Reading the tumour at single-cell resolution: Original research in the Precision Therapeutics special issue Authors: Mohammad Asim Published: 2026 Precision oncology rests on a simple commitment that has proven difficult to honour, namely that the molecular features distinguishing one tumour from another can be measured with enough fidelity to change how a patient is assessed and treated. This virtual special issue of Translational Oncology was assembled to record where that commitment currently stands across a deliberately broad set of malignancies and methods. The collection contains 28 peer-reviewed papers; this editorial concerns the 27 original research articles, and I treat the single Review separately so that the experimental and computational work can be read as a continuous body of evidence. The 27 studies cover gastrointestinal (colorectal, gastric, hepatocellular and pancreatic), thoracic (lung adenocarcinoma and pleural mesothelioma), genitourinary (renal cell), central nervous system (glioma and primary CNS lymphoma), cutaneous (melanoma), skeletal (osteosarcoma), gynaecological (cervical) and haematological (NK/T-cell lymphoma) disease. Their methods extend from single-cell transcriptomics and integrative multi-omics, through perturbation-based mechanistic biology with in vitro and in vivo confirmation, to systems-pharmacology-driven drug discovery and, at the clinical end, biomarker analyses embedded in a randomised phase III trial. Fig. 1 arranges these papers along that discovery-to-clinic continuum and is the organising structure for the sections that follow.Fig. 1Thematic organisation of the 27 original research articles along a discovery-to-clinic continuum. Each node summarises one mechanistic theme and the number of contributing studies; chips give the tumour type and the principal molecular target or analytical method. The single Review in the issue is considered separately.Fig 1 dummy alt text Thematic organisation of the 27 original research articles along a discovery-to-clinic continuum. Each node summarises one mechanistic theme and the number of contributing studies; chips give the tumour type and the principal molecular target or analytical method. The single Review in the issue is considered separately. Single-cell and multi-omic dissection of the tumour microenvironment Five studies use single-cell and integrative analyses to resolve the cellular composition of tumours and to reduce that complexity to a usable signature. In colorectal cancer, an atlas of more than fifty thousand cells defined a transcriptionally active B-cell subset, designated C3, whose marker genes supported a survival-stratifying model [1]. In hepatocellular carcinoma, endothelial cells expressing APOH and exhibiting high PANoptosis activity were identified as a clinically relevant subpopulation, with immunofluorescence and CCK-8 assays anchoring the computational signal in tissue [4]. A breast-cancer analysis linked glycolytic metabolic reprogramming to immune suppression and inferior survival and assigned a functional role to PDCD1 in proliferation, migration and invasion [5]; a pancreatic-adenocarcinoma study converged on GBP4 and a five-gene programme (GBP2, KRT6A, MMP7, BCAT1, SPRR1A) validated in cell and xenograft models [19]; and a lung-adenocarcinoma study described palmitoylation-defined cellular states and an ASPH-linked twelve-gene risk signature trained on TCGA and tested in GEO cohorts [20]. The recurring workflow, resolving heterogeneity, isolating a cell state, deriving a signature and nominating a target, defines much of contemporary translational discovery; its principal limitation is that signatures built on reference cohorts require external and, ultimately, prospective confirmation. Functionally validated targets and mechanisms A second and larger group of papers moved beyond association to perturbation, applying knockdown, overexpression, xenografts, chromatin immunoprecipitation and luciferase reporter assays to establish causality. Several biological hubs recurred. Ferroptosis featured repeatedly: miR-4537 restrained ferroptosis by targeting the 3′-UTR of MIOX in renal cell carcinoma [6]; TRIM26 acted as a dual regulator of ferroptosis and chemoresistance in gastric cancer through HSF1 ubiquitination and cancer-associated-fibroblast-derived exosomal miR-24-3p [11]; and a CHPT1–pSTAT3–SLC7A11 axis governed progression and ferroptosis susceptibility in pancreatic cancer, with ferroptosis induction (Erastin) acting in concert with gemcitabine [15]. The m6A epitranscriptome recurred through ZC3H13-dependent regulation of LMO2 in clear-cell renal carcinoma [2] and YTHDF2-mediated stabilisation of SLC16A9 that conferred adriamycin resistance in NK/T-cell lymphoma [9]. Transcriptional control of stemness was addressed by the PRDM1-driven transactivation of Myc in gastric cancer [8] and by MANF signalling through STAT3/TGF-β/SMAD4/p38 in glioma [13]. These were complemented by a KRAS–ETV4–PD-L1 immune-escape mechanism in lung adenocarcinoma [10], the prognostic role of PRMT8 in BCL2L11-deleted multiple primary lung cancer [3], and NMNAT3-dependent NAD⁺ reprogramming in melanoma [23]. Read together, these papers supply druggable nodes and defined resistance mechanisms rather than correlations alone, which is the contribution mechanistic translational work should make. Multi-omics prognostic and drug-response signatures A third cluster constructed and benchmarked risk models that couple prognosis to the immune microenvironment and to predicted drug sensitivity. ALAD was proposed as a prognostic biomarker associated with metabolic pathways and CD8⁺ T-cell-mediated immunity in renal cell carcinoma [7]; a combined ferroptosis-and-senescence model in cervical cancer was reinforced with clinical-sample validation and correlates of immunotherapy response, including TIDE and IPS scores [18]; an ACTA2-based clinical score was advanced for colorectal adenocarcinoma [21]; a glycerophospholipid-metabolism (GAS) score for osteosarcoma was reported to outperform 42 published signatures (C-index > 0.7) [25]; a shared PANoptosis programme linked lung adenocarcinoma and osteoarthritis through five genes [26]; and a regulated-cell-death framework predicted prognosis and drug sensitivity in gastric adenocarcinoma using oncoPredict and GDSC pharmacogenomics [27]. The appeal of these models is immediate, since they offer stratification together with a therapeutic direction. The corresponding limitation, which the field must keep in view, is that most rest on retrospective public cohorts (TCGA, GEO and comparable resources), so their generalisability and the distance from a risk score to a clinically actionable decision remain the central unresolved questions. Computational drug discovery and repurposing A fourth group applied network and systems pharmacology to discovery and repurposing. Two studies interrogated natural-product chemical space: indole-derived molecules against NAFLD-derived hepatocellular carcinoma, converging on JUN and the AGE-RAGE signalling axis [12], and Platycladi Folium phytochemicals against hepatocellular carcinoma, with ESR1 as a hub and stability assessed by molecular docking and density functional theory [14]. Two others pursued repurposing of approved agents: Erianin against glioma and concurrent morphine tolerance through the JAK2-STAT3 pathway, supported by a glioma-bearing morphine-tolerant mouse model and miRNA integration (miR-375, miR-20a) [22], and the antiparasitic thiabendazole in lung adenocarcinoma, with macrophages implicated and four diagnostic genes (ACE, ALDOA, MME, PMP22) nominated [24]. Repurposing is attractive because it can shorten the path to the clinic, though in silico predictions remain hypotheses until confirmed at the bench and characterised pharmacologically. The animal data in the Erianin study illustrate the bridging that converts a prediction into a credible lead. Clinically anchored biomarkers and predictive models Two contributions sit closest to practice and, to my reading, represent the strongest material in the issue. In pleural mesothelioma, a setting with very few validated biomarkers, twelve angiogenesis-related microRNAs were measured in FFPE samples from 236 of the 448 patients (50.8%) in the MAPS phase III trial (NCT00651456); low miR-193b-3p was independently associated with longer overall survival (adjusted HR = 0.87, 95% CI 0.81–0.93, p < 0.001) and progression-free survival (adjusted HR = 0.91, 95% CI 0.85–0.97, p = 0.0042), and miR-132-3p showed a significant interaction with the bevacizumab-containing regimen (p = 0.004), marking it as predictive rather than only prognostic [16]. Embedding a biomarker in randomised-trial material is the evidentiary standard that precision oncology should aim for. In relapsed primary CNS lymphoma, a C5.0 decision tree built on 209 patients prognosticated the outcome of salvage stereotactic radiosurgery from Karnofsky performance status, deep-brain involvement and IELSG score, with sensitivities of 0.880–1.000 in validation [17]; the model is transparent and clinic-ready, and it produces a decision rather than an association. What the issue indicates Three observations follow from reading the 27 papers together. The first is that the field is converging on a small number of biological hubs, among them ferroptosis and other regulated-cell-death programmes, the m6A epitranscriptome, metabolic reprogramming and tumour-microenvironment crosstalk, that recur across unrelated cancers and suggest genuinely shared vulnerabilities. The second is methodological maturity, in that single-cell resolution and machine learning are now routine, and the stronger papers pair them with functional confirmation rather than presenting computation as an endpoint. The third, and the most consequential, is that a real distance still separates computationally or retrospectively derived signatures from prospectively validated, clinically actionable tools. The contributions that advance the field most are those that close the loop from data to function to patients, and I would have the field hold that closure as the standard; in practical terms this means external and multi-centre validation, prospective cohorts where feasible, wet-laboratory confirmation of in silico hits, and reporting transparent enough to support independent reproduction. Declaration of competing interest The author is Co-Editor-in-Chief of Translational Oncology. No other competing interests are declared.