npj Precision Oncology

Cancer genome projects have produced multidimensional datasets on thousands of samples. Yet, depending on the tumor type, 5-50% of samples have no known driving event. We introduce a semi-supervised method called Learning UnRealized Events (LURE) that uses a progressive label learning framework and minimum spanning analysis to predict cancer drivers based on their altered samples sharing a gene expression signature with the samples of a known event. We demonstrate the utility of the method on the TCGA dataset for which it produced a high-confidence result relating 53 new to 18 known mutation events including alterations in the same gene, family, and pathway. We give examples of predicted drivers involved in TP53, telomere maintenance, and MAPK/RTK signaling pathways. LURE identifies connections between genes with no known prior relationship, some of which may offer clues for targeting specific forms of cancer. Code and Supplemental Material are available on the LURE website https://sysbiowiki.soe.ucsc.edu/lure.

BackgroundAccurate molecular profiling and prognostication from routine histopathology slides could transform precision oncology. We developed a Vision Transformer (ViT)-based multi-instance learning (MIL) framework for combined predictions of 32 solid tumour types, TP53 biomarker detection, and survival prediction directly from Whole Slide Images (WSIs). Methods11,060 primary tumours were curated from the TCGA Pan-Cancer Atlas with corresponding somatic mutations, RNA-seq, and clinical outcome data. TP53 alterations were classified as pathogenic drivers using COSMIC and hotspot annotations. WSIs underwent tissue masking, quality control, stain normalisation, and patch extraction (518 x 518) at 6x downsampling. Each patch was encoded by a ViT into a 768-dimensional embedding, which formed a token sequence for a 6-layer Transformer aggregator with learnable classification and positional embeddings. Seven task heads were developed to generate predictions for various outcomes, including cancer type, TP53 mutation status, TP53 RNA expression levels, overall survival (OS), progression-free interval (PFI), and the corresponding times for OS and PFI. The training process had two stages. First, the model was trained on tumour tissue patches from WSIs at five magnifications. In the second stage, it was fine-tuned using patches from all tissue regions with a content-aware strategy, updating all MIL layers for a maximum of 150 epochs at a learning rate of 1 x 10-. The models performance was evaluated on an independent validation set of 1,729 slides using classification metrics, including the area under the receiver operating characteristic curve (AUROC), regression metrics, and Concordance indices (C-index). ResultsThe multi-resolution ViT-based MIL model achieved an AUROC of 0.775 (95% CI: 0.749-0.801) for TP53 mutation detection on the validation set, demonstrating strong overall performance across classification and survival prediction tasks. The fine-tuned model attained robust performance across the tasks, with 0.7569 accuracy for cancer classification, 0.745 AUROC for TP53 mutation detection, C-indices of 0.686 and 0.650 for OS and PFI, and a mean squared error of 1.072 for TP53 RNA expression level estimation. The fine-tuned model attained an accuracy of 65.9% (95% CI: 0.636-0.681) in tumour classification and an AUROC of 0.766 (95% CI: 0.743-0.789) for detecting TP53 mutations on the external validation set. However, most tumour classes, aside from ovarian cancer, reached an AUROC above 0.88 with class-specific thresholding using the Youden Index. This indicates strong generalisation across 32 tumour types, providing reasonable molecular profiling but offering limited prognostic utility in surgical oncology. ConclusionA ViT-based MIL model can simultaneously infer tumour taxonomy, TP53 mutation status, and TP53 RNA expression levels directly from WSIs, with performance comparable to conventional genomic assays, while prognostic risk remains limited. This integrated, slide-level approach offers a scalable pipeline toward computational pathology.

High-grade serous ovarian cancer (HGSOC) is the most common and lethal subtype of ovarian cancer, characterized by high recurrence rates and limited treatment options following chemotherapy resistance. Its significant heterogeneity poses major challenges for effective therapy and clinical outcomes. In this study, we present a systems-level analysis of spatial transcriptomics data to characterize tumor heterogeneity in post-neoadjuvant chemotherapy HGSOC patients. By integrating gene expression profiles with spatial localization and histological context, we quantified hallmark pathway activities across tissue regions. The computed pathway scores were then used for clustering to investigate intra-tumoral heterogeneity. We also constructed gene co-expression network within tumor-enriched regions. Finally, we examined the association of these co-expressed modules with treatment response. Clustering based on pathway activity scores revealed spatially distinct regions enriched for different hallmark pathways, uncovering functionally diverse cellular subpopulations within the tumor microenvironment. Tumor cell-enriched clusters show difference in pathways related to proliferation, metabolism, immune signaling and stress response, while fibroblast-rich regions exhibit upregulation of epithelial-mesenchymal transition (EMT). Unsupervised co-expression analysis further revealed gene modules associated with both biological processes and clinical phenotypes. Poor responders exhibit higher expression of gene modules involved in stress response, ribosomal function, oxidative phosphorylation, and cell-cycle regulation. In contrast, good responders show elevated activity in modules enriched for immune activation, extracellular matrix (ECM) remodeling, and inflammatory signaling. Our findings provide insights into spatially resolved functional states, tumor heterogeneity, and molecular features associated with treatment response, offering a foundation for precision oncology approaches in ovarian cancer.

Evaluation of bone marrow morphology by experienced hematologists is key in the diagnosis of myeloid neoplasms, especially to detect subtle signs of dysplasia in myelodysplastic neoplasms (MDS). The majority of recently introduced deep learning (DL) models in cytomorphology rely heavily on manually drafted cell-level labels, a time-consuming, laborious process that is prone to substantial inter-observer variability, thereby representing a substantial bottleneck in model development. Instead, we used robust image-level labels for end-to-end DL and trained several state-of-the-art computer vision models on bone marrow smears of 463 patients with MDS, 1301 patients with acute myeloid leukemia (AML), and 236 bone marrow donors. For the binary classifications of MDS vs. donors and MDS vs. AML, we obtained an area-under-the-receiver-operating-characteristic (ROCAUC) of 0.9708 and 0.9945, respectively, in our internal test sets. Results were confirmed in an external validation cohort of 50 MDS patients with corresponding ROCAUC of 0.9823 and 0.98552, respectively. Explainability via occlusion sensitivity mapping showed high network attention on cell nuclei not solely of dysplastic cells. We not only provide a highly accurate model to detect MDS from bone marrow smears, but also underline the capabilities of end-to-end learning to solve the bottleneck of time-consuming cell-level labeling.

Oncogenic viruses like human papilloma virus (HPV) or Epstein Barr virus (EBV) are a major cause of human cancer. Viral oncogenesis has a direct impact on treatment decisions because virus-associated tumors can demand a lower intensity of chemotherapy and radiation or can be more susceptible to immune check-point inhibition. However, molecular tests for HPV and EBV are not ubiquitously available.\n\nWe hypothesized that the histopathological features of virus-driven and non-virus driven cancers are sufficiently different to be detectable by artificial intelligence (AI) through deep learning-based analysis of images from routine hematoxylin and eosin (HE) stained slides. We show that deep transfer learning can predict presence of HPV in head and neck cancer with a patient-level 3-fold cross validated area-under-the-curve (AUC) of 0.89 [0.82; 0.94]. The same workflow was used for Epstein-Barr virus (EBV) driven gastric cancer achieving a cross-validated AUC of 0.80 [0.70; 0.92] and a similar performance in external validation sets. Reverse-engineering our deep neural networks, we show that the key morphological features can be made understandable to humans.\n\nThis workflow could enable a fast and low-cost method to identify virus-induced cancer in clinical trials or clinical routine. At the same time, our approach for feature visualization allows pathologists to look into the black box of deep learning, enabling them to check the plausibility of computer-based image classification.

BackgroundSingle-cell RNA-seq has emerged as an innovative technology used to study complex tissues and characterize cell types, states, and lineages at a single-cell level. Classification of bulk tumors by their individual cellular constituents has also created new opportunities to generate single-cell atlases for many organs, cancers, and developmental models. Despite the tremendous promise of this technology, recent evidence studying epithelial tissues and diverse carcinomas suggests the methods used for tissue processing, cell disaggregation, and preservation can significantly bias gene expression and alter the observed cell types. To determine whether sarcomas - tumors of mesenchymal origin - are subject to the same technical artifacts, we profiled patient-derived tumor explants (PDXs) propagated from three aggressive subtypes: osteosarcoma, Ewing sarcoma (ES), desmoplastic small round cell tumor (DSRCT). Given the rarity of these sarcoma subtypes, we explored whether single-nuclei RNA-seq from more widely available archival frozen specimens could accurately be identified by gene expression signatures linked to tissue phenotype or pathognomonic fusion proteins. ResultsWe systematically assessed dissociation methods across different sarcoma subtypes. We compared gene expression from single-cell and single-nucleus RNA-sequencing of 125,831 whole-cells and nuclei from ES, DSRCT, and osteosarcoma PDXs. We detected warm dissociation artifacts in single-cell samples and gene length bias in single-nucleus samples. Classic sarcoma gene signatures were observed regardless of dissociation method. In addition, we showed that dissociation method biases can be computationally corrected. ConclusionsWe highlighted transcriptional biases, including warm dissociation and gene-length biases, introduced by the dissociation method for various sarcoma subtypes. This work is the first to characterize how the dissociation methods used for sc/snRNA-seq may affect the interpretation of the molecular features in sarcoma PDXs.

Cancer remains a leading cause of morbidity and mortality worldwide. Despite advances in genomics, identifying clinically relevant subtypes of cancer remains challenging due to its complex and heterogeneous nature. In this work, we propose DeepGraphMut (DGM), a novel graph-based deep-learning pipeline that integrates somatic mutation data with protein-protein interaction (PPI) networks. By employing a graph autoencoder with a graph attention layer and a node-level attention decoder, DGM generates patient-specific clinically relevant encodings for unsupervised and supervised tasks. We demonstrate the effectiveness of DGM across 16 cancer types comprising of 7352 samples from The Cancer Genome Atlas (TCGA). Unsupervised clustering reveals distinct subtypes with significant survival differences in 11 cancer types. In supervised analysis using a Cox regression model, DGM demonstrates excellent performance in predicting survival outcomes, achieving a high concordance index (c-index) value in the range of 0.7 across most cancers, underscoring its robust predictive performance using only somatic mutation data. Furthermore, DGM outperforms its lightweight variant and the network-based stratification method in both unsupervised and supervised analyses. In summary, this study presents a promising approach for cancer subtype identification and prognosis, especially in resource-limited settings where multi-omics data may not be readily available. By leveraging the strengths of graph learning and network biology, DGM offers a valuable tool for advancing personalized medicine.

Clinical research relies on high-quality patient data, however, obtaining big data sets is costly and access to existing data is often hindered by privacy and regulatory concerns. Synthetic data generation holds the promise of effectively bypassing these boundaries allowing for simplified data accessibility and the prospect of synthetic control cohorts. We employed two different methodologies of generative artificial intelligence - CTAB-GAN+ and normalizing flows (NFlow) - to synthesize patient data derived from 1606 patients with acute myeloid leukemia, a heterogeneous hematological malignancy, that were treated within four multicenter clinical trials. Both generative models accurately captured distributions of demographic, laboratory, molecular and cytogenetic variables, as well as patient outcomes yielding high performance scores regarding fidelity and usability of both synthetic cohorts (n=1606 each). Survival analysis demonstrated close resemblance of survival curves between original and synthetic cohorts. Inter-variable relationships were preserved in univariable outcome analysis enabling explorative analysis in our synthetic data. Additionally, training sample privacy is safeguarded mitigating possible patient re-identification, which we quantified using Hamming distances. We provide not only a proof-of-concept for synthetic data generation in multimodal clinical data for rare diseases, but also full public access to synthetic data sets to foster further research. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=107 SRC="FIGDIR/small/23298247v1_ufig1.gif" ALT="Figure 1"> View larger version (20K): org.highwire.dtl.DTLVardef@1da27bborg.highwire.dtl.DTLVardef@166982dorg.highwire.dtl.DTLVardef@90cfcborg.highwire.dtl.DTLVardef@13a5957_HPS_FORMAT_FIGEXP M_FIG C_FIG

Therapeutic drug treatments of solid tumors are often undermined by various resistance mechanisms. Identification of drug-resistance phenotypes at the single cell level is challenging because conventional molecular methods are cell-destructive, labor-intensive, and cost-prohibitive. To overcome these challenges, we developed an orthogonal approach to drug-resistance phenotyping, through the use of deep-learning-driven morphology analysis of single, high resolution cell images. Specifically, we trained deep learning-based drug resistance classifiers using cell images from 5 different cell lines that were rendered resistant to 5 different therapeutic agents, using a foundation model framework. With high accuracy, the classifier correctly predicted naive or resistance phenotypes across all cancer types and across all the therapeutic agent types (chemotherapeutic, targeted) tested. These results showed that morphology can capture complex phenotype information in the context of drug treatment. To demonstrate the potential clinical utility of the drug resistance classifier, it was applied to a dissociated tumor biopsy and the resulting phenotype predictions were in close concordance with scRNASeq analysis of the biopsy. Our study highlights the potential of deep-learning-driven morphology analysis to provide complex phenotype information, and ultimately shape oncology drug treatment strategies at the patient-level in a clinical context.

Progress in precision oncology, including biomarker discovery and individualized treatment selection, is limited by the complexity of clinico-genomic data and the scarcity of large multimodal patient cohorts. Here, we introduce PanoraOnc, a pan-cancer artificial intelligence (AI) model pretrained on real-world clinical, genomic, and imaging data from 84,131 patients spanning 66 cancer types. PanoraOnc enables transferable treatment outcome prediction through pan-cancer pretraining and generalizes to unseen cohorts across cancer types, institutions, and therapeutic settings. Evaluation and fine-tuning were performed on cohorts comprising diverse modalities, including clinical features, targeted gene panels, immunofluorescence imaging, whole-exome sequencing, and transcriptomic profiles. Across these settings, PanoraOnc consistently outperforms statistical, machine-learning, survival, and AI baselines, with the largest improvements observed in zero- and few-shot scenarios, demonstrating that large-scale clinico-genomic pretraining enables robust and generalizable outcome predictions across previously unseen conditions. In addition, PanoraOnc supports biomarker discovery through explainable AI, revealing both established and underappreciated features, including tumor-infiltrating clonal hematopoiesis, oncogenic signaling pathways, and DNA damage response mechanisms in immunotherapy-treated melanoma and non-small cell lung cancer. Furthermore, PanoraOnc enables the identification of patient subgroups potentially benefitting from alternative treatments by estimating personalized treatment outcomes across therapeutic scenarios. These findings establish pan-cancer multimodal pretraining as a scalable paradigm for AI-assisted discovery in precision oncology.

PurposeTo evaluate whether open-source histopathology foundation model pipelines, paired with attention-based multiple instance learning (MIL), can accurately classify molecular subtypes of endometrial cancer (EC) from whole-slide images (WSIs) and maintain performance in a real-world, independent cohort. MethodsWe assembled a public discovery cohort of 815 patients (1,195 WSIs) from The Cancer Genome Atlas and Clinical Proteomic Tumor Analysis Consortium, and an independent external cohort of 720 patients (1,357 WSIs) with molecular subtyping determined by mismatch repair immunohistochemistry plus TP53 and POLE sequencing. Four ImageNet-pretrained convolutional neural networks (CNNs) and six open-source foundation encoders using two MIL aggregation strategies (TransMIL and CLAM) were benchmarked within the STAMP pipeline. Models were trained with five-fold cross-validation and evaluated on an independent cohort. Macro-area under the receiver operating characteristic curve (AUC) was the primary outcome. ResultsIn cross-validation, foundation models outperformed CNNs (macro-AUC 0.799-0.860 vs 0.715-0.829). The best configuration (Virchow2 with CLAM) achieved macro-AUC 0.860 (95%CI, 0.839-0.880), macro-F1 score 0.607, and balanced accuracy 0.647. External validation showed substantial degradation for CNNs, while foundation models retained higher discrimination (macro-AUC 0.667-0.780). UNI2 with CLAM had the highest external macro-AUC (0.780), and Virchow2 with CLAM had the best balanced accuracy (0.525). Subtype-level AUCs for UNI2 with CLAM were highest for p53abn (0.851). ConclusionsOpen-source foundation model pipelines with attention-based MIL can deliver accurate and generalizable molecular subtyping of EC directly from WSIs. These models outperform CNNs in real-world validation, supporting their potential as scalable, cost-effective tools to guide precision oncology and triage confirmatory molecular testing.

BackgroundCancer morbidity disproportionately affects patients in low- and middle-income countries (LMICs), where timely and accurate tumor profiling is often nonexistent. Immunohistochemistry-based assessment of estrogen receptor (ER) status, a critical step to guide use of endocrine therapy (ET) in breast cancer, is often delayed or unavailable. As a result, ET is often prescribed empirically, leading to ineffective and toxic treatment for ER-negative patients. To address this unmet need, we developed ESPWA (Estrogen Receptor Status Prediction for Haitian patients using deep learning-enabled histopathology Whole Slide Imaging Analysis), a deep-learning (DL) model that predicts ER status directly from hematoxylin-and-eosin (H&E)-stained whole slide images (WSIs). MethodsWe curated two cohorts of H&E WSIs with tissue-matched ER status: The Cancer Genome Atlas (TCGA, n = 1085) and Zanmi Lasante (ZL, n = 3448) from Haiti. We trained two models using weakly supervised attention-based multiple instance learning: a "TCGA" model, trained on TCGA data, and ESPWA, trained on the ZL dataset. Model performance was evaluated using 10-fold cross validation. ResultsPerformance of the "TCGA" model was sensitive to the domain shift between the TCGA and ZL datasets, with a performance of an area under receiver operating characteristic (AUROC) of 0.846 on the TCGA test sets and 0.671 on the ZL test sets. Compared to the "TCGA" model, ESPWA demonstrated improved performance on the ZL cohort (AUROC=0.790; p=0.005). Subgroup analyses revealed clinically relevant populations in which ESPWA demonstrated improved performance relative to the overall cohort. Finally, ESPWA outperformed an academic breast pathologist (accuracy: 0.726 vs 0.639 respectively; p <0.001) in determining ER status from H&E WSIs. ConclusionESPWA ("Hope" in Haitian Creole) offers an accessible framework to identify individualized therapeutic insights from H&E WSIs in LMICs. We have initiated clinical trials, using ESPWA, in ZL and sub-Saharan African countries to inform precision-based use of ET for prospective patients.

BackgroundGastrointestinal stromal tumor (GIST) is the most common gastrointestinal mesenchymal tumor, driven by tyrosine-protein kinase KIT and platelet-derived growth factor receptor A (PDGFRA) mutations. Specific variants, such as KIT exon 11 deletions, carry prognostic and therapeutic implications, whereas wild-type (WT) variants derive limited benefit from tyrosine kinase inhibitors (TKIs). Given the limited reproducibility of established clinicopathological risk models, deep learning (DL) applied to whole-slide images (WSIs) emerged as a promising tool for molecular classification and prognostic assessment. Patients and methodsWe analyzed 8398 GIST cases from 21 centers in 7 countries, including 7238 with molecular data and 2638 with clinical follow-up. DL models were trained on WSIs to predict mutations, treatment sensitivity, and recurrence-free survival (RFS). ResultsDL predicted mutational status in GIST from WSIs, with area under the curve (AUC) of 0.87 for KIT, 0.96 for PDGFRA. High performance was observed for subtypes, including KIT exon 11 delinss 557-558 (0.67) and PDGFRA exon 18 D842V (0.93). For therapeutic categories, performance reached 0.84 for avapritinib sensitivity, 0.81 for imatinib sensitivity. DL models predicted RFS, with hazard-ratios (HR) of 8.44 (95%CI 6.14-11.61) in the overall cohort and 4.74 (95%CI 3.34-6.74) in patients receiving adjuvant therapy. Prognostic performance was comparable to pathology-based scores, with highest discrimination in the overall cohort and in patients without adjuvant therapy (9.44, 95%CI (5.87-15.20)). ConclusionDL applied to WSIs enables prediction of molecular alterations, treatment sensitivity, and RFS in GIST, performing comparably to established risk scores across international cohorts, providing a baseline for future multimodal predictors. HighlightsO_LIDeep learning on histology predicts KIT and PDGFRA mutations in a large international cohort of GISTs from multiple centers C_LIO_LIWhole-slide image models stratify recurrence-free survival comparable to pathology-based risk scores C_LIO_LIPrognostic value of deep learning is preserved in adjuvant therapy subgroups, supporting treatment duration decisions C_LI O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=117 SRC="FIGDIR/small/26345350v1_ufig1.gif" ALT="Figure 1"> View larger version (36K): org.highwire.dtl.DTLVardef@652548org.highwire.dtl.DTLVardef@729a2borg.highwire.dtl.DTLVardef@1e7b6b9org.highwire.dtl.DTLVardef@18d6721_HPS_FORMAT_FIGEXP M_FIG O_FLOATNOGraphical abstract.C_FLOATNO Overview of study design and dataset characteristics. (A) Multinational collection of WSIs from seven countries (Spain, France, Italy, Germany, the Netherlands, Poland, and Japan), followed by standard image preprocessing with the STAMP pipeline and clinical data preprocessing/standardization via the Grammar Data Curation framework. The workflow was divided into two main branches: (i) molecular mutation and treatment sensitivity prediction, and (ii) RFS prediction. Model performance was evaluated using AUROC and F1 score for classification tasks, and Kaplan-Meier survival curves with hazard ratios for RFS. Model explainability was assessed through heatmaps of WSIs and identification of top predictive tiles. (B) Summary of clinical dataset composition: proportion of cases receiving adjuvant therapy, tumor location distribution, mutation distribution at the exon level, and mutation distribution at the codon level. C_FIG

1Accurate subtype diagnosis is essential for guiding therapy and predicting patient outcome in malignant mesothelioma. Most computational pathology models are trained on large tissue images from resection specimens, which maximises information for training but limits model relevance in real-world diagnostic settings where small biopsies are the most usual tissue source. In this work, we assembled a large multicentre cohort of HES- and HPS-stained mesothelioma biopsy slides. We used a self-supervised learning model to evaluate the associations of biopsy-driven morphology patterns with histological subtype, molecular markers, and survival. The discovered histomorphology patterns captured a continuum of tissue phenotypes spanning epithelioid, sarcomatoid, and non-tumour morphologies. Also, patient-level HPC representations achieved excellent performance for distinguishing epithelioid from non-epithelioid mesothelioma (AUC = 0.94) and demonstrated predictive value for immunohistochemistry (IHC) markers. Additionally, HPC-derived features alone achieved performance comparable to established clinical and molecular variables (C-index = 0.65), while integration of HPCs with clinical and marker information improved performance to a C-index of 0.69. Several HPCs were significantly associated with favourable or adverse prognosis and reflected known subtype-specific biological patterns. In conclusion, self-supervised learning can discover interpretable histomorphological phenotypes directly from routine mesothelioma biopsies without further training. These AI-derived phenotypes capture clinically and biologically relevant information, linking tissue architecture to molecular characteristics, histological subtypes, and patient outcomes. The proposed framework provides a thorough evaluation of real-world biopsy data using a pre-trained model, without the need for computationally intensive retraining, and addresses the question of whether SSL-based AI can be deployed out of the box in clinical settings.

A key challenge for single cell discovery analysis is to identify new cell types, describe them quantitatively, and seek these novel cells in new studies often using a different platform. Over the last decade, tools were developed to address identification and quantitative description of cells in human tissues and tumors. However, automated validation of populations at the single cell level has struggled due to the cytometry fields reliance on hierarchical, ordered use of features and on platform-specific rules for data processing and analysis. Here we present Velociraptor, a workflow that implements Marker Enrichment Modeling in three cross-platform modules: 1) identification of cells specific to disease states, 2) description of hallmark features for each cell and population, and 3) searching for cells matching one or more hallmark feature sets in a new dataset. A key advance is that Velociraptor registers cells between datasets, including between flow cytometry and quantitative imaging using different, overlapping feature sets. Four datasets were used to challenge Velociraptor and reveal new biological insights. Working at the individual sample level, Velociraptor tracked the abundance of clinically significant glioblastoma brain tumor cell subsets and characterized the cells that predominate in recurrent tumors as a close match for rare, negative prognostic cells originally observed in matched pre-treatment tumors. In patients with inborn errors of immunity, Velociraptor identified genotype-specific cells associated with GATA2 haploinsufficiency. Finally, in cross-platform analysis of immune cells in multiplex imaging of breast cancer, Velociraptor sought and correctly identified memory T cell subsets in tumors. Different phenotypic descriptions generated by algorithms or humans were shown to be effective as search inputs, indicating that cell identity need not be described in terms of per-feature cutoffs or strict hierarchical analyses. Velociraptor thus identifies cells based on hallmark feature sets, such as protein expression signatures, and works effectively with data from multiple sources, including suspension flow cytometry, imaging, and search text based on known or theoretical cell features.

Despite development of several effective therapies for multiple myeloma (MM), the prognosis of patients with partial deletion of chromosome 17 (del(17p)) and TP53 aberrations remains poor. By applying comprehensive multi-omics profiling analyses (whole exome and transcriptome sequencing plus proteomics) and functional ex vivo drug screening to samples from 167 patients with MM, we uncovered novel therapeutic vulnerabilities specific to TP53 mutated MM. Our findings revealed a distinct sensitivity profile to a range of inhibitors (mitotic, topoisomerase, HDAC, HSP90, IGF1R and PI3K/AKT/mTOR inhibitors) irrespective of 17p deletion status. Conversely, no increase in sensitivity was observed for monoallelic TP53 (del(17p) with WT TP53) when compared to WT TP53 samples, highlighting the remaining unmet clinical need. Notably, plicamycin, an RNA synthesis inhibitor linked to modulation of chromatin structure and increased transcription, emerged as particularly efficacious for TP53 mutated MM. The increased sensitivity correlated with higher protein expression of the drug targets: HDAC2, HSP90AA1 and multiple ribosomal subunits. Additionally, we observed increased RNA expression of G2M checkpoint, E2F targets and mTORC1 signaling in our cohort and the MMRF-CoMMpass (NCT01454297) study in TP53 mutated MM. Harmonization of multi-omics data with ex vivo drug screening results revealed that TP53 mutated MM is functionally distinct from MM with monoallelic TP53, and demonstrates that MM with mutated TP53, with and without del(17p), may be targetable by approved drugs. These results further indicate the need for regular monitoring by sequencing to identify these patients. KEY POINTSTP53 mutation in myeloma confers sensitivity to multiple compounds, including approved drugs, irrespective of del(17p) status. TP53 mutated myeloma links to higher expression of drug targets involved in cell proliferation, mRNA processing, and chromatin modulation.

Background: Current risk stratification for Ewing sarcoma relies heavily on clinical factors such as metastatic status, failing to capture histologic heterogeneity as a potential prognostic indicator. Although pathology reports contain rich biological data, this information remains locked in unstructured narrative text, limiting large-scale retrospective analyses. We aimed to validate the utility of Large Language Models (LLMs) for scalable data abstraction and to identify prognostic histologic features from a large multi-institutional cohort. Methods: We conducted a retrospective cohort study using data from six Children's Oncology Group (COG) clinical trials. We utilized an LLM-based pipeline (OpenAI o3) to extract structured variables, including immunohistochemical (IHC) markers and CD99 staining patterns - from digitized, Optical Character Recognition (OCR)-processed pathology reports. Extraction accuracy was validated against a human-annotated ground truth (n=200) and cross-validated against senior experts (n=48). We assessed the association between extracted features and Overall Survival (OS) using Kaplan-Meier analysis and multivariable Cox proportional hazards regression, adjusting for metastatic status. Findings: We analyzed 931 diagnostic pathology reports spanning over 19-years. The LLM achieved a weighted average accuracy of 94% across 17 IHC markers; in a cross-validation subset, the LLM outperformed human annotators (weighted average accuracy over 15 IHC markers: LLM o3: 98.1%, a resident specialist 91.4%, and a senior expert 95.9%). Survival analysis identified Neuron-Specific Enolase (NSE) and S100 as significant prognostic biomarkers. After adjusting for metastatic status, NSE positivity was associated with significantly inferior survival (HR 2.15, 95% CI 1.15 - 4.02, p=0.016); this risk was most pronounced in patients with non-metastatic disease (HR 5.64, p=0.0055). Conversely, S100 positivity was associated with improved survival (HR 0.58, 95% CI 0.34-1.00, p=0.046). Interpretation: LLM-assisted extraction of pathology variables is highly accurate and scalable, capable of unlocking "dark data" from historical clinical trials. We identified NSE as a potent risk factor and S100 as a protective marker in Ewing sarcoma, particularly in localized disease. These findings suggest that AI-derived histologic data can refine risk stratification and, if validated, warrant inclusion in future prospective trials.

The histological heterogeneity of primary tumours across the pan-cancer spectrum poses a formidable barrier to accurate lymph node metastasis assessment, often causing AI systems to make "overconfident errors" on rare variants that lead to missed diagnoses. To address this, we present UPATHLN, a unified diagnostic platform that synergizes a pathology foundation model-based encoder with a decoupled uncertainty estimation mechanism. We developed and validated the system using a large-scale multicentre dataset of 26,229 lymph nodes from 14 distinct primary origins. In internal validation, UPATHLN achieved an area under the curve (AUC) of 0.986. Crucially, the uncertainty module functioned as a decisive fail-safe: by flagging potential false-negative predictions for mandatory pathologist review, it intercepted all missed diagnoses, securing 100% conditional sensitivity across both the development and independent test cohorts--even for tumours from seven unseen primary origins. Concurrently, this mechanism reduced the review burden on negative lymph nodes by 73.2%. Ultimately, UPATHLN sets a new benchmark for safety-critical AI, demonstrating that explicitly modelling uncertainty is key to unlocking reliable, workload-efficient diagnostics at the pan-cancer scale.

Next-generation sequencing (NGS) for the detection of somatic variants has become the method of choice in a variety of molecular oncology fields and in the clinic. Its use ranges from sequencing entire tumor genomes and transcriptomes to targeted clinical diagnostic gene panels. The NYU Langone Genome PACT (Profiling of Actionable Cancer Targets, LG-PACT) assay is a qualitative in vitro diagnostic test that uses targeted next generation sequencing (NGS) of formalin-fixed paraffin-embedded (FFPE) tumor tissue matched with normal specimens from patients to detect gene alterations in a targeted panel covering 606 genes and the TERT promoter. Indications for testing are cancer (solid tumors and hematological malignancies) where a mutational profile from multiple genes would be informative for disease stratification, prognosis, or treatment options including targeted therapies and eligibility for clinical trials. The test is intended to provide information on somatic mutations including point mutations, small insertions/deletions (indels), and copy number aberrations for diagnostic and treatment decisions. LG-PACT is a United States Food and Drug Administration (FDA) cleared diagnostic test (510K: K202304). The clinical interpretation of sequencing data of molecular tumor markers from NGS encompasses automated variant calling tools with human interpretation. This final mostly manual review of data step is intensive, involving highly trained scientists, encompassing literature review, interpretation and clinical tier classification by pathologists, who then provide a complete molecular diagnostic report to the treating oncologists. We provide analysis of 1339 clinical genomic profiles from 31 different cancers and their subtypes, comprising of central nervous system (CNS) 792 (59%) cases (incl. meningioma, glioma and glioblastoma), with 267 (20%) cases predominantly of lung, pancreatic and colorectal and 280 of others (21%). Here, we present the technical challenges of validating an NGS oncological diagnostic targeted assay for clinical grade accuracy and sensitivity for patient care. We show how copy number alterations provide a more comprehensive description of the tumors genomic profile. We then outline the utility of targeted panel sequencing based on certified pathologist selection of reportable variants for our current patient cohort. Where analysis of variant detection has led to 49.4% (661/1339) of our clinical tumor samples containing mutations in known therapy targeted genes, 35.6% (477/1339) with mutation detected in other genes, and 15% (201/1339) cases being negative.

BackgroundAccurate quantification of cancer hallmark activity is essential for understanding tumor progression, tailoring treatments, and improving patient outcomes. Traditional methods, such as histopathological grading and immunohistochemistry for protein expression, often overlook the complex interplay between cancer cells and the tumor microenvironment and provide limited insight into hallmark-specific mechanisms. We aimed to develop OncoMark, a high-throughput deep learning-enabled neural multi-task learning framework capable of systematically quantifying integrative hallmarks activities using transcriptomics data from routine tumor biopsies. MethodsIn this study, we acquired single-cell transcriptomics data from 941 tumor samples across 14 tissue types, comprising nearly 3.1 million cells from 56 studies conducted worldwide, to form a large multicenter dataset. Our model employs a supervised neural multi-task learning method designed to predict multiple cancer hallmarks present in the biopsy samples simultaneously. The OncoMark model was developed and tested on 90% of the studies (patients from 51 studies) using repeated five-fold cross-validation performed twice. For further evaluation, the model was assessed on the remaining 10% of the studies (patients from 5 studies) that were excluded from the initial training and testing dataset. Additionally, we included patients from publicly available datasets, including TCGA, GTEx, ANTE, MET500, POG570, CCLE, TARGET, and PCAWG to validate the models performance. The primary objective was to evaluate the performance of the model in identifying cancer hallmarks in cancer datasets and ensure no hallmark predictions were made in normal samples across the four prespecified groups: (i) internal test set, (ii) external test set, (iii) normal samples (real-world), and (iv) cancer samples (real-world). FindingsOncoMark demonstrated exceptional performance in predicting cancer hallmark states, achieving near-perfect accuracy across internal test data and five external test datasets. Internal testing consistently showed accuracy, precision, recall, and F1 scores exceeding 99%, underscoring the models reliability across hallmarks. External test further confirmed these findings, with accuracy, precision, recall, F1 scores, and balanced accuracy consistently exceeding 96{middle dot}6%, and multiple datasets achieving perfect scores, highlighting the models exceptional generalizability and robustness. Specificity tests using GTEx and ANTE datasets accurately classified normal tissues, while sensitivity analysis on TCGA, MET500, CCLE, TARGET, PCAWG, and POG570 datasets effectively identified cancer hallmarks. InterpretationWe developed an AI-based framework that enables accurate, efficient, and cost-effective quantification of cancer hallmark activity directly from transcriptomics data. The framework demonstrated significant potential as an assistive tool for guiding personalized treatment strategies and advancing the clinical management of cancer patients. FundingAshoka University, S.N. Bose National Centre for Basic Sciences, Mphasis F1 Foundation, DST SERB Core Research Grant. Research in ContextO_ST_ABSEvidence before this studyC_ST_ABSWe conducted an extensive literature search using Google Scholar and PubMed without language restrictions, employing search terms such as "(Predicting OR Classifying OR Annotating) and (cancer hallmarks) AND (Deep OR Machine Learning) OR (Artificial Intelligence OR AI)." While there have been advancements in molecular oncology and computational methodologies over the two decades since the concept of cancer hallmarks was first introduced, a comprehensive machine learning or deep learning framework to annotate all cancer hallmarks simultaneously from tumor biopsy samples remains to be developed. Additionally, the scarcity of hallmark-annotated datasets has posed a significant challenge, hindering the development of robust predictive models. Added value of this studyThis study introduces OncoMark, a novel high-throughput neural multi-task learning (N-MTL) framework designed to predict all cancer hallmark activities simultaneously from biopsy samples. OncoMark addresses the lack of annotated hallmark-specific data by generating synthetic biopsy (pseudo-bulk) datasets annotated with hallmark activity, meticulously modeled to reflect real-world tumor biology while maintaining clinical relevance. The framework employs a multi-task learning approach to capture interdependencies among hallmarks, advancing beyond isolated predictions to offer a holistic view of tumor biology. Validation on five independent datasets comprising 95 patient samples demonstrated its generalizability and reproducibility. Further external validation using eight datasets, encompassing over 11,679 cancer and 8348 normal patient samples, reinforced its robustness. To promote clinical integration, a user-friendly web-based tool was developed, enabling seamless access for oncologists and researchers. Implications of all the available evidenceThe OncoMark framework represents a transformative advancement in cancer diagnostics and treatment planning. By enabling accurate and reproducible prediction of all hallmark activities simultaneously from biopsy samples, this model paves the way for precision oncology at scale. Its ability to systematically capture hallmark interdependencies provides deeper insights into tumor behavior, guiding the development of individualized targeted therapies. The incorporation of a web-based interface ensures the accessibility of this innovation to clinicians worldwide, bridging the gap between computational oncology and clinical practice. Following further validation and integration into healthcare workflows, OncoMark has the potential to improve cancer outcomes by delivering timely, cost-effective, and precise tumor analyses, facilitating informed therapeutic decision-making with unparalleled precision. Cancer progression is driven by a set of well-defined biological principles--collectively termed the "hallmarks of cancer"--yet current diagnostic approaches seldom incorporate these distinct molecular features into clinical practice. Despite substantial progress in molecular oncology, traditional methods like histopathological grading and immunohistochemical assays often fail to capture the complex interplay between cancer cells and the tumor microenvironment, emphasizing the need for robust computational frameworks capable of systematically quantifying hallmark-specific activity. Here, we address this gap by developing OncoMark, a high-throughput neural multi-task learning (N-MTL) framework designed to simultaneously quantify hallmark activities in tumor biopsies using transcriptomics data. We show that OncoMark achieves near-perfect accuracy, precision, recall, and F1 scores (>99%) in cross-validation, with external validation consistently exceeding 96.6% on five independent datasets. Further evaluation on eight additional datasets--including large-scale cancer cohorts (TCGA, MET500, CCLE, TARGET, PCAWG, POG570) and normal tissue datasets (GTEx, ANTE)--demonstrated high specificity for normal samples and robust sensitivity for hallmark prediction in cancer. By delivering a comprehensive and cost-effective molecular portrait of tumor biology and providing a user-friendly web platform accessible at https://oncomark-ai.hf.space/, OncoMark has the potential to guide tailored treatment strategies and advance precision oncology. More broadly, this framework signifies a transformative step toward routine hallmark-based diagnostics, promising to improve patient outcomes by facilitating timely and precise tumor profiling.