Abstract Ovarian endometrioid carcinoma (OEC) is the second most common histotype associated with the most favorable prognosis among ovarian carcinomas. Similar to endometrial carcinoma, OEC exhibit significant molecular diversity. We aimed to develop a practical immunohistochemical (IHC) prognosticator by integrating IHC biomarkers with clinical substage, and surgical outcomes. This multi-institutional study included participants with primary invasive OEC from the Alberta Cancer Registry (AOVT), Mayo Clinic, and Disease of the Ovary and their Evaluation Study (DOVE). Pathology review excluded misclassified high-grade serous and mesonephric carcinomas leaving 422 OEC cases (AOVT N=158; DOVE N=143; Mayo N=121) with available tumor biospecimens who did not receive neoadjuvant treatment. Clinical characteristics included age, stage, grade, residual disease and 5-year survival. Clinical risk group was defined by combining stage and residual disease: Low (FIGO stage IA/IB & no macroscopic disease), Intermediate (stage IC-II& no macroscopic disease), High (stage III/IV or macroscopic disease). Tissue microarrays were stained for TP53, PMS2, MSH6, PGR, and CTNNB1 using IHC and were scored by a single pathologist. Tumors were hierarchically categorized as TP53-abnormal (TP53abn)/PGR-loss, mismatch-repair deficient (MMRd, via PMS2 and MSH6), nuclear-CTNNB1 (nCTNNB1), or no-specific-immunohistochemical-profile (NSIP). Overall survival (OS) at 5 years was compared across clinical risk and biomarker groups using Cox regression, adjusted for age and site to generate hazard ratios (HR), and 5-year survival rates (5-YSR). High clinical risk group was associated with worse 5-year OS compared to low & intermediate risk (p<0.0001). Hierarchical IHC biomarker groups were also associated with 5-year OS, adjusted for age and study site (p<0.0001), with consistently worse survival for combined TP53abn/PGR-loss (n=17, HR=6.07, 95% CI 2.78-13.26), PGR-loss only (n=55, HR=4.17, 95% CI 2.19-7.94), TP53abn only (n=29, HR=2.04, 95% CI 0.85-4.90), and better survival for nuclear-CTNNB1 (n=146, HR=0.22, 95% CI 0.07-0.65) compared to NSIP (n=132). Within the low-risk group, MMRd, nCTNNB1, or NSIP had greater than 97% 5-YSR, while OEC with TP53abn/PGR-loss or PR-loss only had less than 75% 5-YSR. Within the intermediate group, only nCTNNB1 exceeded a 5-YSR of greater than 97%. Within the high-risk group, 5YSR for nCTNNB1 OEC was 90.9%, compared to only 16.7% for TP53abn/PGR-loss. This IHC-based algorithm refines prognosis beyond clinical substage. It identifies low-risk patients with unfavorable prognosis (TP53abn/PGRloss, PGR loss only) who may benefit from adjuvant therapy, and also patients with favorable prognosis within the intermediate group for whom adjuvant therapy could be de-escalated. Further patient selection for chemo vs. hormone therapy in the high-risk group may be improved by biomarkers. Citation Format: Gamaliel Taengwa, Hunter J. Atkinson, Bryan M. McCauley, Sebastian M. Armasu, Chen Wang, Jennifer A. Doherty, Holly R. Harris, Ellen L. Goode, Martin Koebel, Stacey J. Winham. IHC-based prognostic sub-stratification of ovarian endometrioid carcinoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 1184.

Human genomic studies link reduced CUB domain-containing protein 1 (CDCP1) expression with myocardial recovery in heart failure. While CDCP1 regulates cardiac fibroblast proliferation in vitro, it's in vivo role in cardiac fibrosis remains unclear. Using a Cdcp1-knockout (KO) angiotensin II/phenylephrine mouse model, we show that Cdcp1 deletion reduces echocardiographic left ventricular mass, histologic cardiac fibrosis, and pro-fibrotic gene expression, along with decreased fibroblast activation and inflammatory markers. Spatial transcriptomics identified a pressure overload-expanded fibroblast subpopulation enriched for growth factor and TGF-β signaling (FB5), which was markedly attenuated in Cdcp1-KO hearts, alongside reduction of a pro-inflammatory cardiomyocyte subtype (CM4). Complementary studies in human ventricular fibroblasts demonstrate that CDCP1 knockdown reduced extracellular matrix gene expression and collagen I deposition. These findings establish CDCP1 as a regulator of cardiac fibrotic remodeling in vivo and open avenues for its further investigation as a potential therapeutic target.

Epstein-Barr virus-positive inflammatory follicular dendritic cell sarcoma (EBV+ IFDCS), recently reclassified in the 5th edition of the WHO classification of lymphoid neoplasms, is a rare mesenchymal and dendritic cell tumor distinct from conventional follicular dendritic cell sarcoma (FDCS) in cellular origin and clinicopathological features. To address this, we present the largest sequencing cohort of EBV+ IFDCS to date, combining targeted next-generation sequencing (NGS) of 12 cases and whole-exome sequencing (WES) of 3 cases. This study aims to elucidate the molecular characteristics of EBV+ IFDCS to better understand its pathogenesis and identify potential therapeutic targets. 12 EBV+ IFDCS cases were analyzed using next-generation sequencing (NGS) with a 506-gene panel, and whole-exome sequencing (WES) was performed on 3 cases. The analysis focused on identifying copy number variations (CNVs), gene fusions, and pathogenic mutations. KEGG pathway analysis was conducted to explore enriched oncogenic pathways. CNV analysis via WES identified focal chromosomal deletions in 2 of 3 cases: 7p and 14q deletions in Case 1, and 17p deletion plus deep deletions in NPRL2 (chromosome 3) and STK11 (chromosome 19) in Case 6. While pathogenic/drug-sensitive SNVs varied across the 12 patients (Fig. 2). Only Patient 3 (48-year-old female, splenic classical subtype EBV+ IFDCS) was TMB-H (11.2 mutations/Mb, ≥ 10 mutations/Mb as threshold), harboring NQO1 p.P187S (VAF = 32.6

Colorectal cancer (CRC) is a prevalent malignancy with high mortality. Neutrophil extracellular traps (NETs) are implicated in metastasis and chemotherapy resistance, making them potential biomarkers for prognosis and treatment response. We extracted NETs-related genes (NRGs) from neutrophil transcriptome data derived from in vitro-treated cells and used LASSO regression to construct a NETs risk score model. Then we validated it in multiple public CRC datasets. Serum citrullinated histone H3 (CitH3) levels were measured in 146 CRC patients and 49 healthy controls by enzyme-linked immunosorbent assay (ELISA) to evaluate diagnostic and prognostic utility, as well as chemotherapy response prediction. The model demonstrated robust prognostic performance (AUC = 0.745-0.762 for 1-5 year survival), with high-risk patients exhibiting significantly reduced survival (p < 0.0001) and compromised treatment efficacy. CitH3 levels were markedly elevated in CRC patients, particularly in those with poor prognosis and chemotherapy resistance, suggesting diagnostic and predictive utility. This study establishes NETs-based risk scoring as an effective tool for CRC prognosis and treatment response prediction, while serum CitH3 emerges as a promising biomarker for diagnosis and monitoring.

Digital pathology has transformed how pathologists review and interpret tissue specimens, enabling remote access, efficient storage, and advanced visualization. Systematically analyzing pathologists’ slide reviewing interactions can uncover opportunities to design AI-assisted tools that integrate seamlessly into routine practice. We present a proof-of-concept suite, PathInteract, for extracting pathology interaction signals from recorded slide-review videos. Mouse cursor movements, viewport actions (i.e., zooming and panning), and verbal narratives are extracted using multiple deep learning and computer vision moduals. Unlike prior methods requiring specialized software or equipment, our approach operates on screen recordings, enabling broader applicability. We developed PathInteract using QuPath-recorded diagnostic sessions with view-tracking logs and applied it to ten educational YouTube videos. Results revealed distinct interaction patterns between two pathologists with different levels of experience, tissue type, and use context. Cursor tracking and viewport detection achieved strong agreement with ground truth, while speech analysis highlighted differences in cell-level focus across diseases. PathInteract enables scalable analysis of reviewing interactions in digital pathology through video recording and supports repurposing of existing pathology videos towards building interpretable, multimodal pathology AI datasets.

Loss-of-function mutations in the genes encoding PINK1 and PRKN result in early-onset Parkinson disease (EOPD). Together, the encoded enzymes direct a neuroprotective pathway that ensures the elimination of damaged mitochondria via autophagy. We performed a genome-wide high-content imaging miRNA screen for inhibitors of the PINK1-PRKN pathway and identified all three members of the miRNA family 29 (miR-29). RNA sequencing revealed target genes regulated by miR-29 and identified ATG9A as a candidate gene. SiRNA-mediated ATG9A silencing phenocopied the effects of miR-29 and suppressed the initiation of PINK1-PRKN–mediated mitophagy. In addition, expression of ATG9A was able to rescue the effects of miR-29a, suggesting that ATG9A is primarily responsible for the inhibitory effect of miR-29. In an EOPD patient cohort, we further discovered two rare, potentially deleterious, ATG9A missense variants (p.R631W and p.S828L) and tested them experimentally in cells. Strikingly, neither EOPD ATG9A variant was able to rescue the phenotype suggesting they both act as loss-of-function mutations and might contribute to the etiology of disease. Together, our study validates miR-29 and its target gene ATG9A as novel regulators of PINK1-PRKN signaling. It further serves as proof-of-concept with the identification of novel, potentially disease-relevant EOPD variants specifically in mitophagy-regulating genes. The nomination of biological pathways is important for the stratification and treatment of patients that suffer from devastating diseases, such as EOPD.

Recovering high-quality microbial genomes from metagenomic sequencing data is essential for accurate profiling and understanding microbial variation. However, existing clustering methods often suffer from limited accuracy and scalability. Here we present MetaCAT (Metagenome Clustering and Association Tool), a framework that combines recovery of microbial genomes from metagenomic data and analysis of their associations with host traits. MetaCAT incorporates a Sparse Weighted Dirichlet Process Gaussian Mixture Model (SWDPGMM) to accurately and efficiently decompose complex datasets and combines k-mer frequency with read coverage to improve genome reconstruction. It also provides a dedicated workflow for microbial single-nucleotide polymorphism identification and metagenome-wide association studies with the host. MetaCAT outperforms existing methods in both clustering accuracy and computational efficiency across diverse datasets. Using metagenomic data from colorectal cancer cohorts, it revealed previously unrecognized marker species and microbial single-nucleotide polymorphisms associated with colorectal cancer. MetaCAT provides a scalable framework for microbial community profiling and advances our understanding of host-microbe interactions.

Abstract Background: Most colorectal cancer (CRC) arises from polyps and is mainly prevented by polypectomy. The most important polyps to manage with colonoscopy are those with highest CRC risk- namely, advanced polyps (> 1cm, villous histology or high-grade dysplasia (HGD). Yet, 48% of advanced polyps recur within 1 to 3 years of removal, and up to 5% of advanced polyps under surveillance still progress to CRC. We performed this pilot study to investigate molecular and microenvironment heterogeneities of polyps and how they might impact their clinical behavior. Methods: GeoMx spatial transcriptomics was performed on FFPE tissues from three different polyp outcome phenotypes (POPs) including the polyp that does not recur (POP-NR), that recurs following polypectomy but cured by colonoscopy (POP-R) or the polyp despite polypectomy develops CRC at the polypectomy(ies) site (POP-CRC). Normal colon and polyp with low or HGD from the index polyp were assessed from 6 patients with POP-NR, 9 with POP-R and 12 with POP-CRC with a minimum of 2 follow up colonoscopies at 3-year intervals. Epithelium was identified as PanCK positive, and stroma identified as PanCK negative and positive nuclear staining. Cell type deconvolution was implemented using single-cell adult human intestinal tract catalogue (Elmentaite et al., https://www.gutcellatlas.org/). Differential gene expression (DEG), cell type abundance and functional enrichment analyses was performed using linear mixed effects model, and observations with p-value < 0.05 and log2FC >= |1| are reported. Results: The greatest number of DEGs were identified in stroma of the index POP-CRC compared to POP-NR polyps (n = 17 down and 11 upregulated genes), followed by epithelium of the index POP-CRC vs POP-NR polyps (n = 10 down and one upregulated gene(s)). Between the index POP-CRC and POP-R, epithelium showed downregulation of 7 genes, and upregulation of no genes. In stroma two genes were down and none upregulated. Four genes were downregulated in stroma of the index POP-R vs POP-NR polyp and 3 in epithelium, and one gene was upregulated in stroma and another in epithelium. MZT2B was upregulated in stroma of both the index POP-R (log2FC = 1.09, p-value = 0.0014) and POP-CRC (log2FC = 1.29, p-value = 0.0001) and has been implicated as a prognostic marker associated with worse prognosis in certain cancers. IgM and IgA plasma cells, proximal progenitors, MMP9+ inflammatory macrophages and myofibroblasts were found to be downregulated in epithelium of index POP-CRC and POP-R compared to POP-NR polyps; while microfold cells were downregulated in stroma of index POP-CRC and POP-R compared to POP-NR polyps. Conclusions: The stromal microenvironment and less so, epithelial features present in the index polyp differ based on a polyp’s future clinical behavior. The polyp-immune interaction warrants further study as a potential prevention target against polyp progression. Citation Format: Mrunal Dehankar, Lisa A. Boardman, Daniel O'Brien, E. Aubrey Thompson, Jennifer M. Kachergus, Ji Shi, Alexej Abyzov, Milovan Suvakov, Rondell P. Graham, Chen Wang. Stromal immune composition significantly contributes to adenomatous polyp recurrence or progression to cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 4018.

Computational pathology has advanced rapidly in recent years, driven by domain-specific image encoders and growing interest in using vision–language models to answer natural-language questions about diseases. Yet, the core problem behind pathology question-answering remains unsolved, considering that a gigapixel slide contains far more information than necessary for a given question. Pathologists naturally navigate tissue and morphology complexity by scanning broadly, and zooming in selectively according to the clinical questions. Current models, in contrast, rely on uniform patch sampling or broad attention maps, often attending equally to irrelevant regions while overlooking key visual evidence. In this work, we try to bring models closer to how humans actually examine slides. We propose a question-guided, tissue-aware, and coarse-to-fine retrieval framework, HistoSelect, that consists of two key components: a group sampler that identifies question-relevant tissue regions, followed by a patch selector that retrieves the most informative patches within those regions. By selecting only the most informative patches, our method becomes significantly more efficient: reducing visual token usage by 70% on average, while improving accuracy across three pathology QA tasks. Evaluated on 356,000 question–answer pairs, our approach outperforms existing methods and produces answers grounded in interpretable, pathologist-consistent regions. Our results suggest that bringing human-like search and attention patterns into WSI reasoning is a promising direction for building practical and reliable pathology VLMs.

Deep learning models enable the prediction of clinical endpoints from whole-slide images (WSIs), but many such models function as "black boxes", lacking transparency about whether and which histomorphological patterns drive their predictions, hindering interpretability and clinical adoption. Here we propose a human-in-the-loop explanation framework, MorphoXAI, which provides both local and global interpretability for deep learning models by incorporating human-expert interpretations. At the global level, it reveals the histomorphological patterns on which the model consistently relies to distinguish between classes of WSIs, as well as the patterns associated with confusion between classes. At the local level, it indicates which of these patterns are used in the prediction of an individual WSI and which regions within the slide correspond to such patterns. We validated our method across multiple deep learning-based WSI analysis tasks spanning different tissue types. The results show that our framework generates explanations that accurately reflect the histomorphology underlying the model's predictions at both global and local levels. For interpretability and clinical utility in diagnostic contexts, human evaluation results showed that our explanations were easy to interpret, rich in diagnostic features, and directly helpful for diagnostic decision-making, thereby enhancing pathologist-AI collaboration. Our work highlights that unifying global and local explanations and grounding them in expert-interpreted morphology enhances the interpretability and verifiability of deep learning models, thereby facilitating the transparent deployment of such models in clinical practice.

Background:Multiplexed immunofluorescence (MxIF) enables high-dimensional immune cell phenotyping and detailed characterization of the tumor immune microenvironment (TIME), but complex cyclic staining can introduce signal degradation and uncertainty in cell-level measurements. In routine practice, pathologists rely on hematoxylin and eosin (H&E) staining as the primary reference for tissue morphology and frequently cross-reference H&E to interpret MxIF findings. Methods:We present a framework for aligning H&E and MxIF images at the tissue microarray (TMA) core level to support cross-modal analysis. Cell nucleus detections from each modality are used as anchor points, formulating the task as a point-set alignment problem. Coherent Point Drift (CPD) is applied for global alignment, followed by graph-matching-based refinement to improve local consistency. Evaluation on ovarian TMAs demonstrates consistent alignment performance for both restained and serial sections. We further explore the use of restained H&E as a reference for MxIF interpretation and investigate virtual H&E generation from MxIF data as a complementary approach when restained H&E is unavailable. Results:The aligned images enable quantitative assessment of cross-modal feature concordance between MxIF-derived measurements and H&E morphology. The generated virtual H&E demonstrated similar cell population compared to the real H&E. Conclusions:Cell-centric alignment can facilitate integrative multimodal histopathology analysis for TMA cores in spatial co-localization, cell feature validation, and virtual H&E generation.

Cancer stemness is a critical determinant of tumor progression and poor prognosis in non-small cell lung cancer (NSCLC), however, the mechanisms by how long noncoding RNAs (lncRNAs) regulate stemness-associated signaling pathways remain incompletely understood. Here, we identify long-intergenic non-coding RNA for kinase activation (LINK-A) as a key regulator of NSCLC stemness that is markedly upregulated in tumor tissues and serum. Functional analyses demonstrate that LINK-A enhances cancer stem-like properties and accelerates tumor growth in vivo. Mechanistically, LINK-A predominantly localizes to the cytoplasm, where it binds to the RNA-binding protein heterogeneous nuclear ribonucleoprotein K (hnRNPK) and promotes its cytoplasmic retention. Cytoplasmic hnRNPK associates with beta-catenin (β-catenin) and the deubiquitinase ubiquitin carboxyl-terminal hydrolase 9× (USP9X), preventing β-catenin ubiquitination and subsequently enhancing Wnt/β-catenin signaling. This activation induces the transcription of Nanog homeobox (NANOG) and POU class 5 homeobox 1 (POU5F1/OCT4), thereby sustaining NSCLC stemness. Collectively, these findings identify a previously unrecognized LINK-A/hnRNPK/USP9X/β-catenin signaling axis and highlight LINK-A as a potential biomarker and therapeutic target for NSCLC.

In antitumor activities, baicalin and astragaloside IV inhibit tumor growth, induce cell death, and restrain metastasis in various cancers. Generally, a mixture of massive herbs like scutellaria or astragalus matches with other drugs to reach a curative effect in traditional Chinese prescription. Therefore, researchers aspire to an effective type of drug combination that shows promoted absorption and higher bioavailability in preclinical studies. Here, we report an optical method to detect chiral baicalin and astragaloside IV and also monitor the absorption of different chirals in ovarian cells. Eventually, R-Baicalin-Astragaloside IV dual drugs combination shows promoted absorption of each other compared with single chiral drugs or another. Based on the optical method results, we designed a series of in vitro and in vivo experiments to explain and analyze the mechanism of the curative effect. Therein, the result reveals that the tumor-associated neutrophils were reduced via the down-regulated TLR4/MYD88/NF-κB pathway to increased PD-1/PD-L1 immune response in epithelial ovarian cancer under the influence of R-Baicalin-Astragaloside IV. Thus, this work offers a comprehensive report on structure-activity relationships of chiral and dual drug strategies to improve its bioavailability in therapy of ovarian cancer.

Abstract BRCA-associated homologous recombination deficiency (HRD) is present in ~50% of high-grade serous carcinomas (HGSC) and predicts sensitivity to platinum-based therapy. However, there is little understanding of why some patients with BRCA-deficient tumors experience poor outcomes. In a large HGSC cohort (n = 1389) including 282 individuals with pathogenic germline BRCA variants (gBRCApv), residual disease after primary surgery has limited prognostic effect in gBRCApv-carriers compared to non-carriers, and prognostic outcomes differ based on the mutation location within functional domains of the BRCA genes. Multi-omic profiling is performed on 154 tumors, enriched for patients with BRCA-deficient tumors that experienced short overall survival ( ≤ 3 years, n = 42). Patients with BRCA2-deficient HGSC and loss of NF1 survive twice as long as those without NF1 loss, whereas PIK3CA, RAD21 and MYC amplification define BRCA2-deficient HGSC with exceptionally short survival. Patients with BRCA1-deficient HGSC and a more elevated HRD score survive significantly longer. BRCA1-deficient tumors in short survivors have evidence of immunosuppressive c-kit signaling and EMT. Our findings confirm that outcome is not determined by BRCA status alone, but rather a combination of co-occurring genomic alterations, the extent of DNA repair deficiency, and the tumor-immune microenvironment.

Cartilage and synovium are essential tissues involved in joint-related diseases and traits, including osteoarthritis (OA), rheumatoid arthritis (RA), and human height. Although genome-wide association studies (GWAS) identify numerous risk loci for these traits, the molecular mechanisms underlying these associations, particularly those involving alternative splicing, remain poorly understood due to the lack of splicing-related genetic data in relevant tissues. To address this limitation, we generate a splicing quantitative trait loci (sQTL) resource for cartilage and synovium. We identify 2,796 independent cis-sQTLs and six trans-sGenes across the two tissues, including 179 tissue-specific cis-sQTLs. Fine-mapping analysis identifies 116 high-confidence functional sVariants predicted to affect splicing through splice site gain or loss, with approximately half located outside canonical splice site motifs. Integration of sQTL data with GWAS summary statistics reveals 12 osteoarthritis, 6 rheumatoid arthritis, and 183 height effector genes in joint tissues. Notably, seven of the 12 osteoarthritis effector genes show joint tissue-specific colocalization. Together with the finding that tissue-specific sGenes play crucial roles in tissue-related biological processes and diseases, our work highlights the significant impact of tissue-specific alternative splicing regulation on disease etiology and provides a valuable resource for further mechanistic validation.

CTNNB1- mutated hepatocellular carcinomas are characterized by a distinctive morphology and activation of the Wnt pathway. AXIN1 also plays a key role in the Wnt pathway, but the morphology of AXIN1-mutated tumors has not been examined. In addition, there are ongoing questions on the ability of AXIN1 mutations to activate the Wnt pathway in hepatocellular carcinoma. AXIN1 mutated tumors (N=18) were studied, along with control groups: CTNNB1 (N=17), APC (6), or "Other" genes in the Wnt pathway (5). Wnt pathway activation was studied by immunostains for beta-catenin and glutamine synthetase. Findings were supplemented by gene expression analysis using TCGA data. On histologic examination, the classic morphology associated with beta-catenin mutations was found in all 4 groups: 8/18 AXIN1 (44%), 10/17 CTNNB1 (59%), 4/6 APC (67%), and 1/5 Other (20%). By immunohistochemistry, Wnt pathway activation was found in 11/18 AXIN1 (61%), 15/17 CTTNB1 (88%), 6/6 APC (100%), and 5/5 (100%) of Other. In AXIN1- mutated tumors, the Wnt pathway was weakly activated. Glutamine synthetase stains also highlighted a new "progressed pattern" associated with distinct subnodules of staining. Tertiary lymphoid structures were uncommon except for cases with CTTNNB1 mutations plus additional mutations in the Wnt pathway. In summary, the classic morphology associated with CTNNB1 mutations is found in hepatocellular carcinomas with mutations in AXIN1 , APC , and other Wnt genes. AXIN1 mutated tumors have Wnt activation that is detectable but at lower levels than CTNNB1 mutated tumors. As tumors progress, their level of Wnt activation can change.

Abstract Spatial transcriptomics links gene expression to tissue architecture, providing a mechanistic view of cellular organization. Yet existing datasets cover few donors and miss the complexity of human disease. Experimental costs remain prohibitive, and large-scale profiling is impractically slow for population-level studies. Accurate computational methods are urgently needed. Predicting gene expression from standard histology, however, remains an open problem, as current approaches transfer poorly to unseen cohorts and diseases. Here, we present Phoenix, a (latent) flow matching generative model that infers pan-cancer spatially resolved single-cell gene expression with high accuracy. Phoenix analyzes treatment response in silico: Applied to 763 head and neck cancer patients, it identified three new spatial biomarkers that we validated across two cancers (breast cancer, n = 84; ovarian cancer, n = 157) and treatment regimens (platinum, trastuzumab). Phoenix generalizes beyond carcinomas: In a large sarcoma cohort (802 tissue microarray cores), it accurately predicted cell-type-specific signatures in held-out samples and captured chemotherapy-induced immune remodeling. Phoenix also extends across species: In a mouse model, it accurately predicted the expression of pancreatic cancer lineage markers and the mutant mKras^G12D allele in silico. In total, we evaluated Phoenix on over 10,000 patients. Our results establish virtual spatial transcriptomics as a scalable framework for studying tissue organization, therapeutic response, and disease mechanisms.