Browse Articles

Discover research articles across all indexed journals

Isolation and preliminary characterization of extracellular vesicles from bottlenose dolphin (Tursiops truncatus) and long-finned pilot whale (Globicephala melas) blow

PLoS ONE Valentina Moccia, Cinzia Centelleghe, Andrea Zendrini et al. Jul 01, 2026 DOI: 10.1371/journal.pone.0352853

Cetaceans are key sentinel species for environmental health monitoring. Although sampling from free-ranging animals is challenging, the analysis of cetacean blow offers a minimally invasive approach to assess their health status. Extracellular vesicles (EVs) are cell-derived nanostructures present in biological fluids and widely studied as disease biomarkers in humans. Despite the potential for similar uses, EVs have not been studied in cetacean blow to date. This proof-of-concept study aims to assess the feasibility of the isolation and characterization of EVs from blow samples collected from five bottlenose dolphins ( Tursiops truncatus ) kept under human care and from a free ranging specimen of long-finned pilot whale ( Globicephala melas ). EVs were purified from bottlenose dolphin blows by ultracentrifugation (UC) or size exclusion chromatography (SEC) and from the long-finned pilot whale by SEC. Particle concentration and size distribution were assessed by Nanoparticle Tracking Analysis (NTA), morphology by Air-atomic force microscopy (AFM) and protein expression by Western Blotting (WB). NTA revealed a higher mean particle concentration in bottlenose dolphin EVs isolated by UC compared to SEC, while EVs isolated from the long-finned pilot whale presented a lower particle concentration. AFM confirmed the presence of EV-like particles within the typical EV size range in bottlenose dolphin’s EVs obtained both by SEC or UC. All EV samples were positive for CD9 and integrin-β and negative to Calnexin. SEC was more sensitive to detect OmpA, a membrane protein of Gram-negative bacteria, in EVs from both species. Our pilot study demonstrates that EV-like particles are present in cetacean blow and can be isolated and characterized. Future investigations focused on characterizing and quantifying a wider array of EV associated molecules may further the application of blow EV analysis for cetacean health assessments.

Cytoskeleton‐Inspired Mechanically Interlocked Catenane Framework Enabling Robust yet Dynamic Polymer Networks

Advanced Materials Yuhang Liu, Wenbin Wang, Yudong Chen et al. Jul 01, 2026 DOI: 10.1002/adma.73742

ABSTRACT The red blood cell cytoskeleton, featuring a highly ordered yet dynamically reconfigurable skeletal topology, offers an important structural inspiration for designing high‐performance polymeric materials that integrate mechanical robustness, structural stability, and dynamic adaptability. Inspired by this topology and functionality, we report a class of dynamic polymer networks featuring a continuous mechanically interlocked catenane framework ( CF MIN). The catenane framework is constructed via a sequential self‐assembly strategy, involving metal‐coordination‐driven formation of a supramolecular polyhexagonal network, followed by host–guest complexation with linear poly(crown ether). The resulting architecture integrates densely distributed mechanical bonds with a continuous rigid skeleton, thereby enabling simultaneous dynamic adaptability and structural stability. Compared with its structurally analogous but non‐interlocked control, CF MIN demonstrates significantly improved stiffness, strength, and toughness. These enhancements arise from hierarchical force‐triggered dynamic processes within catenane framework, involving host–guest dissociation, ring sliding, and metal–ligand bond rupture, together facilitating efficient energy dissipation. Meanwhile, topological constraints of catenane framework, along with robust metal coordination, impart superior structural stability to CF MIN, allowing network to retain integrity at temperatures up to 180°C. This work demonstrates that integrating mechanical bonds into ordered skeletal architectures offers a promising strategy for designing robust, dynamically adaptive, and thermally stable polymeric materials.

An Extravascular Synergistic Cocktail Therapeutic Strategy Based on Lanifibranor Loaded Self‐Healing Bioelastomer for Abdominal Aortic Aneurysms

Advanced Materials Jiateng Hu, Wufei Dai, Zhijue Xu et al. Jul 01, 2026 DOI: 10.1002/adma.202511786

ABSTRACT Abdominal aortic aneurysm (AAA) is a fatal cardiovascular disease with hidden onset and high risk of rupture death, which is a major public health concern worldwide. Notably, aortic adventitia is the earliest lesions and the last barrier to rupture, and adventitial fibroblasts (AFs) are the key cellular components of aortic adventitia. Current treatments mainly focus on surgical resection with prosthetic graft replacement and endovascular stent grafting, yet failed to reconstruct the adventitia and abdominal aorta's natural structure effectively. In this study, we proposed an extravascular synergistic cocktail (ESC) therapeutic strategy to treat AAA by incorporating mechanical supporting and vascular remodeling inhibition based on a self‐healing bioelastomer with antifibrotic drug of lanifibranor. The single‐cell RNA‐sequencing revealed potential myogenic transformation of AFs in AAA with the release of chemokines like monocyte chemoattractant protein‐1 (MCP‐1) . The underlying mechanism of lanifibranor in maintaining the phenotype of AFs via promoting lipid metabolism was disclosed. The bioelastomer with multiple reversible dynamic bonds exhibits remarkable arterial‐like mechanical properties and ultra‐fast self‐healing ability under blood condition, and enables rapid in situ AAA wrapping treatment. In both the rat and dog AAA models, it has been fully demonstrated that the ESC therapy can provide effective mechanical support to prevent vascular dilation while continuously releasing lanifibranor to regulate myogenic transformation of AFs to inhibit the development of AAA. Additionally, we investigated the feasibility of minimally invasive laparoscopic implantation in bama pigs, emphasizing its potential for clinical translation and application.

Leucyl-tRNA synthetase as a molecular target of isobutanol-mediated growth inhibition in Saccharomyces cerevisiae

Journal of Biological Chemistry Mano Hasegawa, Nodoka Oshimura, Kaho Hitomi et al. Jul 01, 2026 DOI: 10.1016/j.jbc.2026.113228

Lactate metabolic reprogramming based on LOX/LDHi-Cu NP to enhance copper death in hepatocellular carcinoma cells.

Journal of Clinical Oncology Jiashuo Li, Shasha Wang Jul 01, 2026 DOI: 10.1200/jco.2026.44.19_suppl.297

297 Background: Cuproptosis—copper-triggered mitochondrial proteotoxicity—has emerged as a promising anti-HCC modality, yet intratumoral lactate accumulation restrains TCA flux and suppresses this death pathway. We therefore engineered a lactate-oxidase/high-density-dopamine–copper nanoplatform (LOX/LDHi-Cu NP) that simultaneously depletes lactate and delivers Cu²⁺ to reprogram tumor metabolism and potentiate cuproptosis. Methods: LOX/LDHi-Cu NP was synthesized via one-step dopamine polymerization with Cu²⁺ chelation; size, zeta potential, and Cu-release kinetics were characterized. In vitro, HepG2 and Huh-7 cells were treated with NPs ± lactate supplementation; lactate, ATP, ROS, mitochondrial membrane potential, and copper-dependent protein aggregation were quantified. Cuproptosis was verified by propidium-iodide staining and rescue with TTM. In vivo, orthotopic HCC-bearing mice received intravenous NPs; tumor lactate, Cu accumulation, ICD markers, and growth inhibition were assessed. Results: LOX/LDHi-Cu NP exhibited uniform nanostructure and acid-responsive copper release; the platform efficiently scavenged lactate, reinstated oxidative phosphorylation, and enabled robust copper-dependent DLAT aggregation, ROS burst, and cuproptotic death in vitro. In orthotopic HCC mice, the nanoplatform selectively accumulated within tumors, markedly lowered intratumoral lactate, evoked immunogenic cell death, and achieved significant tumor suppression without systemic toxicity, collectively demonstrating that lactate-metabolic reprogramming sensitizes hepatocellular carcinoma to cuproptosis. Conclusions: Metabolic reprogramming via LOX/LDHi-Cu NP effectively overcomes lactate-mediated cuproptosis resistance, offering a translatable strategy for potentiating copper-based HCC therapy.

Real-time breath volatile organic compound analysis to detect newly diagnosed lung cancer.

Journal of Clinical Oncology Prabhpreet Kaur, Kumaran Kavyadharshini, Jason Ng et al. Jul 01, 2026 DOI: 10.1200/jco.2026.44.19_suppl.268

268 Background: Non-small cell lung cancer (NSCLC) is highly lethal as most cases are late stage at diagnosis. Screening efforts with computed tomography aimed at reducing NSCLC mortality is limited to specific sub-populations with limited implementation globally. Thus, rapid scalable screening tools to detect NSCLC continue to be an unmet need. Detection and analysis of volatile organic compounds (VOCs) from exhaled breath with proton transfer reaction-time-of-flight mass spectrometry (PTR-MS) is real-time, non-invasive, and highly sensitive. With the hypothesis that PTR-MS analytics of exhaled VOC can discriminate between patients with lung cancer and healthy volunteers, we performed a pilot study to identify and define a unique exhaled-breath VOC signature representative of newly diagnosed, treatment-naïve NSCLC. Methods: A single-centre, prospective, pilot study was conducted. Treatment-naïve patients with newly diagnosed, histologically confirmed NSCLC of any stage were recruited and the control group comprised age, sex, and smoking history matched healthy volunteers with no known malignancy. Exhaled-breath VOC profiles were analysed with PTR-MS testing. A proprietary machine learning algorithm was employed to analyse breath samples to identify a VOC signature that could distinguish between patients with newly diagnosed, treatment naïve NSCLC and healthy controls. Results: 185 newly diagnosed treatment naïve lung cancer patients were recruited over 18 months with over 500 distinct VOC measured in each breath. Amongst these VOC, 20 were identified that differed significantly in concentration within breath of patients with lung cancer compared to breath of healthy controls. To make a 20 VOC multiplex predictive biomarker for lung cancer, a random forest model with 10-fold cross validation was adopted to build the classifier. The optimised VOC signature distinguished patients with newly diagnosed treatment naïve NSCLC from healthy controls with an area under the curve of 0.9 and 0.87 in the training and validation sets respectively with corresponding sensitivity and specificity of 93% and 86% in the training set, and 89% and 85% in the validation set. Conclusions: This study represents the largest known cohort of newly diagnosed, treatment naïve NSCLC patients with exhaled-breath VOC samples. A high-performance VOC signature unique to newly diagnosed NSCLC was identified. The high sensitivity and specificity demonstrated provides a strong signal for the feasibility of breath-based screening for NSCLC. Further large-scale external validation is ongoing to test the utility of exhaled-breath VOC as a real-time, point-of-care non-invasive screening tool for NSCLC.

Few-atom-thick silver films for enhanced nanoscale nonlinear optics

Nature Communications Philipp K. Jenke, Saad Abdullah, Andrew P. Weber et al. Jul 01, 2026 DOI: 10.1038/s41467-026-74804-4

Abstract The inherently weak nonlinear optical response of bulk materials remains a fundamental limitation in advancing photonic technologies. Nanophotonics addresses this challenge by tailoring the size and morphology of nanostructures to manipulate the optical near field, thus modulating the nonlinear response. Here, we explore a complementary strategy based on engineering the electronic band structure in the mesoscopic regime to enhance optical nonlinearities. Specifically, we demonstrate an increase in second-harmonic generation (SHG) from crystalline silver films as their thickness is reduced down to just a few atomic monolayers. Operating at the boundary between bulk and two-dimensional systems, these ultra-thin films exhibit a pronounced enhancement of SHG with decreasing thickness. This enhancement stems from quantum confinement effects that modify the interaction between electronic states and incident light, which we explain based on quantum-mechanical calculation. Our atomically-thin crystalline silver films provide a new means to overcome the small interaction volumes inherent to nanophotonic platforms, enabling efficient nanoscale nonlinear optics with potential applications in photonics, sensing, and quantum technologies.

Collaborative Pt–Ni Sites Enable CO-Tolerant Hydrocracking for Upcycling Polyolefin Waste

Journal of the American Chemical Society Lu Sun, Xiaomei Wang, Jing Xu et al. Jul 01, 2026 DOI: 10.1021/jacs.6c00426

Evaluation of early trabecular changes around implants using fractal analysis and panoramic indices

PLoS ONE Mert Keles, Ibrahim Burak Yuksel, Dilek Ozkan Sen et al. Jul 01, 2026 DOI: 10.1371/journal.pone.0352984

Early trabecular changes around dental implants may not be adequately captured by conventional panoramic morphometric indices. Fractal analysis has been proposed as a quantitative method for detecting subtle changes in bone microarchitecture. This study aimed to evaluate early peri-implant trabecular structural changes using fractal analysis and panoramic morphometric indices during the unloaded healing period. This retrospective study included 60 single-tooth dental implants. Standardized panoramic radiographs were obtained at baseline and 3 months postoperatively. Mandibular cortical width, panoramic mandibular index, and fractal dimension values were measured in mesial, distal, and apical regions using ImageJ software. As all implants healed submerged and unloaded, periodontal parameters—plaque index, gingival index, bleeding on probing, and probing pocket depth—were recorded from natural teeth in the same quadrant. Baseline and 3-month values were compared using paired t-tests, with statistical significance set at p < 0.05. Fractal dimension values increased significantly in all evaluated regions at 3 months compared with baseline (p < 0.05), indicating measurable early trabecular structural changes. In contrast, mandibular cortical width and panoramic mandibular index showed no significant changes. Among periodontal parameters, only gingival index demonstrated a significant improvement (p < 0.001), reflecting reduced gingival inflammation. Fractal analysis detected early trabecular changes during the 3-month unloaded healing period, whereas conventional panoramic morphometric indices showed no significant change. These findings should be interpreted as indicators of trabecular reorganization rather than direct evidence of osseointegration. Fractal analysis may provide supplementary, non-invasive information during early radiographic follow-up.

Bio‐Inspired Full‐Spectrum Bidirectional Optoelectronic Synaptic Transistor for In‐Sensor Neuromorphic Vision

Advanced Materials Meiqiu Dong, Yu Zhang, Mingsheng Luo et al. Jul 01, 2026 DOI: 10.1002/adma.73698

ABSTRACT Inspired by avian visual systems, wide‐spectrum bidirectional optoelectronic synaptic transistors (OSTs) have emerged as a transformative platform for neuromorphic vision processing. However, current artificial synaptic technologies remain constrained by fragmented control paradigms that rely on wavelength‐selective or electrical polarity‐dependent modulation, restricting their capacity for complex visual information processing. In this study, an organic 2D molecular crystal (2DMC) heterojunction is developed to achieve high‐performance full‐spectrum bidirectional OSTs. The high‐quality 2DMC heterostructure facilitates efficient exciton separation and diffusion, yielding remarkable ambipolar characteristics with an exceptional responsivity of 6.3 × 10 4 A W −1 under weak light. The bipolar heterostructure transistor design enables dynamic reconfiguration between excitatory and inhibitory synaptic behaviors through dual modulation mechanisms of spectral tuning and gate polarity control, successfully emulating bidirectional neuroplasticity features. The practical viability of this technology is demonstrated through implementation in a broad‐spectrum intelligent‐vehicle‐mounted vision system that achieves a classification accuracy of >90% for traffic objects across complex scenarios. This work not only provides an effective strategy for developing high‐performance bidirectional optoelectronic synapses but also paves the way for advanced neuromorphic computing and intelligent machine vision technologies.

Impaired cortical actin dynamics via lanosterol-dependent HMG-CoA reductase downregulation mediates IFN-α-induced mast cell stabilization

Journal of Biological Chemistry Rema Naskar, Liu Ye, Sanu Karan et al. Jul 01, 2026 DOI: 10.1016/j.jbc.2026.113161

Adebrelimab plus TACE and/or HAIC for primary liver cancer: A retrospective study.

Journal of Clinical Oncology Ruipeng Zheng, Jiarui Li, Haifeng Zhang Jul 01, 2026 DOI: 10.1200/jco.2026.44.19_suppl.296

296 Background: This retrospective analysis reports on the use of adebrelimab—a novel PD-L1 inhibitor—in combination with transarterial chemoembolisation (TACE) and/or hepatic arterial infusion chemotherapy (HAIC) for the treatment of primary liver cancer. Methods: The study was approved by the Ethics Committee of the First Hospital of Jilin University and conducted within the interventional department. Inclusion criteria were: (1) pathological or clinical diagnosis of primary liver cancer; (2) at least one cycle of adebrelimab plus TACE and/or HAIC; (3) complete medical records and patient cooperation for follow-up. Cases were enrolled between March 2023 and January 2026. The primary endpoint was progression-free survival (PFS); secondary endpoints included overall response rate (ORR), disease control rate (DCR), overall survival (OS), and safety. Analyses were performed using R software (version 4.5.2). Results: Seventy-two patients were included. The cohort was predominantly male (83.3%), aged < 65 years (75.0%), with hepatitis B virus infection in 55.6%. Most patients had BCLC stage B disease (66.7%) and Child-Pugh class B liver function (59.7%); portal-vein tumour thrombus was present in 88.9%, and 19.4% had extrahepatic metastases. Median PFS was 7.1 months (95% CI 5.1–9.9 months). Median OS was 17.9 months (95% CI 10.8–not reached). ORR was 65.3% (47/72) and DCR was 70.8% (51/72). The most frequent any-grade adverse events (AEs) associated with interventional therapy (≥10%) were fever, hypoalbuminemia, and abnormal liver-function tests. Any-grade AEs attributed to adebrelimab included elevated bilirubin, increased creatinine, decreased white-blood-cell count, and fatigue. Grade ≥3 interventional-therapy-related AEs comprised elevated ALT (5.6%), elevated AST (15.3%), and raised blood bilirubin (2.8%). Grade ≥3 adebrelimab-related AEs were limited to decreased white-blood-cell count (1.4%). No fatal AEs were reported. Conclusions: The combination of adebrelimab with TACE and/or HAIC showed encouraging efficacy and a manageable safety profile in patients with primary liver cancer. Prospective clinical trials are warranted to further evaluate this regimen.

Performance of the hybrid tumor-informed and tumor-agnostic ctDNA testing to predict breast cancer recurrence and progression: A real-world study in Southeast Asia.

Journal of Clinical Oncology Van-Anh Hoang Nguyen, Ngoc Nguyen, Tu Nguyen et al. Jul 01, 2026 DOI: 10.1200/jco.2026.44.19_suppl.42

42 Background: ctDNA is a novel biomarker to monitor treatment response and predict cancer recurrence. Tumor-informed ctDNA assays require high-quality tumor tissues which are unobtainable in many contexts. We developed a hybrid tumor-informed and tumor-agnostic approach to overcome this constraint and evaluate its performance in real-world samples. Methods: We retrospectively analyzed 262 real-world blood samples of 180 patients diagnosed with stage I-IV breast cancer (BC). ctDNA was detected by tumor-informed personalized mutations, combined with a tumor-agnostic BC-specific hotspot panel (K-TRACK, Gene Solutions). The longitudinal ctDNA status was then compared with clinical outcomes documented for 69 early-stage and 17 metastatic-stage patients. Results: The overall pre-treatment ctDNA detection rates were 56.1% and 70.9% for early- and metastatic- stages respectively, and highest in HER2+ and triple negative BC. Compared to the tumor-informed approach, the hybrid approach improved ctDNA detection rate by 20.4% for cases with suboptimal FFPE quality. In the early stage, post-operative ctDNA was detected in 80.0% (8/10) of the patients having recurrence, with lead time up to 11.0 months; while 98.3% (58/59) of those with no recurrence had negative results. ctDNA positivity was an independent prognostic factor (p<0.001) that increased risk of recurrence (HR=147.8, 95% CI: 26.3-831.3). In the metastatic stage, at 12 weeks after therapy initiation, 85.7% (6/7) of the molecular responders, who had ctDNA cleared or reduced by >50% of baseline VAF level, achieved clinical partial response or stable disease; while 70.0% (7/10) of the molecular non-responders had confirmed progressive disease. In a subgroup of metastatic ER+ patients, tumor-agnostic hotspot panel identified new resistance mutations including ESR1 Y537N (7.1%), ESR1 Y538G (3.6%), ESR1 P535H (1.8%). Conclusions: Our hybrid tumor-informed and tumor-agnostic ctDNA assay improved ctDNA detection rate, and predicted cancer recurrence and progression with high accuracy in real-world samples.

A carboxylate switch point controls long-range energy transduction in respiratory Complex I

Nature Communications Adel Beghiah, Patricia Saura, Terezia Kovalova et al. Jul 01, 2026 DOI: 10.1038/s41467-026-74767-6

Abstract Complex I is a highly intricate membrane-bound protein complex that powers the cellular energy metabolism by a long-range ( > 300 Å) proton-coupled electron transfer (PCET) reaction. Here, we investigate the highly debated coupling mechanism of Complex I by probing the charge transfer reaction along its functionally central carboxylate pathway (E-channel). By combining biophysical and site-directed mutagenesis experiments with high-resolution (2.6-2.8 Å) cryo-electron microscopy (cryo-EM) and multiscale simulations, we identify a conserved carboxylate switch point (D79 NuoA ) that mediates proton transfer by establishing a kinetic gate and couples the redox chemistry to proton pumping. We find that mutation of the identified site, as found in patients suffering from severe neurodegenerative disorders, drastically perturbs the charge transfer mechanism, and results in a 20% PCET activity. Our combined findings illustrate mechanistic principles of molecular gates underlying long-range charge transfer reactions, and show how disease mutations perturb the function of conserved switch points in energy transduction.

Unified <i>N</i> -Sulfonylated Urea Catalyzed Enantioselective Dimerization, Arylation, and Amination of Tryptamines: Total Synthesis of Five Pyrroloindoline Alkaloids

Journal of the American Chemical Society Jin-Rui Tian, Chun-Yu Mi, Ka Lu et al. Jul 01, 2026 DOI: 10.1021/jacs.6c04688

Developing a natural language processing system using transformer-based models for adverse drug event detection in electronic health records

PLoS ONE Jingyuan Wu, Xiaodi Ruan, Elizabeth McNeer et al. Jul 01, 2026 DOI: 10.1371/journal.pone.0350516

Objective To develop a transformer-based natural language processing (NLP) system for detecting adverse drug events (ADEs) from clinical notes in electronic health records (EHRs). Materials and Methods We fine-tuned BERT Short-Formers and Clinical-Longformer using the processed dataset from the 2018 National NLP Clinical Challenges (n2c2) shared task Track 2. Two data processing methods, window-based and split-based approaches, were compared to identify the optimal processing method. Model generalizability was evaluated on a dataset extracted from Vanderbilt University Medical Center (VUMC) EHRs. Results On the n2c2 dataset, the best 5-fold cross-validation AUPRC, micro F1, and macro F1 scores were 0.840 (Clinical-Longformer, 4-chunk split), 0.965 (BioBERT, 15-word window), and 0.852 (Clinical-Longformer, 10-chunk split). On the VUMC dataset, the best AUPRC, micro F1, and macro F1 scores were 0.536 (Clinical-Longformer, 6-chunk split), 0.966 (BERT-base-uncased, 6-chunk split), and 0.762 (Clinical-Longformer, 4-chunk split). Discussion Transformer-based models demonstrated strong performance for ADE detection, with split-based processing generally outperforming window-based methods. Clinical-Longformer combined with a practical split-based approach showed promise for real-world implementation. Beyond token limits, chunk size substantially influenced model performance, even when text length remained within limits. Conclusion Our findings provide practical guidance for developing transformer-based ADE detection systems from clinical notes. The selection of both text preprocessing strategies and model architectures should be guided by note characteristics and practical considerations such as annotation burden.

Shear‐Stiffening Damping Ionogels Enabled by the Synergy of Dynamic Bonds and Steric Hindrance

Advanced Materials Shilong Zhang, Junjie Yu, Lingling Li et al. Jul 01, 2026 DOI: 10.1002/adma.73746

ABSTRACT High‐rate impact loading causes structural failure and insufficient energy dissipation in protective materials. The intrinsic stiffness–damping trade‐off in polymer materials makes it highly challenging to simultaneously achieve strong damping and broad frequency energy dissipation. Here, we developed tough ionogels with enhanced shear‐stiffening and damping capabilities through the synergy of dynamic bonds and steric hindrance of hyperbranched polymeric ionic liquids (HPILs). The synergistic coupling of dynamic‐bond dissociation in the high‐frequency regime and the topological constraint of viscous‐flow HPILs in the low‐frequency regime endows the ionogels with broad‐frequency damping and pronounced shear‐stiffening behavior. The prepared ionogel shows an 842‐fold shear‐stiffening response and high damping (tan δ &gt; 1) across a wide frequency range (10 − 3 –10 5 rad s − 1 ). At a high impact rate of 4000 s − 1 , it also exhibits high impact strength (248.6 MPa) and toughness (86.5 MJ m − 3 ). This strategy provides a reference for the design of next‐generation high‐performance damping ionogels.

Clinically observed RASA1 missense mutants exhibit diverse RasGAP protein behaviors in vitro

Journal of Biological Chemistry Maxum E. Paul, Rediet B. Delelegne, Jocelyn E. Chau et al. Jul 01, 2026 DOI: 10.1016/j.jbc.2026.113201

Automated tumor content analysis from whole slide images using deep learning models for NGS quality control in precision oncology.

Journal of Clinical Oncology Satya Prakash Khuntia, Hitesh Goswami, Kshitij Rishi et al. Jul 01, 2026 DOI: 10.1200/jco.2026.44.19_suppl.23

23 Background: Accurate tumor cell abundance (TCA) estimation is critical for ensuring sample adequacy prior to next-generation sequencing (NGS) in precision oncology workflows. Current standard practice requires manual pathologist review of hematoxylin and eosin stained whole slide images (WSI), creating throughput bottlenecks and inter-observer variability. We developed and evaluated a series of lightweight CNN models for cancer-agnostic TCA estimation to automate NGS quality control. Methods: Over 1,000 WSIs spanning multiple tumor types were processed using Otsu-based tissue segmentation. Tile-level pseudo-labels were generated using an ensemble of CellViT models, with slide-level TCA computed as the minimum of area-based and cell-based estimates across the ensemble. All slide-level TCA labels were independently validated by board-certified pathologists, with discordant cases reviewed and reconciled prior to model training. Four CNN architectures were evaluated: MobileNetV3, GoogLeNet, ResNet, and DenseNet. Models were trained via knowledge distillation from the CellViT ensemble and validated against pathologist-confirmed TCA estimates. Performance was assessed using MAE, AUC, and clinical concordance at a 20% TCA threshold for NGS adequacy determination. Results: All four models achieved AUC greater than 0.91 for NGS adequacy classification at the 20% TCA threshold, with MAE under 7% against pathologist-confirmed reference standards across all tumor types evaluated. Higher capacity architectures demonstrated marginal accuracy gains while MobileNetV3 achieved comparable performance at significantly faster inference speed, processing a complete WSI in under 5 seconds. Conclusions: Lightweight CNN models trained via knowledge distillation from a CellViT ensemble and pathologist-confirmed labels can achieve clinically acceptable cancer-agnostic TCA estimation suitable for automated NGS QC workflows. Compared to large Vision Transformer models requiring substantial computational resources, the distilled CNN models offer dramatically reduced inference cost while preserving clinical-grade accuracy. These models have potential to substantially reduce pathologist workload in molecular pathology laboratories while maintaining accuracy required for NGS adequacy determination, with direct applicability to precision oncology treatment selection pipelines and broad accessibility across resource-varied clinical settings.

Machine learning-based ICU mortality prediction across hematologic malignancy subtypes: A comparative analysis using MIMIC-IV.

Journal of Clinical Oncology Shankar Biswas, Yashasvi Srivastava, Ayman Hamadttu Jul 01, 2026 DOI: 10.1200/jco.2026.44.19_suppl.262

262 Background: Patients with hematologic malignancies face high ICU mortality, yet prediction models largely treat these diseases as a homogeneous group. Whether risk profiles differ across subtypes is poorly understood, particularly for multiple myeloma (MM) where ICU-specific prediction tools are absent. We developed machine learning (ML) models to predict in-hospital mortality and compared disease-specific predictive features across non-Hodgkin lymphoma (NHL), MM, and acute myeloid leukemia (AML). Methods: Using MIMIC-IV v3.1 (2008-2022), we identified 1,372 adult ICU patients (NHL=742, MM=330, AML=298). A CatBoost model was trained on 134 features extracted from the first 24 hours (demographics, labs, vitals, SOFA score, treatments) with temporal validation (train: 2008-2016, n=1,019; test: 2017-2022, n=353). Combined and subtype-specific models were developed. SHAP analysis identified differential predictors. Patients were stratified into low, intermediate, and high risk groups. Results: Overall in-hospital mortality was 24.0%. AML had the highest mortality across all endpoints (in-hospital 40.3% vs NHL 19.7% vs MM 19.1%; 90-day 52.0% vs 33.2% vs 32.4%). The combined model achieved AUROC 0.83 on temporal validation. The MM-specific model outperformed the combined model (AUROC 0.92 vs 0.89), while the AML-specific model underperformed (0.72 vs 0.77), suggesting myeloma harbors distinct predictive signals diluted in mixed cohorts. Risk groups showed observed mortality of 7.9% (low, n=189), 31.6% (intermediate, n=79), and 60.0% (high, n=85). BUN/creatinine ratio and respiratory rate were top predictors across all subtypes; platelet parameters were disproportionately important in AML, while SpO2 carried greater weight in MM. Model performance remained stable in sensitivity analyses excluding transplant patients (0.81), early deaths (0.80), and across age subgroups (Table 1). Conclusions: An ML model using first-24-hour ICU data predicts in-hospital mortality across hematologic malignancy subtypes (AUROC 0.83). Disease-specific models reveal that MM benefits most from tailored prediction, while AML benefits from shared cross-disease features. These findings support subtype-aware risk stratification in critically ill hematologic malignancy patients. Model/Analysis N (Test) AUROC Combined (All subtypes) 353 0.83 NHL - Combined model 191 0.80 NHL - Specific model 191 0.80 MM - Combined model 80 0.89 MM - Specific model 80 0.92 AML - Combined model 82 0.77 AML - Specific model 82 0.72 Sensitivity: Excl. SCT 346 0.81 Sensitivity: Excl. early deaths 340 0.80 Subgroup: Age &lt; 65 92 0.84