Citations

Abstract Scientific foundation models learn high-dimensional representations from diverse data modalities, yet what they encode and how to extract that knowledge remain open questions. Here we show that probing the internal representations of Evo 2, a 7-billion-parameter genomic foundation model, enables accurate and interpretable genetic variant effect prediction. We introduce a covariance-based probe that captures second-order structure from Evo 2 sequence embeddings to predict variant pathogenicity across variant types and functional consequences, matching or exceeding specialized predictors within their domains. To ground these predictions in known biological mechanisms, we train a complementary panel of probes on existing annotations to detect which genomic properties are disrupted by a variant. This categorized evidence is then integrated with each variant’s genomic context through a language model to generate variant-specific mechanistic hypotheses. Our pathogenicity predictions correlate with experimental measures of variant function, clinical penetrance, and biobank disease associations while the mechanistic hypotheses are consistent with expert reviews, known mechanism classes, and downstream molecular readouts. We release pathogenicity scores, disruption profiles, and contextualized interpretations for 4.2 million variants from the ClinVar database as an open resource through the Evo Variant Effect Explorer (EVEE). More broadly, this structured probing approach offers a general framework for interrogating foundation models across scientific disciplines and grounding their outputs in existing domain concepts.

Abstract Objectives Patients referred for specialized care often arrive with outside medical records (OMRs) compiled into multi-report PDFs that include imaging, pathology, and clinical notes in unstructured formats. Reviewing these records is time consuming and mentally taxing, increasing the risk of delayed care, clinician frustration, and missed information affecting quality of care. This study aimed to automate the segmentation, classification, and date extraction of scanned OMRs, with a focus on records relevant to breast cancer care. Materials and Methods We used optical character recognition (OCR) to extract machine-readable text from 1303 scanned PDF documents from 116 distinct external institutions. Gemini 1.5, a large language model (LLM), was then used to segment multi-report files into individual documents, classify them into clinically meaningful categories such as mammograms and pathology reports, and extract study dates to build diagnostic timelines. Document categories were informed by clinical workflows in a breast cancer center. Results The system achieved an F1 score of 0.95 for segmentation, 0.96 for classification, and 0.90 for date extraction. In a pilot of 45 records reviewed by clinicians, only 2 classification errors and 1 date error were reported. Clinicians estimated that the tool reduced OMR review time by 40%, improved workflow efficiency, and increased satisfaction. Discussion Our findings demonstrate that combining OCR with LLMs can significantly enhance the processing of unstructured medical records, reducing manual burden and supporting timely clinical decision-making. Conclusion This study demonstrates the successful application of OCR and LLMs for organizing scanned OMRs within a specialty clinic. By automating a previously manual process, the approach supports scalable review of incoming outside records and has potential for adaptation to other clinical workflows. Future work will focus on evaluating the system across additional specialties and institutions.

Abstract The advent of high-throughput sequencing has led to an exponential increase in genomic data, highlighting the need for efficient and accurate models to analyze and interpret this information. In this study, we introduce a novel, exomic foundational model that leverages a combination of the human reference genome and multispecies data to improve variant detection and interpretation. Our model utilizes a short-range transformer architecture and is trained on a large dataset of human exomic sequences derived from the Tapestry study. Through a series of ablation studies and scaling experiments, we demonstrate the effectiveness of our model in predicting next token accuracy and identifying clinically pathogenic variants. We also show that our model outperforms existing models in a range of downstream tasks, including variant effect prediction and disease state identification. In fact, our largest sequence transformer for range adaptive nucleotide decoding variant (1B parameters) surpassed previous benchmarks, demonstrating a mean accuracy of 0.880 [an 8.2% improvement over the original nucleotide transformer (NT) and a 7% improvement over NT-v2]. Furthermore, we construct a unique exomic ClinVar dataset to evaluate the model’s performance on pathogenicity and disease states. Our results highlight the potential of this model to improve our understanding of the human exome and its role in disease. The model and its applications have significant implications for genomics-based diagnosis and personalized medicine, including tailored therapeutic development.

Abstract Background Hospital discharge summaries are critical for ensuring continuity of care, but their quality often varies. Large language models (LLMs) have the potential to standardize and enhance the efficiency of this documentation process. Objectives To evaluate the quality of hospital discharge summaries created by an LLM‐based hospital course drafting tool created by Epic Systems compared with human‐written summaries. Methods Retrospective study at a single tertiary‐care institution in 2024. The cohort included 100 adult hospitalizations lasting >72 h across medical and surgical dismissing services. No interventions were performed. Summaries (LLM‐generated vs. human‐written) were independently reviewed using a standardized rubric covering nine domains (e.g., comprehensiveness, clarity, relevance). Scores were normalized and compared. Readability was assessed using Flesch Reading Ease. Results LLM‐generated summaries outperformed human‐written summaries across all criteria ( p < .05), with the greatest difference observed in comprehensiveness (LLM median 0.62 vs. human −0.23). Human‐written summaries from surgical services scored lower than those from medical services, but LLM performance was consistent across both. Human summaries had higher Flesch Reading Ease scores (33.11 vs. 26.2; p < .05), reflecting simpler language. Conclusions LLM‐generated summaries demonstrated superior quality, consistency, and clinical utility compared with human‐written summaries, highlighting their potential to improve documentation efficiency and standardization.

Abstract Background: Spatial Reasoning AI (SRAI) systems are emerging as a transformative class of clinical artificial intelligence, enabling real-time interpretation of complex anatomy in imaging, surgery, and chronic disease management. Unlike traditional AI approaches constrained to tabular or 2D inputs, SRAI models incorporate geometric priors, multimodal fusion, and 3D anatomical reasoning to support high-impact tasks such as volumetric segmentation, lesion localization, intraoperative guidance, and prognostic modeling. Objective: This scoping review aimed to (1) map the emerging landscape of SRAI in healthcare, (2) assess its clinical performance across diagnostic, prognostic, and workflow applications, and (3) identify opportunities and barriers for safe translation into practice. Methods: Following PRISMA 2020 guidelines, we systematically searched PubMed, Springer Nature, and IEEE Xplore (including CVPR, ICCV, ISBI, and EMBC) for original studies implementing spatially aware AI models with clinical validation or workflow relevance. Data extraction captured model architecture, input modality, anatomical scope, application domain, and performance outcomes. Results: We identified eligible studies spanning oncology, neurology, cardiology, surgery, and infectious diseases. SRAI systems consistently outperformed conventional methods in segmentation, lesion classification, and risk stratification. A paradigm shift is underway from task-specific CNNs to foundation-level models, including MedSAM and MedCLIP for zero-shot segmentation and retrieval, and NeRF-based pipelines for anatomically faithful 3D reconstruction. Randomized evidence shows that SRAI can improve workflow efficiency without compromising diagnostic accuracy. Conclusion: SRAI models demonstrate strong potential to advance diagnostic precision, prognostic accuracy, and workflow integration across multiple clinical domains. Their safe adoption will require standardized evaluation metrics, prospective multi-institutional validation, and interpretable outputs. As foundation models evolve to better encode spatial priors, SRAI is poised to become a cornerstone of workflow-integrated precision medicine.

Abstract Background: Spatial Reasoning AI (SRAI) systems are reshaping clinical practice by enabling real-time interpretation of complex anatomical structures in medical imaging, surgery, and chronic disease management (Lin et al., 2023, Qiao et al., 2021; Song et al., 2024) . Unlike traditional AI approaches that rely on tabular or 2D data, SRAI models incorporate geometric priors, multimodal fusion, and 3D anatomical reasoning to support high-impact tasks such as volumetric segmentation, disease localization, and intraoperative guidance (Milletari et al., 2016; Hatamizadeh et al., 2022; Zhang et al., 2022) . Objective: This scoping review aimed to (1) characterize the emerging landscape of spatial reasoning models in healthcare, (2) evaluate their clinical performance across diagnostic and prognostic applications, and (3) identify architectural trends that underpin successful translation into practice. Methods: Following PRISMA guidelines, we systematically screened PubMed to identify studies that implemented spatially aware AI models with real-time or clinically validated outcomes (Page et al., 2021) . Data extraction captured model architecture, modality, anatomical scope, clinical application, and performance metrics. Results: We identified 124 eligible studies across the clinical domains of oncology, neurology, cardiology, orthopedic surgery, and infectious diseases (Nie et al., 2019; Hao et al., 2020; Tasken et al., 2023) . SRAI systems consistently outperformed conventional methods in segmentation, lesion classification, and risk stratification. (Lin et al., 2023; Qiao et al., 2021; Mahootiha et al., 2024) . A paradigm shift from task-specific CNNs to foundation models that reason across spatial structures has emerged, beginning with adaptations of generalist models such as MedSAM and MedCLIP for zero-shot segmentation and cross-modal retrieval, and NeRF-based reconstruction for anatomically faithful 3D modeling (Ma et al., 2024; Mildenhall et al., 2020; Radford et al., 2021). Beyond technical performance, randomized evidence shows that SRAI systems can improve workflow efficiency without compromising outcomes (Lang et al., 2023) . Conclusion: SRAI models hold promise for enhancing diagnostic accuracy, prognostic precision, and clinical efficiency across diverse applications. Their successful integration will depend on standardized evaluation metrics, robust validation, and interpretable outputs. As foundation models evolve to better encode spatial priors, SRAI is poised to become a cornerstone of personalized and procedural medicine.

Abstract The advent of high-throughput sequencing has led to an exponential increase in genomic data, highlighting the need for efficient and accurate models to analyze and interpret this information. Here, we introduce a novel, exomic foundational model that leverages a combination of human reference genome and multispecies data to improve variant detection and interpretation. Our model utilizes a short- range transformer architecture and is trained on a large dataset of human exomic sequences derived from the Tapestry study. Through a series of ablation studies and scaling experiments, we demonstrate the effectiveness of our model in pre- dicting next token accuracy and identifying clinically pathogenic variants. We also show that our model outperforms existing models in a range of downstream tasks, including variant effect prediction and disease state identification. In fact, our largest STRAND variant (1B parameters) surpassed previous benchmarks, demonstrating a mean accuarcy of 0.880 (8.2% improvement over the original NT and 7% improvement over NT-v2). Furthermore, we construct a unique exomic ClinVar dataset to evaluate the model’s performance on pathogenicity and disease states. Our results highlight the potential of this model to improve our understanding of the human exome and its role in disease. The model and its applications have significant implications for genomic based diagnosis and personalized medicine including tailored therapeutic development.

Background Patient admissions at a U.S. tertiary care hospital occur via the emergency department (ED), or transfer center. We aim to compare the clinical outcomes of patients admitted from the ED to admissions coordinated by the transfer center. Methods Admissions to Mayo Clinic Hospital, Rochester, MN, between July 2019 to June 2021 were identified in this retrospective study and categorized into two cohorts—transfer center and ED. The two cohorts were then matched for age, sex, admitting service, and Charlson Comorbidity Index. Univariate and multivariate analyses were performed to compare hospital length of stay (LOS), mortality, 30-day mortality, and 30-day readmissions between the two cohorts. Results 73,685 admissions were identified, of which 24,262 (33%) were transfer center admissions. In the matched cohorts (n = 19,093, each), in-hospital mortality (2.4% versus 1.9%), 30-day mortality (5.4% versus 3.9%), 30-day readmission (12.7% versus 7.2%), and LOS (6.4 days versus 5.1 days) were significantly higher (P < 0.001) among the admissions coordinated by transfer center. A higher palliative care consultation rate (9.4% versus 6.2%, P < 0.001), and a lower proportion of home discharges home (76.2% versus 82.5%, P < 0.001) among transfer center admissions was observed. Similar findings were noted in multivariate analysis, even when adjusting for LOS. Conclusions Transfer center admissions had higher in-hospital mortality, LOS, 30-day mortality, and 30-day readmission compared to ED admissions. This study also highlights new considerations for palliative care consultation before transfer acceptance, especially to avoid futile transfers. Additional studies analyzing factors behind the outcomes of transfer center admissions are required.

Enteropathy-associated T cell lymphoma (EATL) is an aggressive subtype of non-Hodgkin’s lymphoma often associated with coeliac disease (CD). We describe a previously healthy man in his 50 s who presented with a history of abdominal pain, failure to thrive and significant weight loss over a 3-month period. Investigations revealed a positive coeliac serology, diffuse duodenal atrophy with multiple duodenal and jejunal ulcers on endoscopy and mesenteric lymphadenopathy on CT imaging. Duodenal tissue biopsy confirmed a diagnosis of EATL Stage IVB. Chemotherapy with cyclophosphamide, doxorubicin, vincristine and prednisone regimen was initiated. This case highlights the need for greater awareness and consideration of EATL in individuals with worsening malabsorption and abdominal pain, irrespective of coeliac history.