ACSCeND offers a robust, interpretable approach to profiling CSC dynamics and establishes CSC state as a clinically meaningful, pan-cancer biomarker for guiding stemness-informed therapies by integrating single-cell precision with bulk-level applicability.
Abstract
Abstract Cancer stem-like cells (CSCs) play a pivotal role in driving tumor heterogeneity, therapeutic resistance, and disease progression. Despite the power of single-cell RNA sequencing (scRNA-seq) to resolve intratumoral hierarchies, there remains a need for robust, scalable tools to consistently profile CSCs across both single-cell and bulk transcriptomic data. To address this, we developed ACSCeND—a unified, machine learning–based framework that enables high-resolution CSC state classification and tissue-level deconvolution. ACSCeND comprises (i) a supervised classifier trained on curated scRNA-seq datasets to assign cells into pluripotent-like, multipotent-like, or unipotent-like states, and (ii) an attention-guided autoencoder that deconvolves CSC subtype proportions from bulk RNA sequencing data. Compared to existing tissue deconvolution tools, ACSCeND achieves superior performance, with higher accuracy across synthetic and real-world samples. Applied to over 25 000 tumor profiles from The Cancer Genome Atlas (TCGA), PREdiction of Clinical Outcomes from Genomics (PRECOG), tumor-relapse, and checkpoint inhibitor studies, ACSCeND reveals that CSC abundance strongly correlates with poor disease-free survival and reduced immunotherapy efficacy. Moreover, it uncovers distinct CSC-state-specific molecular programs, offering insights into CSC-driven heterogeneity and tumor evolution. The model also recapitulates known developmental hierarchies in noncancerous tissues, supporting its broader biological relevance. By integrating single-cell precision with bulk-level applicability, ACSCeND offers a robust, interpretable approach to profiling CSC dynamics and establishes CSC state as a clinically meaningful, pan-cancer biomarker for guiding stemness-informed therapies. ACSCeND is available as a python package (through pip) at https://pypi.org/project/ACSCeND/.
Single-cell sequencing (SCS) technologies have revolutionized cancer research by overcoming the critical limitations of bulk sequencing, enabling unprecedented resolution in deciphering intratumoral heterogeneity and the tumor microenvironment (TME). Unlike bulk approaches that average molecular signals across cell populations, SCS resolves rare subpopulations-such as pre-existing therapy-resistant clones and cancer stem cells-and reconstructs clonal evolutionary trajectories driving metastasis and treatment resistance. By profiling individual cells, SCS reveals dynamic cellular interactions, functional states, and spatial architectures within the TME, including immune-stromal crosstalk and transcriptional plasticity. Applied pan-cancer, SCS uncovers conserved molecular programs that transcend tissue origin, such as universal T-cell exhaustion signatures, recurrent stromal reprogramming, and fundamental pathways like epithelial-mesenchymal transition and angiogenesis. Technological advancements in scalability, cost-effectiveness, and spatial multi-omics integration now enable high-resolution mapping of tumor ecosystems. Large-scale initiatives like the Human Cell Atlas and CancerSCEM harmonize datasets, revealing both universal mechanisms and tissue-specific adaptations. Clinically, SCS enhances diagnostic sensitivity and enables precision prognostication through cellular biomarkers. It informs novel therapeutic strategies, such as targeting specific TME components or optimizing immunotherapy response. Emerging artificial intelligence (AI) tools further amplify SCS's impact by integrating multi-omic data to predict drug responses and resistance mechanisms. Despite challenges in standardization and cost, the future convergence of SCS, spatial technologies, and AI-driven modeling positions SCS as a cornerstone of precision oncology. This review systematically summarizes the advances of SCS technologies and their transformative applications in pan-cancer research and clinical translation.
Zerui Lu, Jiayi Li, Rui Geng et al.· Seminars in Oncology· 0 citations
Glioblastomas are aggressive, heterogeneous tumors that present significant challenges in both diagnosis and treatment. Despite advances in surgical resection, radiotherapy, and chemotherapy with temozolomide (TMZ), the prognosis for glioblastoma patients remains poor, largely due to tumor heterogeneity and resistance mechanisms, such as genetic mutations in DNA repair pathways. To address these specific heterogenous qualities of glioblastoma, single-cell RNA sequencing (scRNA-seq) has emerged as a powerful tool for characterizing glioblastoma tumors, enabling the identification of subpopulations that respond differently to treatment. However, utilizing the vast amount of data generated by scRNA-seq poses challenges in clinical applications. To overcome this challenge, computational models have been introduced to more effectively process patient scRNA-seq data into more digestible information for clinicians. More specifically, advanced deep learning approaches show promise for processing and analyzing patient scRNA-seq data, enhancing informed treatment approaches for highly heterogenous glioblastoma. This review aims to explain how scRNA-seq can be used to identify important areas of glioblastoma treatment resistance, evaluate current glioblastoma scRNA-seq-based deep learning models, and outline relevant training datasets to overcome patient scRNA-seq data availability limitations. Ultimately, these deep learning models can be utilized by researchers and clinicians to provide more informed and precise treatment to glioblastoma patients.
Marybeth G. Yonk, Mainak Mustafi, Megan A Lim et al.· Frontiers in Genetics· 0 citations
Intratumor transcriptional heterogeneity (ITTH), defined as the coexistence of diverse cell states within a single tumor, complicates cancer treatment and contributes to variable therapeutic responses. Although single-cell RNA sequencing (scRNA-seq) can resolve this complexity, its cost and technical demands limit large-scale use. Bulk RNA sequencing (bulk RNA-seq) provides a scalable alternative but requires computational methods to deconvolve bulk transcriptomes into distinct cell states. Existing supervised approaches rely on accurate reference data, which are lacking for many cancer types, while unsupervised methods are not tailored to capture heterogeneity within the malignant compartment. To address these limitations, we developed CDState, an unsupervised deconvolution method based on nonnegative matrix factorization with a sum-to-one constraint and a cosine-similarity-based optimization, which infers malignant cell states using bulk RNA-seq data. CDState demonstrated robustness using pseudobulk scRNA-seq datasets from five cancer types, outperforming existing unsupervised methods in estimating both state-specific gene expression and cell proportions. Applied to 33 cancer types from The Cancer Genome Atlas, CDState revealed recurrent gene programs, including epithelial-mesenchymal transition, MYC targets, and oxidative phosphorylation, as major contributors to malignant ITTH. The malignant state proportions were linked to clinical features, including patient survival and therapeutic response. Finally, mutations and copy number alterations in genes such as TP53, KRAS, PIK3CA, SOX2, and SATB1 were identified as potential genetic drivers of malignant cell ITTH across cancer types. This study demonstrates the utility of CDState for characterization of malignant cell states from bulk RNA-seq data, establishing a framework for investigating malignant cell ITTH in large-scale cancer atlases.
Agnieszka Kraft, J. Yates, Florian Barkmann et al.· Cancer Research· 0 citations
Hepatocellular carcinoma (HCC) remains a leading cause of cancer-related mortality worldwide, underscoring the urgent need for robust molecular signatures that support multiple clinical tasks and generalize across diverse transcriptomic platforms.
We performed an integrative analysis of 1,300 HCC transcriptomic profiles spanning microarray (n = 869), bulk RNA sequencing (n = 370), spatial transcriptomics (n = 10), and single-cell RNA sequencing (n = 51). Using a multi-task machine learning framework, we distilled a foundational 284-gene program into a compact 25-gene transcriptional signature, including the under-characterized gene IPO9. This signature captured the core proliferative state of HCC while maintaining high diagnostic accuracy (AUC = 0.987–0.991) and served as an independent prognostic factor for both overall and progression-free survival, with activity levels that progressively scale with advancing tumor grade. Spatial and single-cell analyses localized the signature almost exclusively to malignant hepatocytes within the tumor core, confirming its cell-intrinsic nature. Extending beyond HCC, the signature demonstrated conserved oncogenic pathway alignment across 31 additional cancer types and prognostic relevance in nine of them, identifying it as a pan-cancer malignant proliferative-proteostatic marker.
This study defines a mechanistically interpretable 25-gene malignant axis that generalizes from HCC to multiple cancer types . Our findings suggest that this signature could serve as a reliable framework for clinical risk assessment and personalized management in HCC.
Tho Ngoc-Quynh Le, Nhi Doan Yen Nguyen, Thanh Quang Nguyen et al.· Discover Oncology· 0 citations
GITIII-scale is presented, a hierarchical, interpretable pan-cancer spatial transcriptomics foundation model for TME representation learning that investigates cell state-niche associations and their underlying ligand-receptor (LR) signaling pathways that recovered niche-associated state changes more accurately than existing spatial transcriptomics foundation models in cancer types unseen during training.
Xiaohui Xiao, Jia-Shu He, Shiyang Zhang et al.· 0 citations
PURPOSE
Circulating tumor cells (CTCs) provide a minimally invasive window into metastatic disease and treatment response, but their clinical utility has been constrained by manual, subjective, low-throughput identification in multi-channel fluorescence microscopy data.
METHODS
To address this limitation, we developed and clinically validated the System for Enhanced Evaluation of Tumor Cells (SEE-TC), a deep learning approach for scalable, reproducible phenotyping of individual circulating cells. SEE-TC was developed and evaluated on more than 8.5 million cells from 3,386 blood samples spanning six cancer types, enabling generalization across heterogeneous imaging conditions, staining panels, and acquisition platforms.
RESULTS
SEE-TC achieved single-cell segmentation accuracy on par with humans, learned biologically relevant latent cellular representations that correlate with established morphological and immunofluorescent biomarkers, and reliably distinguished CTCs from background populations without reliance on arbitrary thresholds. When applied longitudinally, SEE-TC provides a quantitative, patient-level readout of CTC burden over time which was significantly associated with worse overall survival across multiple cancer types.
CONCLUSIONS
To our knowledge, this is the first fully automated AI approach to single-cell segmentation and CTC phenotyping. By transforming CTC analysis from a human-dependent task into a scalable and reproducible digital assay, SEE-TC enables high-fidelity longitudinal monitoring of tumor burden and supports broader clinical deployment of CTC-based liquid biopsies in precision oncology. It is currently being deployed to identify CTCs and quantify target expression for both prognostic and predictive biomarker evaluation in multiple prospective clinical trials on a commercial platform.
M. Bootsma, M. Sharifi, J. Sperger et al.· Clinical Cancer Research· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
