Citations

Abstract With the rapid advancements in large language model technology and the emergence of bioinformatics‐specific language models (BioLMs), there is a growing need for a comprehensive analysis of the current landscape, computational characteristics, and diverse applications. This survey aims to address this need by providing a thorough review of BioLMs, focusing on their evolution, classification, and distinguishing features, alongside a detailed examination of training methodologies, datasets, and evaluation frameworks. We explore the wide‐ranging applications of BioLMs in critical areas such as disease diagnosis, drug discovery, and vaccine development, highlighting their impact and transformative potential in bioinformatics. We identify key challenges and limitations inherent in BioLMs, including data privacy and security concerns, interpretability issues, biases in training data and model outputs, and domain adaptation complexities. Finally, we highlight emerging trends and future directions, offering valuable insights to guide researchers and clinicians toward advancing BioLMs for increasingly sophisticated biological and clinical applications.

Abstract Background Delirium is often underdiagnosed in clinical practice and is not routinely coded for billing. While manual chart review can identify delirium, it is labor-intensive and impractical for large-scale studies. Natural language processing (NLP) can analyze unstructured text in electronic health records (EHRs) to extract meaningful clinical information. Methods To support national integration of NLP for EHR-based delirium identification across different institutions, we launched the Delirium Interest Group within the national Evolve to Next-Gen Accrual to Clinical Trials (ENACT) NLP Working Group. This paper outlines our initial efforts to standardize, evaluate, and translate an NLP-based delirium detection model into the i2b2/ENACT platform. Results Multisite contextual inquiry identified several key challenges, including variations in local screening practices (eg, tools used, documentation frequency, and quality control), the need for harmonized definitions in the context of EHRs, and the complexity of modeling temporal logic. Multisite NLP evaluation revealed variable performance degradation driven by differences in delirium screening practices, clinical documentation patterns and semantics, and note syntactic structures. Conclusion Our work represents an important first step toward enabling scalable and standardized NLP-based delirium detection across institutions. By engaging diverse institutions through the ENACT NLP Working Group, we identified shared challenges and site-specific variations that impact model implementation and performance. Our collaborative approach enabled the development of a more robust framework for delirium identification across heterogeneous EHR systems. Future efforts will build on this foundation to enhance the validity, usability, and translational impact of delirium detection.

PURPOSE Given the extended survivorship of young adults with breast cancer, characterizing health care utilization and adherence to surveillance guidelines can inform gaps in care. PATIENTS AND METHODS Patients age 40 years and younger at diagnosis of stage 0-III breast cancer enrolled in a prospective cohort study completed a survey 7 years after diagnosis assessing receipt of care and surveillance. We evaluated factors associated with transition to a primary care provider (PCP) for cancer-related follow-up and adherence to mammography and bone density scans. RESULTS Among 660 women, 92% reported having a PCP, although for cancer-related follow-up, 57% saw only an oncology provider, 29% saw both a PCP and oncology provider. In the past 2 years, 68% reported getting a flu shot, 76% a pap test/pelvic examination, 56% cholesterol screening, and 42% (n = 232 endocrine therapy [ET] users) a bone density scan. Among women who had a unilateral mastectomy or lumpectomy (n = 356), 82% reported a mammogram in the past year. In multivariate analyses, those with stage 0 ( v I disease) were more likely to see a PCP for follow-up ( v PCP + oncology provider); ET users and those more concerned about recurrence were less likely. Patients with human epidermal growth factor receptor 2-positive (HER2) disease were more likely to adhere to mammography; those with a BRCA mutation and those with a unilateral mastectomy were less likely. Patients with a BRCA mutation and who had chemotherapy were more likely to report a bone density scan. CONCLUSION Although most have a PCP, over half of young survivors continue to see an oncology provider for follow-up in long-term survivorship. Mammography uptake was high; uptake of other health maintenance practices was lower. Identification of barriers/facilitators to receipt of guideline-concordant care may inform strategies to optimize survivorship.

ABSTRACT Objectives To characterize practice patterns and outcomes in the management of low‐ and intermediate‐grade salivary gland carcinoma prior to the existence of treatment guidelines. Methods Analysis of a registry of patients who underwent parotid and submandibular gland resections for low‐and intermediate‐grade carcinomas between 2010 and 2019. Results Of all 786 patients included in the study, 726 (92%) had preoperative imaging and 653 (83%) had preoperative biopsy. Of the 729 patients with parotid gland cancer, the majority underwent superficial ( n = 384, 53%) or total ( n = 254, 35%) parotidectomy. In patients with facial nerve preservation, total parotidectomy was associated with a significant increase in transient facial weakness (72/177 (41%) vs. 82/311 (26%), RR 0.65, 95% CI 0.50–0.84, p < 0.05) and permanent facial nerve weakness (23/176 (13%) vs. 16/301 (5%), RR 0.41, 95% CI 0.22–0.75, p < 0.05) compared to superficial parotidectomy. Adjuvant radiation therapy ( RT ) was delivered to 285 (36%) patients. The proportion of patients receiving adjuvant RT declined significantly over the time period from 2015 to 2019 compared to 2010 to 2014 at 162/504 (32%) and 123/282 (44%), respectively ( RR 0.74, 95% CI 0.61–0.89, p < 0.05). When comparing the time periods from 2015 to 2019 and 2010 to 2014, there was no significant difference in local control rates ( RR 0.52, 95% CI 0.26–1.04, p = 0.06) or regional control rates ( RR 0.75, 95% CI 0.26–2.13, p = 0.58). Conclusions Management of low‐ and intermediate‐grade salivary cancer from 2010 to 2019 was variable, which is expected given the rarity and heterogeneity of the disease and the lack of treatment guidelines prior to 2021. Most patients with parotid malignancies underwent superficial or total parotidectomy. The extent of parotidectomy had an impact on facial nerve function outcomes. Delivery of adjuvant radiation trended down with time. The data presented here will support dissemination of the guidelines and provide data that could inform future trials. Level of Evidence 2b.

Abstract Error analysis is a critical step in evaluating and improving clinical concept extraction models, the most common clinical natural language processing (NLP) task. Unlike corpus annotation, which follows standardized protocols for creating gold-standard datasets, error analysis requires nuanced judgment grounded in both clinical expertise and NLP knowledge. This is especially important given the heterogeneity of clinical text, where variations in documentation style, note structure, and terminology can substantially influence model behavior. Despite its importance, there is currently no standardized, user-level framework to support systematic error analysis in clinical concept extraction task. In this study, we developed and validated MedError, a machine-assisted, human-in-the-loop framework designed to standardize and enhance error analysis for clinical concept extraction tasks. We collected and manually curated a corpus of 1,187 unique errors from a total of 4,237 notes across three different distinct hospitals. The error categories were defined using our previously validated error taxonomy and included 480 false negatives and 707 false positives across 25 error types and 48 clinical concept categories. We evaluated the performance of three proprietary and three open-source large language models (LLMs) in automatically classifying these errors into 26 and 15 predefined categories. We further developed a machine-assisted framework, MedError, which integrates best practices in error analysis, LLM-assisted classification and reasoning, and a user-friendly interface to enable more efficient, reproducible, and context-aware error analysis. The framework supports both single-site and federated multisite error analysis, facilitating the effective translation of clinical NLP systems into real-world settings.

Wastewater treatment is evolving rapidly with the advent of advanced deep-learning AI, graph-based, and physics-informed approaches. This study integrates graph neural networks, physics-informed neural networks, and multi-agent reinforcement learning within a hybrid digital-twin framework, evaluated on multi-scale real-world datasets. The developed models achieved over 90% dissolved organic carbon removal and reduced aeration energy by up to 22% while maintaining process stability. The results demonstrate that graph–physics synergies not only boost operational efficiency but also reveal critical trade-offs between energy savings and hydraulic performance. Our findings establish a new benchmark for resilient, low-carbon wastewater treatment, highlighting the transformative role of data-driven system design.

Clean water security demands autonomous systems that sense, reason, and act at scale. We introduce AquaSurveil, a unified multi-agent platform coupling mobile robots, fixed IoT nodes, and privacy-preserving machine learning for continent-scale water health surveillance. The architecture blends Gaussian-process mapping with distributed particle filtering, multi-agent deep-reinforcement Voronoi coverage, GAN/LSTM anomaly detection, and sheaf-theoretic data fusion; components are tuned by Bayesian optimization and governed by Age-of-Information-aware power control. Evaluated on a 2.82-million-record dataset (1940–2023; five countries), AquaSurveil achieves up to 96% spatial-coverage efficiency, an ROC-AUC of 0.96 for anomaly detection, ≈95% state-estimation accuracy, and reduced energy consumption versus randomized patrols. These results demonstrate scalable, robust, and energy-aware water quality surveillance that unifies robotics, the IoT, and modern AI.

BACKGROUND: Whereas antiphospholipid syndrome (APS) is best known for increasing the risk of macrovascular thrombosis, APS vasculopathy is characterized by the abnormal proliferation of endothelial and smooth muscle cells, leading to occlusion of small blood vessels in the skin, kidneys, and heart, among other organs. The underlying mechanisms remain unclear, and targeted treatment options for patients with APS are lacking. METHODS: To identify and analyze APS microvascular endothelial cells (MVECs), skin biopsies of patients with APS complicated by livedo racemosa were characterized using single-cell RNA sequencing. CCN1 (cellular communication network factor 1) and CCN2 were identified in skin and kidney biopsies by immunofluorescence microscopy and in plasma by ELISA. Healthy dermal MVECs were cultured with APS patient–derived serum or immunoglobulin G (IgG), and relevant signaling pathways were characterized using quantitative PCR, immunoblotting, and immunofluorescence microscopy. The proliferation and migration of vascular smooth muscle cells were determined after exposure to conditioned medium from APS IgG–stimulated MVECs. A mouse model of APS IgG–accelerated neointima formation was developed. Anti-CCN2 monoclonal antibodies were tested in vascular smooth muscle cell functional assays and the mouse model. RESULTS: Increased endothelial cell expression of CCN1 and CCN2 was identified by single-cell RNA sequencing and further confirmed in APS skin by microscopy as well as in APS plasma by ELISA. Exposure of healthy MVECs to patient IgG triggered the upregulation of CCN1 and CCN2 via Toll-like receptor 4– and YAP1 (yes-associated protein 1)–mediated signaling. CCN2, originating from APS IgG–stimulated MVECs led to the proliferation and migration of vascular smooth muscle cells, phenotypes that were inhibited by an anti-CCN2 antibody or depletion of EGFR (epidermal growth factor receptor). We further observed increased expression of CCN2, along with evidence of YAP1 nuclear translocation, in kidney vessels of APS nephropathy biopsies. Finally, CCN2 inhibition with an anti-CCN2 monoclonal antibody significantly reduced neointima thickening and cell proliferation in a mouse model of APS IgG–accelerated neointima formation. CONCLUSIONS: This study revealed activation of YAP1-mediated signaling in APS dermal microvessels and demonstrated that the YAP1 target CCN2 plays a role in facilitating pro-proliferative communication between MVECs and vascular smooth muscle cells. These findings offer insights into the cellular and molecular mechanisms underlying APS vasculopathy, providing potential therapeutic targets for patients.

Importance Women diagnosed with breast cancer at a young age are felt to have a higher risk for locoregional recurrence (LRR) regardless of type of local therapy. Objective To assess the long-term incidence of isolated LRR by molecular subtype in a modern multicenter cohort of young women. Design, Setting, and Participants This cohort study, a multicenter prospective study named the Young Women’s Breast Cancer Study, enrolled 1302 women diagnosed with breast cancer at 40 years or younger from 2006 to 2016. Treatment information and incident LRR (ipsilateral breast/chest or lymph node recurrence) were self-reported on study surveys and confirmed with medical record review; molecular subtype was determined by record review. Analysis was reported from February 2023 to May 2025. Main Outcomes and Measures Cumulative incidence of isolated LRR was calculated using the Kaplan-Meier method; hazard ratios were estimated by Cox proportional hazards regression. Results The cohort included 1135 women with stage I through III breast cancer who had a median follow-up of 10.1 years (range, 0.4-16.3 years). The age at diagnosis was younger than 30 years for 145 patients (12.8%), 31 to 35 years for 318 patients (28.0%), and 36 to 40 years for 672 patients (59.2%). There were 59 isolated local recurrences (5.2%) and 4 isolated regional recurrences (0.4%). Among patients with local therapy and subtype data available (n = 1128), 366 (32%) had luminal A–like tumors; 240 (21%), luminal B–like tumors; 231 (20%) luminal ERBB2 positive (+)–like (formerly HER2 positive); 90 (8%) ERBB2 +-like; and 201 (18%) triple negative. A total of 346 women (30%) had breast-conserving therapy (BCT) (98% of whom had radiation), 296 (26%) unilateral mastectomy, and 487 (43%) bilateral mastectomy. Of women who had mastectomy, 425 (54%) had radiation. The cumulative incidence of LRR at 10.1 years by subtype was as follows: luminal A, 4.4% (range, 1.0%-6.9%); luminal B, 4.7% (range 1.8%-7.7%); luminal ERBB2 +, 6.1% (range, 3.1%-8.3%); ERBB2 +, 2.2% (range, 0%-6.3%); and triple negative, 6.5% (range, 4.2%-10.1%). The cumulative incidence of LRR by locoregional treatment type at 10.1 years was 6.7% after BCT (range, 4.3%-10.1%), 6.5% after mastectomy without radiation (range, 0%-7.7%), and 2.4% after mastectomy with radiation (range, 1%-4.2%). Although mastectomy with radiation was associated with the lowest risk of LRR on multivariable analysis, when examined within molecular subtype, there were no differences seen. Conclusions and Relevance In this contemporary cohort of women diagnosed with breast cancer at age 40 years or younger, risk of isolated LRR was relatively low (5.6%) at a median follow-up of 10.1 years, and significant differences were not seen by tumor subtype. Concerns for long-term risk of LRR should not influence surgical decision-making with young women, irrespective of molecular subtype.