Specificity-driven cell-gene graph learning identifies rare cell states in single-cell and spatial transcriptomic data

Abstract

Detecting rare cell populations that drive development, differentiation, and disease-associated transformation remains a central challenge in biology and medicine. Although these populations often represent promising targets for intervention, they are difficult to resolve from single-cell transcriptomic data because most methods rely on homophily-based cell–cell similarity, which can merge rare cells into dominant populations and mask their subtle transcriptional signatures. The challenge is further amplified in multi-sample analyses, where batch correction can dilute rare-cell-specific signals. Here, we present scFormer, a heterogeneous graph transformer (HGT) framework for sensitive and robust rare-cell discovery. scFormer constructs a Z-score-guided cell-gene heterogeneous graph in which highly specific marker genes serve as informational bridges, embedding rare-cell features directly into the graph topology rather than inferring them from global neighbors. This design provides a clear biological rationale for rare-cell recovery, as low-abundance cells can remain connected through shared high-specificity genes even when local cell–cell neighborhoods are sparse. An integrated optimization strategy jointly performs representation learning, clustering, and optional batch correction, enabling rare-cell discovery while preserving biological structure. Across 125 simulated and 18 real datasets, scFormer consistently achieved competitive or superior performance relative to existing approaches. Applied to diverse multi-sample single-cell and spatial transcriptomics datasets, scFormer recovered known but weakly represented populations and revealed previously obscured cell states, including proliferative club cells in the airway epithelium, revival stem cells during intestinal regeneration, and rare embryonic cell states from spatial transcriptomics. Overall, scFormer provides a unified framework for identifying biologically meaningful rare populations while mitigating batch effects in multi-sample datasets.

Keywords

Rare cell identification / Single-cell RNA sequencing / Heterogeneous graph transformer / Cell-gene graph learning / Spatial transcriptomics

References

[1] Amarachintha SP, Mourya R, Ayabe H, Yang L, Luo Z, Li X, et al. . Biliary organoids uncover delayed epithelial development and barrier function in biliary atresia. Hepatology, 2022, 75(1): 89-103

[2] Argelaguet R, Cuomo ASE, Stegle O, Marioni JC. Computational principles and challenges in single-cell data integration. Nat Biotechnol, 2021, 39(10): 1202-1215

[3] Ascic E, Åkerström F, Sreekumar Nair M, Rosa A, Kurochkin I, Zimmermannova O, et al. . In vivo dendritic cell reprogramming for cancer immunotherapy. Science, 2024, 386(6719 eadn9083

[4] Ayyaz A, Kumar S, Sangiorgi B, Ghoshal B, Gosio J, Ouladan S, et al. . Single-cell transcriptomes of the regenerating intestine reveal a revival stem cell. Nature, 2019, 5697754): 121-125

[5] Barker N, van Es JH, Kuipers J, Kujala P, van den Born M, Cozijnsen M, et al. . Identification of stem cells in small intestine and colon by marker gene Lgr5. Nature, 2007, 449(7165): 1003-1007

[6] Belarif L, Vanhove B, Poirier N. Full antagonist of the IL-7 receptor suppresses chronic inflammation in non-human primate models by controlling antigen-specific memory T cells. Cell Stress, 2018, 2(12): 362-364

[7] Bifari F, Decimo I, Pino A, Llorens-Bobadilla E, Zhao S, Lange C, et al. . Neurogenic radial glia-like cells in meninges migrate and differentiate into functionally integrated neurons in the neonatal cortex. Cell Stem Cell, 2017, 20(3): 360-373.e7

[8] Bodas C, Felipe I, Chanez B, Lafarga M, Lopez de Maturana E, Martinez de Villarreal J, et al. . A common CTRB misfolding variant associated with pancreatic cancer risk causes ER stress and inflammation in mice. Gut, 2025

[9] Chen Y-T, Su Y-C, Kung JT. B cell development sans B cell receptor responsiveness due to unfolded protein response-triggered Mef2c protein degradation. J Immunol Baltim Md 1950, 2018, 201(10): 2885-2898

[10] Choi YH, Kim JK. Dissecting cellular heterogeneity using single-cell RNA sequencing. Mol Cells. 2019;42(3):189–99. https://doi.org/10.14348/molcells.2019.2446.

[11] Chu X, Tian W, Ning J, Xiao G, Zhou Y, Wang Z, et al. . Cancer stem cells: advances in knowledge and implications for cancer therapy. Signal Transduct Target Ther, 2024, 9(1): 170

[12] Davis RE, Ngo VN, Lenz G, Tolar P, Young RM, Romesser PB, et al. . Chronic active B-cell-receptor signalling in diffuse large B-cell lymphoma. Nature, 2010, 463(7277): 88-92

[13] Deng Q, Ramsköld D, Reinius B, Sandberg R. Single-cell RNA-seq reveals dynamic, random monoallelic gene expression in mammalian cells. Science, 2014, 343(6167): 193-196

[14] Dong R, Yuan G-C. GiniClust3: a fast and memory-efficient tool for rare cell type identification. BMC Bioinformatics, 2020, 21(1 158

[15] Dou J, Liang S, Mohanty V, Miao Q, Huang Y, Liang Q, et al. . Bi-order multimodal integration of single-cell data. Genome Biol, 2022, 231 112

[16] Fa B, Wei T, Zhou Y, Johnston L, Yuan X, Ma Y, et al. . GapClust is a light-weight approach distinguishing rare cells from voluminous single cell expression profiles. Nat Commun, 2021, 121 4197

[17] Fedl AS, Tagoh H, Gruenbacher S, Sun Q, Schenk RL, Froussios K, et al. . Transcriptional function of E2A, Ebf1, Pax5, Ikaros and Aiolos analyzed by in vivo acute protein degradation in early B cell development. Nat Immunol, 2024, 259): 1663-1677

[18] Gacita AM, Fullenkamp DE, Ohiri J, Pottinger T, Puckelwartz MJ, Nobrega MA, et al. . Genetic variation in enhancers modifies cardiomyopathy gene expression and progression. Circulation, 2021, 143(13): 1302-1316

[19] Galassi C, Chan TA, Vitale I, Galluzzi L. The hallmarks of cancer immune evasion. Cancer Cell, 2024, 42(11): 1825-1863

[20] Gavin AL, Huang D, Blane TR, Thinnes TC, Murakami Y, Fukui R, et al. . Cleavage of DNA and RNA by PLD3 and PLD4 limits autoinflammatory triggering by multiple sensors. Nat Commun, 2021, 12(1): 5874

[21] Goddard ET, Linde MH, Srivastava S, Klug G, Shabaneh TB, Iannone S, et al. . Immune evasion of dormant disseminated tumor cells is due to their scarcity and can be overcome by T cell immunotherapies. Cancer Cell, 2024, 42(1): 119-134.e12

[22] Haghverdi L, Lun ATL, Morgan MD, Marioni JC. Batch effects in single-cell RNA-sequencing data are corrected by matching mutual nearest neighbors. Nat Biotechnol, 2018, 36(5): 421-427

[23] Hao Y, Hao S, Andersen-Nissen E, Mauck WM, Zheng S, Butler A, et al. . Integrated analysis of multimodal single-cell data. Cell, 2021, 184(13): 3573-3587.e29

[24] Hao Y, Stuart T, Kowalski MH, Choudhary S, Hoffman P, Hartman A, et al. . Dictionary learning for integrative, multimodal and scalable single-cell analysis. Nat Biotechnol, 2024, 42(2): 293-304

[25] Herman J, Sagar, Grün D. FateID infers cell fate bias in multipotent progenitors from single-cell RNA-seq data. Nat Methods. 2018;15(5):379–86. https://doi.org/10.1038/nmeth.4662.

[26] Hewitt RJ, Puttur F, Gaboriau DCA, Fercoq F, Fresquet M, Traves WJ, et al. . Lung extracellular matrix modulates KRT5+ basal cell activity in pulmonary fibrosis. Nat Commun, 2023, 14(1): 6039

[27] Hu Z, Dai M, Chang Y, Hua X, Zhang N, Chen X, et al. . Strategies for arterial graft optimization at the single-cell level. Nat Cardiovasc Res, 2024, 3(5): 541-557

[28] Hung C-T, Su T-H, Chen Y-T, Wu Y-F, Chen Y-T, Lin S-J, et al. . Targeting ER protein TXNDC5 in hepatic stellate cell mitigates liver fibrosis by repressing non-canonical TGFβ signalling. Gut, 2022, 71(9): 1876-1891

[29] Jaitin DA, Kenigsberg E, Keren-Shaul H, Elefant N, Paul F, Zaretsky I, et al. . Massively parallel single-cell RNA-seq for marker-free decomposition of tissues into cell types. Science, 2014, 343(6172): 776-779

[30] Jiang J, Shao X, Liu W, Wang M, Li Q, Wang M, et al. . The mechano-chemical circuit in fibroblasts and dendritic cells drives basal cell proliferation in psoriasis. Cell Rep, 2024, 43(7 114513

[31] Jindal A, Gupta P, Jayadeva, Sengupta D. Discovery of rare cells from voluminous single cell expression data. Nat Commun. 2018;9(1):4719. https://doi.org/10.1038/s41467-018-07234-6.

[32] Jung S, Choe S, Woo H, Jeong H, An H-K, Moon H, et al. . Autophagic death of neural stem cells mediates chronic stress-induced decline of adult hippocampal neurogenesis and cognitive deficits. Autophagy, 2020, 163): 512-530

[33] Korsunsky I, Millard N, Fan J, Slowikowski K, Zhang F, Wei K, et al. . Fast, sensitive and accurate integration of single-cell data with Harmony. Nat Methods, 2019, 16(12): 1289-1296

[34] Levine JH, Simonds EF, Bendall SC, Davis KL, Amir ED, Tadmor MD, et al. . Data-driven phenotypic dissection of AML reveals progenitor-like cells that correlate with prognosis. Cell, 2015, 162(1): 184-197

[35] Li P, Liang D, Yang E, Zeb M, Huang H, Sun H, et al. . Bio-nanopore technology for biomolecules detection. Adv Biotechnol, 2024, 2(4): 45

[36] Liang S, Mohanty V, Dou J, Miao Q, Huang Y, Müftüoğlu M, et al. . Single-cell manifold-preserving feature selection for detecting rare cell populations. Nat Comput Sci, 2021, 1(5): 374-384

[37] Liao S, Zhou X, Liu C, Liu C, Hao S, Luo H, et al. . Stereo-cell: spatial enhanced-resolution single-cell sequencing with high-density DNA nanoball-patterned arrays. Science, 2025, 389(6762 eadr0475

[38] Liu Y, Zhuo S, Zhou Y, Ma L, Sun Z, Wu X, et al. . Yap-Sox9 signaling determines hepatocyte plasticity and lineage-specific hepatocarcinogenesis. J Hepatol, 2022, 763): 652-664

[39] Liu J, Fan X, Gu C, Yang Y, Wu B, Chen G, et al. . scHeteroNet: a heterophily-aware graph neural network for accurate cell type annotation and novel cell detection. Adv Sci Weinh Baden-Wurtt Ger, 2025, 12(16 e2412095

[40] Lu Z, Mo S, Xie D, Zhai X, Deng S, Zhou K, et al. . Polyclonal-to-monoclonal transition in colorectal precancerous evolution. Nature, 2024, 6368041): 233-240

[41] Ma A, Wang X, Li J, Wang C, Xiao T, Liu Y, et al. . Single-cell biological network inference using a heterogeneous graph transformer. Nat Commun, 2023, 14(1): 964

[42] Mariani J, Coppola G, Zhang P, Abyzov A, Provini L, Tomasini L, et al. . FOXG1-dependent dysregulation of GABA/glutamate neuron differentiation in autism spectrum disorders. Cell, 2015, 162(2): 375-390

[43] Martinu T, Todd JL, Gelman AE, Guerra S, Palmer SM. Club cell secretory protein in lung disease: emerging concepts and potential therapeutics. Annu Rev Med, 2023, 74: 427-441

[44] Metcalfe C, Kljavin NM, Ybarra R, de Sauvage FJ. Lgr5+ stem cells are indispensable for radiation-induced intestinal regeneration. Cell Stem Cell, 2014, 14(2): 149-159

[45] Montoro DT, Haber AL, Biton M, Vinarsky V, Lin B, Birket SE, et al. . A revised airway epithelial hierarchy includes CFTR-expressing ionocytes. Nature, 2018, 560(7718): 319-324

[46] Polański K, Young MD, Miao Z, Meyer KB, Teichmann SA, Park J-E. BBKNN: fast batch alignment of single cell transcriptomes. Bioinforma Oxf Engl, 2020, 36(3): 964-965

[47] Potter SS. Single-cell RNA sequencing for the study of development, physiology and disease. Nat Rev Nephrol, 2018, 14(8): 479-492

[48] Rimland CA, Tilson SG, Morell CM, Tomaz RA, Lu W-Y, Adams SE, et al. . Regional differences in human biliary tissues and corresponding in vitro-derived organoids. Hepatology, 2021, 73(1): 247-267

[49] Rock JR, Onaitis MW, Rawlins EL, Lu Y, Clark CP, Xue Y, et al. . Basal cells as stem cells of the mouse trachea and human airway epithelium. Proc Natl Acad Sci U S A, 2009, 106(31): 12771-12775

[50] Ross AA, Cooper BW, Lazarus HM, Mackay W, Moss TJ, Ciobanu N, et al. . Detection and viability of tumor cells in peripheral blood stem cell collections from breast cancer patients using immunocytochemical and clonogenic assay techniques. Blood, 1993, 82(9): 2605-2610

[51] Ruan X, Kang B, Qi C, Lin W, Wang J, Zhang X. Progenitor cell diversity in the developing mouse neocortex. Proc Natl Acad Sci U S A, 2021, 118(10 e2018866118

[52] Saillour Y, Broix L, Bruel-Jungerman E, Lebrun N, Muraca G, Rucci J, et al. . Beta tubulin isoforms are not interchangeable for rescuing impaired radial migration due to Tubb3 knockdown. Hum Mol Genet, 2014, 236): 1516-1526

[53] Sang Q, Kong F. Applications for single-cell and spatial transcriptomics in plant research. New Crops, 2024, 1 100025

[54] Sekine Y, Algarate PT, Cafferty WBJ, Strittmatter SM. Plexina2 and CRMP2 signaling complex is activated by Nogo-A-liganded Ngr1 to restrict corticospinal axon sprouting after trauma. J Neurosci off J Soc Neurosci, 2019, 39(17): 3204-3216

[55] Shaffer SM, Dunagin MC, Torborg SR, Torre EA, Emert B, Krepler C, et al. . Rare cell variability and drug-induced reprogramming as a mode of cancer drug resistance. Nature, 2017, 546(7658): 431-435

[56] Si J, Su X, Jin Z, Duan S. Uncovering essential lncRNAs through transcriptome-scale CRISPR-Cas13 screening. Adv Biotechnol, 2025, 3(3): 27

[57] Srivatsan SR, Regier MC, Barkan E, Franks JM, Packer JS, Grosjean P, et al. . Embryo-scale, single-cell spatial transcriptomics. Science, 2021, 373(6550): 111-117

[58] Suzuki Y, Lutshumba J, Chen KC, Abdelaziz MH, Sa Q, Ochiai E. IFN-γ production by brain-resident cells activates cerebral mRNA expression of a wide spectrum of molecules critical for both innate and T cell-mediated protective immunity to control reactivation of chronic infection with Toxoplasma gondii. Front Cell Infect Microbiol, 2023, 13: 1110508

[59] Tamura K, Yu J, Hata T, Suenaga M, Shindo K, Abe T, et al. . Mutations in the pancreatic secretory enzymes CPA1 and CPB1 are associated with pancreatic cancer. Proc Natl Acad Sci U S A, 2018, 115(18): 4767-4772

[60] Tang S, Huang X, Wang X, Zhou X, Huang H, Qin L, et al. Vital and distinct roles of H2A.Z isoforms in hepatocellular carcinoma. OncoTargets Ther. 2020;13:4319–37. https://doi.org/10.2147/OTT.S243823.

[61] Tao S, Tang D, Morita Y, Sperka T, Omrani O, Lechel A, et al. . Wnt activity and basal niche position sensitize intestinal stem and progenitor cells to DNA damage. EMBO J, 2015, 34(5): 624-640

[62] Valente M, Collinet N, Vu Manh T-P, Popoff D, Rahmani K, Naciri K, et al. . Novel mouse models based on intersectional genetics to identify and characterize plasmacytoid dendritic cells. Nat Immunol, 2023, 24(4): 714-728

[63] Wang X, Duan M, Li J, Ma A, Xin G, Xu D, et al. . MarsGT: Multi-omics analysis for rare population inference using single-cell graph transformer. Nat Commun, 2024, 15(1): 338

[64] Wegmann R, Neri M, Schuierer S, Bilican B, Hartkopf H, Nigsch F, et al. . CellSIUS provides sensitive and specific detection of rare cell populations from complex single-cell RNA-seq data. Genome Biol, 2019, 20(1 142

[65] Wijnands C, Noori S, Donk NWCJ, VanDuijn MM, Jacobs JFM. Advances in minimal residual disease monitoring in multiple myeloma. Crit Rev Clin Lab Sci, 2023, 607): 518-534

[66] Xu Y, Wang S, Feng Q, Xia J, Li Y, Li H-D, et al. . scCAD: Cluster decomposition-based anomaly detection for rare cell identification in single-cell expression data. Nat Commun, 2024, 15(1): 7561

[67] Yong CH, Hoon S, De Mel S, Xu S, Scolnick JA, Huo X, et al. . Mapbatch: conservative batch normalization for single-cell RNA-sequencing data enables discovery of rare cell populations in a multiple myeloma cohort. Blood, 2021, 138(Supplement 1): 2954-2954

[68] Yu Z, Su Y, Lu Y, Yang Y, Wang F, Zhang S, et al. . Topological identification and interpretation for single-cell gene regulation elucidation across multiple platforms using scMGCA. Nat Commun, 2023, 14(1): 400

[69] Zappia L, Phipson B, Oshlack A. Splatter: simulation of single-cell RNA sequencing data. Genome Biol, 2017, 18(1): 174

[70] Zhang G, He P, Tan H, Budhu A, Gaedcke J, Ghadimi BM, et al. . Integration of metabolomics and transcriptomics revealed a fatty acid network exerting growth inhibitory effects in human pancreatic cancer. Clin Cancer Res off J Am Assoc Cancer Res, 2013, 19(18): 4983-4993

[71] Zhang Y, Parmigiani G, Johnson WE. ComBat-seq: batch effect adjustment for RNA-seq count data. NAR Genom Bioinform, 2020, 23 lqaa078

[72] Zhang Y, Narayanan SP, Mannan R, Raskind G, Wang X, Vats P, et al. . Single-cell analyses of renal cell cancers reveal insights into tumor microenvironment, cell of origin, and therapy response. Proc Natl Acad Sci U S A, 2021, 11824 e2103240118

[73] Zhou S, Li Y, Wu W, Li L. ScMMT: a multi-use deep learning approach for cell annotation, protein prediction and embedding in single-cell RNA-seq data. Brief Bioinform, 2024, 252 bbad523

Funding

National Key R&D Program(2022YFC3400401)

National Natural Science Foundation of China(32270604)

Guangdong Basic and Applied Basic Research Foundation(2026A1515010528)

Shenzhen Natural Science Foundation in Basic Research Fund(JCYJ20250604175316022)

RIGHTS & PERMISSIONS

The Author(s)