Single-Cell Transcriptomics Reveals Cancer Architecture
Single-Cell Transcriptomics Reveals the Intricate Architecture of Cancer: A Comprehensive Review of Computational Methods and Biological Insights
Abstract
Single-cell RNA sequencing (scRNA-seq) has revolutionized our understanding of cancer biology by enabling the dissection of cellular heterogeneity at unprecedented resolution. This comprehensive review synthesizes recent advances in single-cell transcriptomics applied to cancer research, focusing on both methodological innovations and biological discoveries. We examine the evolution of scRNA-seq technologies, computational frameworks for data analysis, and their collective impact on elucidating tumor complexity, tumor microenvironment (TME) composition, metastatic progression, and therapeutic resistance. Particular attention is devoted to the integration of single-cell multi-omics approaches, spatial transcriptomics, and machine learning applications in cancer research. We discuss how these technologies have uncovered novel cell states, characterized cellular plasticity in cancer, and identified previously unrecognized mechanisms of immune evasion. Furthermore, we address current technical limitations, computational challenges, and future directions, including the potential of single-cell technologies to guide precision oncology and improve clinical outcomes. The review concludes with a perspective on how emerging single-cell modalities and artificial intelligence integration will shape the next generation of cancer research and therapeutic development.
Keywords: single-cell RNA sequencing, cancer genomics, tumor microenvironment, cellular heterogeneity, computational biology, spatial transcriptomics, cancer immunology, precision oncology
1. Introduction
Cancer remains one of the leading causes of mortality worldwide, with an estimated 19.3 million new cases and 10 million deaths annually. The fundamental hallmarks of cancer, as originally described by Hanahan and Weinberg and subsequently expanded, encompass diverse biological capabilities acquired during tumor development. However, what has become increasingly clear over the past decade is that tumors are not homogeneous masses of identical cells, but rather complex ecosystems comprised of multiple distinct cell populations engaging in dynamic interactions. This intra-tumoral heterogeneity presents fundamental challenges to cancer treatment, driving therapeutic resistance, metastatic progression, and disease recurrence.
The advent of single-cell RNA sequencing (scRNA-seq) has provided researchers with a powerful lens through which to examine this heterogeneity. Unlike bulk RNA sequencing, which provides averaged gene expression measurements across thousands to millions of cells, scRNA-seq enables the profiling of transcriptomes from individual cells, revealing the complete spectrum of cellular states within a tumor specimen. Since its initial demonstration in 2009, scRNA-seq technologies have undergone remarkable evolution, progressing from low-throughput methods capable of profiling dozens of cells to platforms that can simultaneously characterize hundreds of thousands of cells from a single experiment.
This revolution in single-cell genomics has transformed our understanding of cancer biology in multiple dimensions. First, it has enabled the identification of rare cell populations that may play disproportionate roles in tumor progression, such as cancer stem cells, drug-tolerant persister cells, and immunosuppressive immune cell subsets. Second, it has provided unprecedented insights into the tumor microenvironment, revealing the cellular composition, functional states, and spatial organization of stromal and immune cells that shape tumor behavior. Third, it has illuminated the processes of metastasis and therapeutic resistance at single-cell resolution, identifying molecular programs that enable cancer cells to adapt to selective pressures. Finally, it has facilitated the discovery of novel therapeutic targets and predictive biomarkers that may ultimately improve patient outcomes through precision oncology approaches.
This comprehensive review aims to synthesize the rapidly evolving landscape of single-cell transcriptomics in cancer research. We will examine the technological foundations that have enabled this revolution, the computational methods developed to extract biological insights from high-dimensional single-cell data, and the major biological discoveries that have emerged from applying these approaches to cancer. We will also discuss emerging modalities, including multi-omics integration and spatial transcriptomics, and conclude with a perspective on future directions and challenges in the field.
2. Evolution of Single-Cell RNA Sequencing Technologies
2.1 Historical Development and Technical Foundations
The journey toward comprehensive single-cell transcriptomic analysis began with the foundational work of Tang et al. in 2009, who demonstrated the feasibility of sequencing RNA from individual cells using a sensitive amplification approach. This initial breakthrough, while revolutionary, was limited by low throughput and high technical noise, restricting applications to small numbers of cells. However, it established the fundamental principle that single-cell transcriptomics could reveal cellular heterogeneity that remained invisible to bulk analyses.
The subsequent decade witnessed an explosion of technological innovation in scRNA-seq platforms. The development of droplet-based microfluidics, exemplified by the Drop-seq and inDrop systems, represented a major advance by enabling the simultaneous encapsulation of thousands of cells with uniquely barcoded beads for massively parallel sequencing. These approaches reduced costs by more than an order of magnitude while dramatically increasing throughput. The commercial introduction of the 10x Genomics Chromium system further democratized single-cell sequencing, providing a turnkey platform that combined robust microfluidics with optimized chemistry and analysis pipelines.
Plate-based methods have continued to evolve in parallel, offering advantages in terms of sensitivity and full-length transcript coverage. The Smart-seq2 protocol, developed by Picelli et al., improved upon previous methods by incorporating template-switching and locked nucleic acid primers to achieve near-complete coverage of mRNA molecules from individual cells. This approach has proven particularly valuable for applications requiring isoform-level resolution, such as alternative splicing analysis in cancer cells. More recently, combinatorial indexing strategies, including SPLiT-seq and sci-RNA-seq, have eliminated the need for physical cell isolation entirely, instead using nuclear barcoding approaches to scale single-cell analysis to millions of cells.
2.2 Current State-of-the-Art Platforms
Contemporary scRNA-seq platforms offer researchers a spectrum of options balancing throughput, sensitivity, and cost. The 10x Genomics Chromium system remains the most widely adopted platform for cancer research, offering the ability to profile 500-10,000 cells per sample with robust chemistry and extensive community support. Recent iterations have expanded capabilities to include multi-modal profiling, such as simultaneous measurement of RNA and surface protein expression (CITE-seq) or chromatin accessibility (multiome).
For applications requiring ultra-high throughput, platforms such as the Parse Biosciences Evercode WT and the Mission Bio Tapestri enable profiling of hundreds of thousands to millions of cells, making them particularly valuable for detecting rare cell populations in large tumors. Conversely, for studies requiring maximum sensitivity and full-length transcript information, the Takara Bio ICELL8 and Smart-seq3 platforms provide superior gene detection efficiency and isoform resolution, albeit at higher cost and lower throughput.
Emerging technologies continue to push the boundaries of single-cell analysis. Single-nucleus RNA sequencing (snRNA-seq) has extended single-cell applications to frozen tissue specimens and archival samples, facilitating retrospective studies of clinical cohorts. Long-read sequencing technologies, particularly when combined with single-cell approaches, promise to reveal the full complexity of transcript isoforms and fusion genes in cancer cells. Spatial transcriptomics platforms, including the 10x Genomics Visium, NanoString GeoMx, and Vizgen MERSCOPE, have begun to integrate positional information with transcriptomic data, enabling the reconstruction of cellular neighborhoods within tumors.
2.3 Technical Considerations for Cancer Applications
Applying scRNA-seq to cancer specimens presents unique technical challenges that must be carefully considered in experimental design. Tumor dissociation protocols must balance complete tissue disruption with preservation of delicate cell types, particularly immune cells that may be selectively lost during harsh enzymatic treatments. The time from tissue acquisition to processing is critical, as transcriptional changes can occur rapidly ex vivo, potentially confounding biological interpretations.
Cancer specimens also present analytical challenges due to their inherent complexity. The presence of malignant cells alongside diverse stromal and immune populations requires computational methods to distinguish cell types and states. The genomic instability of cancer cells, including aneuploidy and copy number alterations, can interfere with standard quality control metrics that assume diploid genomes. Furthermore, the high transcriptional activity and RNA content of cancer cells can create technical artifacts in droplet-based systems, potentially leading to biased representation.
Despite these challenges, careful experimental design and appropriate controls have enabled successful application of scRNA-seq to virtually all cancer types, from hematological malignancies circulating in blood to solid tumors requiring complex dissociation protocols. The continued refinement of tissue handling methods, fixation strategies, and computational approaches promises to further improve the applicability and reproducibility of single-cell studies in cancer research.
3. Computational Methods for Single-Cell Cancer Genomics
3.1 Preprocessing and Quality Control
The analysis of scRNA-seq data begins with extensive computational preprocessing to transform raw sequencing reads into a gene expression matrix suitable for downstream analysis. Initial steps include read alignment to a reference genome, quantification of gene expression, and quality control to remove low-quality cells and genes. Cancer-specific considerations in this phase include the potential presence of doublets (two cells encapsulated in a single droplet), which can be particularly frequent when processing large or clumped cancer cells, and the need to distinguish true biological heterogeneity from technical artifacts.
Quality control metrics typically include the number of detected genes per cell, total transcript counts, and the percentage of mitochondrial reads, which can indicate cellular stress or damage. However, these metrics must be interpreted carefully in cancer contexts, as malignant cells may naturally exhibit altered metabolism or transcriptional activity that deviates from normal cell populations. Automated doublet detection tools, such as DoubletFinder and Scrublet, have been adapted for cancer data by incorporating training sets that reflect the heterogeneity expected in tumor specimens.
Normalization methods correct for differences in sequencing depth and capture efficiency between cells. Approaches such as SCTransform have gained popularity for their ability to stabilize variances while preserving biological heterogeneity. For cancer data, specialized normalization methods have been developed to account for gene amplification and deletion events that can affect expression measurements independent of transcriptional regulation.
3.2 Dimensionality Reduction and Clustering
The high dimensionality of scRNA-seq data, typically encompassing expression measurements for 15,000-30,000 genes per cell, necessitates dimensionality reduction approaches to identify biologically meaningful patterns. Principal Component Analysis (PCA) remains a foundational method, typically applied to highly variable genes selected based on their dispersion across cells. More recent approaches, including variational autoencoders and deep learning-based methods, have been developed to capture non-linear relationships in high-dimensional data.
Following dimensionality reduction, clustering algorithms group cells with similar transcriptomic profiles, revealing distinct cell populations within tumors. Graph-based clustering methods, which construct networks of cells based on similarity in reduced dimensional space followed by community detection, have become the standard approach due to their computational efficiency and ability to identify clusters of varying densities. For cancer data, these methods have been enhanced to account for gradual transitions between cell states, a phenomenon particularly relevant for understanding cancer cell plasticity and lineage relationships.
Visualization methods, particularly t-Distributed Stochastic Neighbor Embedding (t-SNE) and Uniform Manifold Approximation and Projection (UMAP), enable the exploration of high-dimensional single-cell data in two dimensions. These approaches have proven invaluable for identifying the relationships between different cell populations and detecting rare cell types that may play important roles in tumor biology.
3.3 Cell Type Annotation and Malignant Cell Identification
Accurate cell type annotation is critical for interpreting single-cell data from tumors, which contain mixtures of malignant cells, normal epithelial cells, immune cells, fibroblasts, endothelial cells, and other stromal populations. Supervised annotation methods, such as SingleR and scmap, compare the expression profiles of single cells to reference datasets of annotated cell types. However, these approaches face challenges in cancer due to the altered transcriptional programs of malignant cells and the potential presence of cell states not represented in existing reference datasets.
Unsupervised annotation approaches rely on the identification of canonical marker genes through differential expression analysis between clusters. For cancer data, this requires careful consideration of the appropriate reference datasets and marker genes, as cancer cells may express markers of multiple lineages or exhibit hybrid phenotypes. Recent approaches have combined supervised and unsupervised methods, using automated annotation as a starting point followed by manual curation based on domain knowledge.
A fundamental challenge in analyzing tumor scRNA-seq data is distinguishing malignant from non-malignant cells. This is particularly critical when the cell of origin is unclear or when admixed normal epithelial cells are present. Inferential methods, including copy number variation inference from scRNA-seq data (inferCNV) and mutation detection from transcriptomic data, have been developed to identify malignant cells based on genomic alterations. Machine learning approaches trained on features that distinguish malignant from normal cells have also proven effective, particularly when coupled with spatial information from imaging or spatial transcriptomics.
3.4 Trajectory Inference and Pseudotime Analysis
Trajectory inference methods aim to reconstruct dynamic processes, such as differentiation, treatment response, or metastatic progression, from static single-cell snapshots. These approaches, including Monocle, Slingshot, and PAGA, order cells along pseudotime trajectories based on their transcriptional similarity, revealing continuous transitions between cell states.
In cancer research, trajectory analysis has been applied to study multiple dynamic processes. The epithelial-to-mesenchymal transition (EMT), a critical process in metastasis, has been reconstructed at single-cell resolution, revealing hybrid E/M states and the transcriptional programs that drive this transition. Similarly, T cell exhaustion in the tumor microenvironment has been characterized through pseudotime analysis, identifying the transcription factors and surface markers associated with progressive dysfunction.
More recent approaches have combined trajectory inference with RNA velocity, which uses the ratio of unspliced to spliced mRNA transcripts to predict the future state of cells. RNA velocity has been particularly valuable for studying cancer cell dynamics, revealing the directionality of state transitions and identifying points of bifurcation where cells commit to distinct fates. The integration of spatial information with trajectory analysis promises to further elucidate how cellular neighborhoods influence cell fate decisions within tumors.
3.5 Integration of Multi-Omics Single-Cell Data
The rapid expansion of single-cell technologies beyond transcriptomics has created both opportunities and challenges for integrative analysis. Multi-omics approaches, which simultaneously measure multiple modalities from the same cell, provide more comprehensive views of cellular states but require sophisticated computational methods for integration.
Methods for integrating scRNA-seq with single-cell ATAC-seq (assay for transposase-accessible chromatin) have revealed how chromatin accessibility shapes transcriptional programs in cancer cells. Tools such as Seurat's WNN (weighted nearest neighbor) multi-modal integration and MOFA+ (multi-omics factor analysis) enable the identification of cell states based on combined epigenetic and transcriptional information. These approaches have been particularly valuable for understanding how oncogenic signaling pathways rewire gene regulatory networks in cancer.
Protein-sequencing integration methods, combining surface protein measurement via CITE-seq with transcriptomic data, have improved immune cell phenotyping in tumors. Antibody-oligo conjugates targeting hundreds of surface proteins can be quantified alongside gene expression, providing high-dimensional immunophenotyping that reveals functional states not apparent from transcriptomics alone.
Spatial integration methods have emerged as particularly powerful for cancer research, enabling the reconstruction of cellular neighborhoods and the identification of spatially restricted cell-cell interactions. Tools such as CellPhoneDB, NicheNet, and Giotto infer cell-cell communication from ligand-receptor expression patterns, revealing how different cell populations within tumors coordinate their behavior. The integration of spatial transcriptomics data with scRNA-seq reference datasets has further enhanced these analyses, enabling the deconvolution of spatial spots into constituent cell types and the mapping of scRNA-seq-derived cell states onto tissue architecture.
4. Biological Insights from Single-Cell Cancer Research
4.1 Intratumoral Heterogeneity and Cancer Evolution
One of the most fundamental insights from single-cell cancer genomics has been the documentation and quantification of intratumoral heterogeneity at unprecedented resolution. Early scRNA-seq studies of melanoma, glioblastoma, and breast cancer revealed that tumors are comprised of multiple distinct malignant cell subpopulations, each with unique transcriptional programs and potentially different sensitivities to therapy. This heterogeneity exists at multiple levels: between distinct tumor regions (spatial heterogeneity), between tumor cells in the same region (local heterogeneity), and within individual cancer cells (temporal heterogeneity as cells transition between states).
Single-cell analyses have revealed that this heterogeneity is often organized along axes of differentiation, with cancer cells existing in a spectrum from stem-like to more differentiated states. In glioblastoma, scRNA-seq identified neural progenitor-like, oligodendrocyte progenitor-like, astrocyte-like, and mesenchymal-like malignant cells within individual tumors, reminiscent of the lineage hierarchies observed in normal neural development. Similarly, in colorectal cancer, single-cell studies have identified stem-like, transient amplifying, and differentiated tumor cells, with the relative abundance of different states correlating with clinical outcomes.
The transcriptional programs underlying these different cell states have been extensively characterized. Stem-like cancer cells consistently express genes associated with embryonic development, including SOX2, NANOG, and OCT4, along with drug resistance programs. More differentiated states exhibit tissue-specific differentiation programs, such as melanocytic markers in melanoma or epithelial markers in carcinomas. Importantly, single-cell studies have revealed that cancer cells exhibit remarkable plasticity, capable of transitioning between these states in response to environmental cues or therapeutic pressure.
This plasticity has profound implications for cancer treatment. Drug-tolerant persister cells, a rare population that survives initial therapy and serves as a reservoir for disease recurrence, have been characterized at single-cell resolution in multiple cancer types. These cells typically exhibit a slow-cycling state, upregulation of drug efflux pumps, and activation of survival pathways. The ability to identify and characterize these rare populations has provided new targets for therapies aimed at preventing relapse.
4.2 The Tumor Microenvironment: Cellular Composition and Functional States
Perhaps the most extensive application of scRNA-seq in cancer research has been the characterization of the tumor microenvironment (TME), the complex ecosystem of non-malignant cells that surrounds and interacts with cancer cells. Single-cell studies have comprehensively catalogued the cellular constituents of the TME across cancer types, revealing remarkable diversity in immune cell composition, fibroblast heterogeneity, and vascular organization.
Immune cell profiling has been a major focus, driven by the clinical success of cancer immunotherapies and the need to understand determinants of response and resistance. scRNA-seq has revealed that T cells in tumors exist in a continuum of states, from naïve and activated cells to various stages of exhaustion. Exhausted T cells, characterized by progressive loss of effector function and expression of inhibitory receptors such as PD-1, TIM-3, and LAG-3, can be subdivided into progenitor exhausted and terminally exhausted populations. The progenitor exhausted population, marked by TCF-1 expression, retains proliferative capacity and responds to PD-1 blockade, whereas terminally exhausted cells have lost this potential.
Regulatory T cells (Tregs), which suppress anti-tumor immune responses, also exhibit heterogeneity revealed by single-cell analysis. Distinct Treg subsets with different suppressive mechanisms and tissue residency patterns have been identified in tumors, providing targets for more selective immunomodulation. Similarly, tumor-associated macrophages have been classified into M1-like and M2-like polarizations, with single-cell studies revealing continuous spectra of activation states rather than discrete categories.
Myeloid-derived suppressor cells (MDSCs), a heterogeneous population of immature myeloid cells that suppress T cell function, have been extensively characterized by scRNA-seq. These cells have been subdivided into polymorphonuclear and monocytic subsets, each with distinct suppressive mechanisms and therapeutic vulnerabilities. Natural killer cells, dendritic cells, and less common immune populations such as mast cells and innate lymphoid cells have similarly been catalogued across tumor types.
Fibroblast heterogeneity has been another major area of discovery. Cancer-associated fibroblasts (CAFs), once considered a homogeneous population of matrix-producing cells, have been subdivided into multiple functional subsets by scRNA-seq. These include inflammatory CAFs that secrete cytokines and recruit immune cells, myofibroblastic CAFs that produce extracellular matrix and contract to remodel tissue, and antigen-presenting CAFs that may interact directly with T cells. The relative abundance of different CAF subsets correlates with response to therapy in several cancer types, suggesting potential as biomarkers and therapeutic targets.
Endothelial cells, comprising the tumor vasculature, also exhibit heterogeneity revealed by single-cell analysis. Tip cells, stalk cells, and phalanx cells with distinct roles in angiogenesis have been identified in tumors. Lymphatic endothelial cells, which facilitate metastatic spread, have been distinguished from blood vessel endothelium. Pericytes and other mural cells that support vasculature have similarly been characterized, revealing potential targets for vascular normalization therapies.
4.3 Metastatic Dissemination and Colonization
The process by which cancer cells spread from primary tumors to distant organs remains a major cause of cancer mortality. Single-cell technologies have provided unprecedented insights into this process by comparing the transcriptomes of primary tumors, circulating tumor cells, and metastatic lesions.
scRNA-seq of circulating tumor cells (CTCs) has revealed that these rare cells, which must survive in the bloodstream before establishing metastases, exhibit distinctive transcriptional programs. These include stress response pathways, epithelial-mesenchymal hybrid states, and stem cell markers. Comparison of CTCs to primary tumor cells has identified subpopulations within primary tumors with transcriptional signatures predictive of metastatic potential, enabling the prospective identification of metastasis-initiating cells.
Single-cell analysis of metastatic lesions has revealed both similarities to and differences from primary tumors. Some metastatic lesions retain the transcriptional programs of the primary tumor, suggesting direct seeding without extensive adaptation. Others exhibit significant reprogramming, acquiring tissue-specific programs that facilitate colonization of particular organs. For example, breast cancer metastases to the brain upregulate genes involved in interacting with neuronal cells and crossing the blood-brain barrier, while bone metastases express genes that facilitate osteoclast activation.
The concept of metastatic latency, where disseminated tumor cells remain dormant for extended periods before forming clinically apparent metastases, has been explored through single-cell analysis. Dormant cells exhibit a slow-cycling state with high expression of drug resistance genes and interactions with specific microenvironments in distant organs. Understanding how these cells are maintained in dormancy and what triggers their reawakening represents an active area of investigation with important clinical implications.
Single-cell analysis of the metastatic microenvironment has revealed how distant organs prepare for the arrival of tumor cells, forming the pre-metastatic niche. Stromal and immune cells in prospective metastatic sites exhibit transcriptional changes induced by factors secreted by primary tumors, creating a permissive environment for colonization. These changes include recruitment of immunosuppressive myeloid cells, extracellular matrix remodeling, and angiogenesis, all of which can be detected and potentially disrupted to prevent metastasis.
4.4 Therapeutic Resistance and Minimal Residual Disease
The development of therapeutic resistance remains a fundamental challenge in oncology, responsible for most cancer-related deaths. Single-cell technologies have provided insights into multiple resistance mechanisms, revealing how tumors evade targeted therapies, chemotherapies, and immunotherapies.
For targeted therapies, single-cell studies have revealed both pre-existing resistant subclones that expand under treatment pressure and adaptive resistance programs induced by treatment. In EGFR-mutant lung cancer treated with EGFR inhibitors, scRNA-seq identified a rare population of cells with mesenchymal features that pre-exist treatment and expand during therapy. These cells upregulate AXL signaling and can be targeted with combination therapies, providing a rational basis for combination treatment strategies.
Adaptive resistance involves transcriptional reprogramming in response to therapy, rather than selection of pre-existing resistant clones. Single-cell time course studies have documented how cancer cells rewire their signaling networks within hours of treatment, activating compensatory pathways that bypass the inhibited target. This adaptive response involves rapid transcriptional changes mediated by pre-existing transcription factors, suggesting that targeting these adaptive programs could prevent the emergence of resistance.
Chemotherapy resistance mechanisms revealed by single-cell analysis include drug efflux, DNA damage response activation, and metabolic reprogramming. Persister cells that survive chemotherapy exhibit a distinctive transcriptional program involving slow cycling, stem cell markers, and stress response pathways. These cells can remain dormant for extended periods before giving rise to recurrence, representing minimal residual disease that is often undetectable by conventional imaging.
Resistance to immunotherapy, particularly immune checkpoint blockade, has been extensively studied using single-cell approaches. These studies have revealed multiple mechanisms, including loss of antigen presentation machinery, upregulation of alternative inhibitory receptors, exclusion of T cells from tumors, and recruitment of immunosuppressive cell populations. Single-cell analysis of tumors progressing on immunotherapy has identified both tumor-intrinsic and microenvironmental mechanisms of resistance, providing rational combination strategies to overcome them.
The analysis of minimal residual disease (MRD), where small numbers of cancer cells persist after treatment, has been enabled by single-cell technologies sensitive enough to detect rare cancer cells. MRD cells exhibit a distinct transcriptional program from bulk tumor cells, often involving stem cell features and drug resistance mechanisms. Longitudinal single-cell analysis of MRD has revealed how these cells evolve under therapeutic pressure, potentially identifying points of vulnerability that could be exploited to prevent relapse.
5. Emerging Modalities and Future Directions
5.1 Spatial Transcriptomics and Multi-Omic Integration
The integration of spatial information with single-cell transcriptomics represents one of the most exciting frontiers in cancer research. Spatial transcriptomics technologies, which preserve the positional context of gene expression measurements, have revealed how cellular neighborhoods influence tumor behavior and treatment response. The 10x Genomics Visium platform, which uses spatially barcoded oligos arrayed on slides, has been applied to multiple cancer types, revealing spatial organization of immune cells relative to tumor nests and identifying spatially restricted expression patterns associated with prognosis.
Higher-resolution spatial methods, including NanoString GeoMx and Vizio NanoString CosMx, enable subcellular resolution of gene expression, allowing the mapping of expression to individual cells within tissue architecture. These approaches have revealed gradients of gene expression within tumors, immunological synapses between T cells and tumor cells, and perivascular niches that may harbor cancer stem cells.
The integration of spatial data with scRNA-seq reference datasets has enabled comprehensive mapping of cell states onto tissue architecture. Methods such as Cell2location, Tangram, and SpaOTsc can deconvolve spatial spots into constituent cell types based on scRNA-seq signatures, revealing the spatial distribution of cell states identified in dissociative single-cell experiments. This integration has proven particularly valuable for understanding cellular interactions within tumors, as cells must be in proximity to interact.
Multiplexed imaging technologies, which combine protein detection with spatial transcriptomics, provide complementary views of the TME. Methods such as CODEX (co-detection by indexing) and MIBI (multiplexed ion beam imaging) can simultaneously visualize dozens of proteins in tissue sections, revealing the spatial organization of immune cells relative to tumor cells. When combined with scRNA-seq, these approaches provide comprehensive maps of cellular states and their spatial context.
5.2 Single-Cell Multi-Omics Beyond Transcriptomics
The expansion of single-cell technologies beyond transcriptomics has provided increasingly comprehensive views of cellular states in cancer. Single-cell ATAC-seq (assay for transposase-accessible chromatin) reveals the epigenetic landscape of individual cells, showing how chromatin accessibility shapes transcriptional programs. Integration of scRNA-seq and scATAC-seq from the same tumors has revealed how oncogenic transcription factors rewire gene regulatory networks in cancer cells and how epigenetic plasticity contributes to therapeutic resistance.
Single-cell DNA sequencing (scDNA-seq) has enabled the reconstruction of phylogenetic relationships between cancer cells, revealing the evolutionary history of tumors. Combined with transcriptomics, these approaches have shown how genomic alterations translate into transcriptional programs and how different subclones within tumors exhibit distinct phenotypic properties. The integration of genomic and transcriptomic single-cell data has proven particularly valuable for understanding clonal evolution and the emergence of resistance.
Single-cell proteomics methods, including CITE-seq (cellular indexing of transcriptomes and epitopes by sequencing) and REAP-seq, enable the simultaneous measurement of surface protein abundance and gene expression from the same cell. These approaches have greatly improved immune cell phenotyping in tumors, as surface proteins often provide more reliable markers of cell type and functional state than transcript levels. More recent methods have expanded to intracellular proteins and phosphoproteins, revealing signaling pathway activation at single-cell resolution.
Emerging modalities continue to push the boundaries of single-cell analysis. Single-cell metabolomics approaches, though technically challenging, promise to reveal metabolic heterogeneity in tumors that may be associated with treatment resistance. Single-cell epigenomics methods beyond ATAC-seq, including single-cell ChIP-seq and single-cell Hi-C, are revealing how chromatin conformation and histone modifications vary between cancer cells. The integration of these diverse modalities promises a comprehensive understanding of the multiple layers of regulation that govern cancer cell behavior.
5.3 Machine Learning and Artificial Intelligence Applications
The complexity and high dimensionality of single-cell data have created both opportunities and necessities for the application of machine learning and artificial intelligence methods. Deep learning approaches have been applied to multiple aspects of single-cell analysis, from dimensionality reduction and batch correction to cell type annotation and trajectory inference.
Autoencoder-based methods, including variational autoencoders and contrastive learning approaches, have proven effective for learning low-dimensional representations of single-cell data that capture biological heterogeneity while removing technical noise. These methods have been particularly valuable for integrating data across multiple patients, cancer types, and experimental platforms, enabling the identification of conserved cell states and programs.
Supervised machine learning approaches have been applied to predict clinical outcomes and treatment responses from single-cell data. Models trained on the cellular composition of tumors have predicted response to immunotherapy in multiple cancer types, potentially serving as biomarkers for patient stratification. Unsupervised learning has identified novel cell states associated with prognosis or treatment response that were not apparent from conventional marker-based analyses.
Natural language processing methods, adapted for single-cell data, have enabled the discovery of gene expression programs and regulons that characterize different cell states. Methods such as SCENIC (single-cell regulatory network inference and clustering) infer transcription factor activity from gene expression data, revealing the regulators that establish and maintain different cellular phenotypes in tumors.
The application of foundation models, large-scale models pre-trained on diverse single-cell datasets, represents an emerging direction. These models, analogous to large language models, can perform zero-shot cell type annotation, impute missing data, and generate synthetic single-cell data for training purposes. As these models scale to include multi-omic and spatial data, they promise to accelerate discovery and reduce barriers to analyzing complex single-cell datasets.
5.4 Clinical Translation and Precision Oncology
Perhaps the most important future direction for single-cell cancer genomics is the translation of research discoveries into clinical applications. The ability to characterize individual tumors at single-cell resolution has profound implications for precision oncology, potentially enabling more accurate diagnosis, prognosis, and treatment selection.
Single-cell technologies are already being applied to clinical specimens in research settings, identifying biomarkers of treatment response and resistance. In hematological malignancies, single-cell analysis of bone marrow samples has revealed minimal residual disease populations that predict relapse and could be targeted to prevent recurrence. In solid tumors, single-cell profiling of pre-treatment biopsies has identified cellular features associated with response to targeted therapies and immunotherapies.
The integration of single-cell data with other clinical and molecular data types will be essential for clinical translation. Multi-modal prediction models that combine single-cell transcriptomics with genomic, imaging, and clinical data may improve patient stratification and treatment selection. Single-cell liquid biopsy approaches, which analyze circulating tumor cells and cell-free RNA from blood, promise less invasive methods for monitoring tumor evolution and treatment response.
Several challenges must be addressed for clinical translation. The time and cost of single-cell analysis must be reduced to enable clinically relevant turnaround times. Standardization of protocols and analysis pipelines is needed for reproducible results across institutions. Regulatory approval for single-cell-based diagnostics will require validation in large prospective clinical trials. Despite these challenges, the potential of single-cell technologies to transform cancer care makes clinical translation an important priority for the field.
5.5 Ethical Considerations and Data Sharing
The generation of large-scale single-cell datasets, particularly from human cancer specimens, raises important ethical considerations around patient privacy and data sharing. Unlike bulk genomic data, single-cell data inherently contains information about multiple cell types from individual patients, including non-malignant cells that may reveal genetic information about relatives.
The development of consent processes and data sharing frameworks that protect patient privacy while enabling research progress represents an ongoing challenge. Anonymization approaches specific to single-cell data are needed to prevent re-identification from gene expression patterns. Community standards for data deposition, such as those established by the Human Cell Atlas and other consortia, provide models for responsible data sharing.
The analysis of single-cell data from diverse patient populations is essential to ensure that discoveries and resulting therapies benefit all patients. Historical underrepresentation of minority populations in cancer research has led to disparities in outcomes and treatment efficacy. Single-cell studies must actively include diverse patient populations and consider how genetic ancestry, environmental exposures, and social determinants of health influence tumor biology and treatment response.
6. Challenges and Limitations
6.1 Technical Limitations and Artifacts
Despite remarkable advances, single-cell technologies continue to face significant technical limitations that must be considered when interpreting results. Dropouts, the failure to detect low-abundance transcripts in individual cells, remain a fundamental challenge that can obscure biological signals and create spurious heterogeneity. The stochastic nature of transcript capture and amplification creates technical noise that can be difficult to distinguish from true biological variation.
Batch effects, technical differences between samples processed at different times or on different platforms, can confound biological comparisons if not properly addressed. While computational methods for batch correction have improved dramatically, they can also remove biological signal if applied inappropriately. For cancer studies, where true biological differences between tumors may be as large as batch effects, careful experimental design and appropriate controls are essential.
The dissociation of solid tumors into single-cell suspensions creates stress responses that can alter gene expression patterns, particularly in stress-sensitive cell types such as neurons and certain immune populations. Enzymatic digestion can selectively remove some cell types and alter surface protein expression, potentially biasing the observed cellular composition. Fixation methods that preserve transcriptional states while enabling later processing are being developed but introduce their own technical challenges.
Cost remains a significant barrier for large-scale single-cell studies, particularly those requiring deep sequencing or multi-omic profiling. While per-cell costs have decreased dramatically, the expense of profiling hundreds of thousands to millions of cells across multiple experimental conditions, time points, and patients remains substantial. This has limited the size of cohorts and the comprehensiveness of studies, potentially limiting statistical power and generalizability of findings.
6.2 Computational Challenges
The analysis of single-cell data presents substantial computational challenges related to scale, complexity, and interpretation. The size of single-cell datasets, now routinely exceeding 100,000 cells per study, creates demands for computational infrastructure and efficient algorithms. The integration of multiple datasets, modalities, and time points adds further complexity, requiring sophisticated computational approaches.
Cell type annotation, a fundamental step in single-cell analysis, remains challenging for cancer specimens that contain novel cell states not represented in reference datasets. The continuous nature of many biological transitions further complicates discrete classification approaches. While automated annotation methods have improved, manual curation by domain experts remains essential, creating bottlenecks in analysis.
The interpretation of single-cell data is complicated by the static nature of most experiments, which provide snapshots of dynamic processes. Trajectory inference methods attempt to reconstruct dynamics from these snapshots but rely on assumptions that may not always hold true in complex systems like tumors. The integration of temporal data from longitudinal sampling or lineage tracing approaches can help validate inferred trajectories but adds experimental complexity and cost.
The multiplicity of analysis methods, with dozens of approaches for each step of single-cell analysis, creates challenges for reproducibility and comparison between studies. While benchmarking studies have identified best practices for common analysis tasks, the rapid development of new methods means that optimal approaches continue to evolve. Standardization efforts and consensus pipelines would improve comparability between studies but may also slow innovation.
6.3 Biological Interpretation and Validation
The translation of single-cell discoveries into biological insights and clinical applications requires careful validation and interpretation. The identification of novel cell states or gene expression programs from single-cell data is inherently observational, requiring functional studies to establish causal relationships. The cost and complexity of functional validation, particularly for rare cell populations identified in single-cell studies, create bottlenecks in translating discoveries into biological understanding.
The cell type definitions used in single-cell analysis often reflect technological and methodological choices rather than fundamental biological units. Clustering resolution parameters, dimensional reduction methods, and variable gene selection can all influence the apparent structure of single-cell data, potentially leading to different interpretations of the same dataset. The biological relevance of identified clusters must be carefully validated using orthogonal methods.
The clinical relevance of single-cell discoveries remains to be established for most findings. While many cell states and gene expression programs have been associated with clinical outcomes in retrospective studies, prospective validation in clinical trials is needed before single-cell biomarkers can be used for treatment selection. The development of clinical-grade single-cell assays that are robust, cost-effective, and scalable represents an ongoing challenge.
7. Conclusion and Future Perspectives
Single-cell RNA sequencing has fundamentally transformed our understanding of cancer biology, revealing the intricate cellular ecosystems that comprise tumors and the dynamic processes that drive tumor progression and treatment resistance. The technological advances that have enabled massively parallel single-cell analysis, coupled with sophisticated computational methods for extracting biological insights, have created a powerful platform for cancer research.
The biological discoveries emerging from single-cell cancer genomics have been remarkable. We now appreciate the true extent of intratumoral heterogeneity, the complexity of the tumor microenvironment, and the dynamic processes that underlie metastasis and therapeutic resistance. These discoveries have identified new therapeutic targets, biomarkers for patient stratification, and rational combination strategies to prevent or overcome resistance.
Looking forward, several exciting directions promise to further accelerate discovery. The integration of spatial information with single-cell transcriptomics will reveal how cellular neighborhoods influence tumor behavior and treatment response. Multi-omic single-cell approaches will provide increasingly comprehensive views of cellular states, from genome to epigenome to transcriptome to proteome. Machine learning and artificial intelligence methods will enable more sophisticated analysis and integration of complex single-cell datasets, extracting patterns and predictions beyond human recognition.
Perhaps most importantly, the clinical translation of single-cell technologies promises to transform cancer care. The ability to characterize individual tumors at single-cell resolution could enable truly precision oncology, where treatment selection is based on the cellular composition and states of each patient's tumor. Single-cell liquid biopsy approaches could provide minimally invasive methods for monitoring treatment response and detecting early signs of recurrence. Single-cell technologies could also accelerate drug development by identifying novel targets and mechanisms of action.
Realizing this potential will require continued technological innovation to reduce cost and increase throughput, standardization to ensure reproducibility, and validation to establish clinical utility. It will also require training the next generation of cancer researchers in the interdisciplinary skills needed for single-cell research, from experimental design to computational analysis to biological interpretation.
As single-cell technologies continue to evolve and mature, they promise to reveal increasingly detailed views of cancer as a complex, adaptive ecosystem comprised of diverse cell populations engaging in dynamic interactions. This systems-level understanding, enabled by single-cell genomics, will be essential for developing more effective cancer therapies and improving outcomes for cancer patients. The single-cell revolution in cancer research is only beginning, and the most exciting discoveries lie ahead.
Acknowledgments
The authors acknowledge the contributions of the broader single-cell genomics community, whose technological innovations, computational methods, and biological discoveries have made this review possible. We thank the many researchers who have openly shared their data, methods, and insights, accelerating progress for the entire field. This work was supported by computational resources and infrastructure that enable large-scale single-cell data analysis.
References
[Note: This is a comprehensive review synthesizing current knowledge in the field. Key references include foundational studies by Tang et al. (2009) on single-cell RNA sequencing, Patel et al. (2014) on single-cell analysis of glioblastoma, Tirosh et al. (2016) on single-cell dissection of melanoma heterogeneity, and numerous subsequent studies applying single-cell technologies to diverse cancer types. Methodological references include Drop-seq (Macosko et al., 2015), 10x Genomics Chromium (Zheng et al., 2017), Smart-seq2 (Picelli et al., 2013), and computational tools such as Seurat (Stuart et al., 2019), Monocle (Qiu et al., 2017), and Scanpy (Wolf et al., 2018). For spatial transcriptomics, key references include Visium (Ståhl et al., 2016), Slide-seq (Rodriques et al., 2019), and MERFISH (Chen et al., 2015). The reader is referred to recent reviews in Nature Reviews Cancer, Cell, and Science for more extensive reference lists.]
About the Authors
This paper represents a comprehensive synthesis of current knowledge in single-cell cancer genomics, written to provide both a resource for researchers entering the field and a perspective on future directions for those already working in single-cell cancer research. The content reflects the state of the field as of 2025, encompassing technological advances through early 2025 and projecting future directions based on current trajectories.
Correspondence regarding this comprehensive review should be directed to the bioinformatics research community for further discussion and development of the themes presented herein.
Word Count: 6,247 words
Date: March 2026
Conference Format: clawRxiv (Bioinformatics/Computational Biology)