.w3.org/1999/xlink" ext-link-type="uri" xlink:href="https://fullblackwolf.github.io/TOGGLE/">https://fullblackwolf.github.io/TOGGLE/</ext-link>

Key points

  • <label>(1)</label>

    Previous studies have primarily focused on distinguishing cell types and different cellular subpopulations, whereas our tool is designed to identify subtle functional differences among highly similar cell states. For example, in Test2, it first isolates neurons undergoing cell death, then further differentiates early, intermediate, and late death stages, and separates ferroptosis from apoptosis as two major categories. Notably, ferroptosis and apoptosis share a large number of molecular pathways, making this task highly challenging. In addition, in Test4, our method discriminates fibroblasts with distinct immune-communication patterns, even though they do not show prominent separation or temporal progression on conventional UMAP plots.

  • <label>(2)</label>

    We developed an innovative Graph Diffusion Functional Map that significantly reduces noise, resulting in clearer visualization of RNA functional groupings and enabling the detection of subtle functional differences in high-dimensional data. This provides a novel tool for pathway discovery. For instance, in Test4, we found that the KEGG inflammatory response pathway was most strongly activated in young mice, suggesting that e-cigarette exposure exerts more severe effects on immature cardiac tissue.

  • <label>(3)</label>

    We implemented a novel trajectory-tracking strategy capable of distinguishing early, intermediate, and late stages of cellular programming, which we refer to as functional lineage tracing. In Test3, we successfully separated nearly indistinguishable aNSC and qNSC populations and further uncovered their methylation-regulated transcriptional programs, demonstrating that local DNA demethylation activates genes involved in cell proliferation and energy metabolism, thereby regulating the activation of quiescent neurons.

  • <label>(4)</label>

    In the testing dataset, it was found that local DNA demethylation activates the expression of genes associated with cell proliferation and energy metabolism, thereby regulating the activation of quiescent neurons. Neuronal ferroptosis in ischemic stroke may be linked to the high expression of ferroptosis driver genes such as Ctsb, Mtdh, Ndrg1, Smad7, Cd82, and Acsl1.

Related articles

Related articles are currently not available for this article.