NLM - National Library of Medicine Grants

NLM - National Library of Medicine

Project Summary and Abstract I am applying for an NLM Grant for Scholarly Works in Biomedicine and Health to complete my history of medicine monograph: Bringing Up Baby: Race, Infant Mortality, and the Creation of Prenatal Care, 1900-1930. This book will trace the intertwining threads of public health, eugenics, racial science, Progressive Era philanthropy, and professionalizing obstetrics in a story of how Americans became aware of and sought to fix the problem of infant mortality in the early twentieth century. In the 1910s and 1920s myriad groups and organizations, both those interested in health and those interested in social reform, studied the extent and causes of infant mortality, lobbied state and federal governments for maternal and infant welfare funding, and attempted to convince the American public that pregnancy was a condition that required medical surveillance and intervention. This will be the first work of history to dive into these movements and determine how nationalism, race, and medical professionalism efforts shaped the development of prenatal health care in this country. There have been no historical studies devoted solely to prenatal care and my findings into the emergence of this medical specialty and public health concern show it to be rooted in particular racial politics and national health concerns of the early 1900s. Relying on a range of sources including federal infant mortality studies, public health journals, personal correspondence, medical reports, meeting transactions, sociological reports, and popular health pamphlets, I illustrate that prenatal health care originated in a time of eugenics, Jim Crow, and medical misogyny, and perhaps never fully left those values behind. Investigating the history of prenatal health care will expand the historical field of American reproduction and medicine as well as inform current racial disparities in maternal and infant health care and mortality. In addition, this study draws together and speaks across multiple fields in history including women’s history, medical history, political history, and social history. I have plans to publish with Rutgers University Press, the press that published my well-received and widely-read first book Lost: Miscarriage in Nineteenth-Century America.

NLM - National Library of Medicine

1 Medical errors are the third leading cause of death in the United States yet estimates of their total 2 burden and epidemiology remain largely unknown, with few comprehensive assessments 3 available. To address this gap, we propose leveraging the Retract-and-Reorder (RAR) method, 4 an existing health information technology (IT) tool that detects near-miss, self-caught order errors, 5 to better understand the underlying causes of medical errors. The RAR method has been reliably 6 used to detect wrong-patient and certain types of medication prescribing order errors. We 7 expanded its application to diagnostic imaging, identifying additional error types such as wrong- 8 site, wrong-contrast, wrong-side, and wrong-modality, using logic-based natural language 9 processing (NLP). However, over 42% of detected errors remained unclassified, requiring labor- 10 intensive manual review for further categorization. In this proposal, we aim to develop a scalable 11 pipeline that automatically classifies order errors and addresses unknown error types using 12 generative large language models (LLMs). To accomplish this, we will first (AIM 1) develop and 13 validate a generative LLM-based classification model for categorizing RAR events into predefined 14 error types, focusing on imaging order errors. We will compare its performance against the current 15 logic-based NLP approach, hypothesizing that the LLM will achieve equal or better accuracy by 16 correctly classifying known error and identifying previously missed error types, thereby improving 17 overall classification. Then, we will (AIM 2) demonstrate the scalability of the LLM pipeline by 18 applying it to medication order errors and developing a dissemination plan. We hypothesize that 19 LLMs can be readily adapted to diverse large sets of order types across various domains without 20 requiring fine-tuning. This study will establish the feasibility of developing an advanced, 21 automated, and scalable open-source tool for classifying and characterizing RAR events across 22 different medical orders. By identifying and understanding various order error types across 23 domains, this research will support the development of measures and targeted interventions to 24 improve patient safety. Furthermore, our privacy-preserving approach, achieved by deploying an 25 open-source LLM along with comprehensive documentation and structured dissemination, will 26 enable adoption across institutions and diverse healthcare settings. Beyond imaging and 27 medication orders, this framework could support cross-institutional implementation, facilitating its 28 expansion into other order domains.

NLM - National Library of Medicine

SUMMARY The cell type composition and cellular metabolism jointly shape a tissue’s microenvironment, and its function as a consequence. Over the past few years, spatial transcriptomics (ST) has become an important and commonly used method for mapping cellular states in tissues. As powerful as this approach is, however, transcriptomes constitute only one of many crucial biological modalities, including DNA, protein, and small molecules, which, upon integration, would provide a more comprehensive view of tissue architecture. Recently, spatial metabolomics (SM) by mass spectrometry imaging has become available and promises to enable the elucidation of entire metabolomes at high spatial resolution. At present though a robust approach is missing for the integration of spatial metabolomics with spatial transcriptomics. In this proposal we aim to develop novel algorithms for such an integration in the context of an important problem in cancer biology. When studying drug-treated tumors, ST and SM can reveal cellular states and the precise concentration of the drug that they are experiencing, respectively. We previously found that as cancer cells adapt to therapy, they undergo a set of cell state transitions that we have referred to as the ‘resistance continuum’. We also found in vivo evidence for the states along this continuum, however new insight requires an integration of spatial and temporal analysis. In our preliminary results, we showed the power of joint ST and SM analysis, but we were not able to track the clonal and drug treatment history of the cells over time. Thus an open question, with immense clinical relevance, is what is the effective concentration of a drug experienced by an adapting cell lineage. We propose to address this question here by deploying two independent frameworks. In Aim 1 we describe a method using an optimal transport framework to integrate data from a time-course comprising tumors adapting to a drug from different animals. Our computational framework will be designed to identify cellular transitions and propose specific hypotheses for testing. A second approach described in Aim 2 exploits novel lineaging technology and our established serial passaging approach for studying the same tumor over time. Analyzing this data will allow us to reveal the longitudinal history of a clone and reveal whether cells with lower or higher dose concentrations in their early adaptations were selected for higher drug resistance. Overall, the approaches developed in this proposal specifically address the challenges of spatial metabolomics and spatial transcriptomics data and we expect them to be of high value for many in the large community of researchers using spatial analyses.

NLM - National Library of Medicine

This project supports a Research Software Engineer (RSE) to significantly advance in-hospital patient care by developing and disseminating cutting-edge, AI-driven software tools for the early prediction of clinical deterioration. The broad, long-term objective is to transform patient monitoring by enabling timely interventions, thereby improving patient outcomes and reducing healthcare costs associated with acute deterioration events in non-critical care settings. This work directly supports Aim 2 of NIH grant R01NR020774. A primary specific aim is to develop next-generation, personalized deterioration prediction models leveraging an extensive clinical wearable dataset. The research design involves employing generative foundation models and multimodal learning. Key methods include self-supervised pre-training on unlabeled physiological time series data using frameworks such as SimCLR and BYOL, followed by fine-tuning on labeled deterioration events. Multimodal foundation models will be developed to integrate continuous vital signs from wearables with Electronic Health Record (EHR) data, utilizing novel fusion techniques to capture complex interactions. To address data scarcity for rare clinical events, cross-location data synthesis techniques, including Generative Adversarial Networks (GANs) and optimal transport-based methods, will be investigated to generate realistic synthetic physiological data. These models will be rigorously validated using 10-fold cross-validation for both short-term (4-hour) and mid-term (24-hour) prediction windows, aiming for superior accuracy, timeliness, and generalizability. The models' capabilities will also be assessed for predicting missing continuous vital values and demographic features based solely on recorded vitals. A second major aim, directly aligning with the RSE's short-term career goal, is the creation and public release of a robust, open-source Python package for comprehensive validation of wearable sensor data against multiple ground truth sources. This package will incorporate time alignment algorithms, visualization tools (e.g., scatterplots, Bland-Altman plots), and automated statistical tests. Its development will adhere to research software engineering best practices, including a modular architecture for interoperability, extensive unit and integration testing for robustness, comprehensive documentation for user adoption and tools for distributed computing to handle large datasets. The RSE's long-term career objective involves building a platform to operationalize the developed deterioration foundation model, specifically the Continuous Clinical Alert System (CCAS). This platform will provide secure, scalable infrastructure for real-time data streaming, EHR integration, and seamless deployment of CCAS outputs into clinical workflows, supporting the entire AI/ML Software as a Medical Device (SaMD) lifecycle and eventual FDA submission. These RSE activities are critical for translating advanced AI research into clinically impactful tools, enhancing patient safety, and establishing sustainable research software.

NLM - National Library of Medicine

PROJECT SUMMARY Preventable medication errors are a global problem that can cause significant patient harm and annually incur costs of $42 billion worldwide. In the United States, 3 million outpatient medical appointments, 1 million emergency department visits, and 125,000 hospital admissions each year are the result of medication errors. Medication errors result in 3 million outpatient medical appointments, 1 million emergency department visits, and 125,000 hospital admissions each year. Astoundingly, over 4 billion prescriptions are dispensed every year in the United States alone. Although dispensing error rates are generally low at 0.06%, the sheer volume of dispensed medications translates to 2.4 million incorrectly dispensed medications each year. In the pharmacy, dispensing errors arise when pharmacists do not detect that the medication filled inside a prescription vial is different from the medication ordered on the prescription’s label. These dispensing errors can result in patient harm, added strain on the healthcare system, and costly legal action against the pharmacy. Artificial intelligence (AI) can be employed to assist in the verification process to help avoid dangerous and costly pharmacy dispensing errors. However, for the human-AI partnership to function optimally, the AI should be capable of determining the relative risks of medication errors (e.g., warfarin vs. vitamin C, kidney function, pregnancy status) while encouraging providers to make sound cognitive decisions such that optimal trust is maintained (i.e., catching errors), and temporal and cognitive demand is reduced (i.e., improving efficiency and avoiding alert fatigue). Risk-sensitive classification is critical when misclassification errors widely vary in frequency and severity. Imperative to this goal is to design AI from which risk-sensitive information can be extracted and conveyed to calibrate user’s trust in AI as either over-trust or under-trust can lead to near miss and incident errors. This proposed project will further our knowledge for designing risk- sensitive AI outputs and inform the development of AI models that encourage pharmacy staff to make sound clinical decisions that lead to better patient outcomes while improving work-life at lower costs of care. This study develops risk-sensitive AI methods in the context of medication images classification and designs effective AI advice and reasoning that lead to lower cognitive demand and increased trust in the AI. Our hypothesis is that risk-sensitive AI will lead to improved pharmacist work performance and more calibrated trust. The objectives of this proposal are to: 1) design risk-sensitive artificial intelligence to double-check dispensed medication images in real-time; 2) evaluate changes in pharmacy staff trust due to the use of risk- sensitive artificial intelligence; and 3) determine the effect of risk-sensitive artificial intelligence on pharmacy staff work performance.

NLM - National Library of Medicine

PROJECT SUMMARY A major challenge in modern biological data analysis is integrating and reasoning over the vast volumes of unstructured, multimodal data now available, such as images and text. While each modality offers complementary biological insight, they are not encoded in a shared format or language, making it difficult to align them or reason across them computationally. A core unmet need is the development of AI systems that can bridge this divide by learning shared representations across data types. This project targets building AI systems for aligning and reasoning jointly over biological image and text data. We focus on microscopy and the challenge of interpreting microscopy images often requiring integration with broader biological context—relating observed phenotypes to those seen in other experiments, identifying plausible mechanisms, and connecting to relevant prior studies. This knowledge is frequently buried in unstructured images and text scattered across publications, databases, and annotations. Conventional AI systems rely on supervised learning, which demands large amounts of manually annotated data and cannot scale to the complexity or breadth of modern biology. Training AI models using weak supervision offers a promising alternative: by learning from loosely aligned image-text pairs, models can capture cross-modal associations from noisy but abundant sources. Vision-language models (VLMs) built on this principle embed images and text into a shared semantic space and support flexible reasoning tasks such as retrieval and question answering. However, current models often suffer from “blurry vision”—they can identify broad semantic matches between images and text but fail to resolve the fine-grained visual distinctions essential for biological interpretation. The goal of our project is to overcome this limitation by advancing weak supervision methods that enable fine-grained alignment between biological images and text, with a focus on microscopy. We will curate a large and diverse dataset of fine-grained image-text pairs and train a visual encoder using multi-scale contrastive learning to integrate both global and local alignment signals. This encoder will power an agentic AI system for query-based interpretation of microscopy images, that can retrieve relevant biological evidence and generate natural-language interpretations of microscopy images in response to researcher queries. We will validate the system in expert-driven use cases spanning single-cell perturbation and tissue-level pathology, and disseminate it through integration into widely used imaging workflows. By building AI tools that help researchers connect microscopy image content to pathways, phenotypes, and prior studies, we aim to support flexible, biologically grounded exploration and accelerate data-driven discovery. By open-sourcing our datasets, methods, and trained models for fine-grained image-text alignment, we also aim to advance the broader capabilities of multimodal AI for biological data analysis.

NLM - National Library of Medicine

PROJECT SUMMARY Human gut microbiomes influence health by producing metabolites and enzymes that modulate immunity, transform drugs, and digest nutrients. However, most of these enzymes remain functionally unknown. Current annotation tools rely mainly on sequence similarity searches, which can only assign meaningful functions to less than 30% of microbial proteins. Although recent approaches incorporate protein language models and structural comparison, they still rely on predefined pipelines, manual literature or database searches, and specialized expertise in microbial research. This makes the annotations time-consuming without intelligent automation for context-aware insights and limits their scalability across diverse microbial ecosystems. Large language models (LLMs) have emerged as powerful tools in scientific research by analyzing data, answering complex questions, and generating new hypotheses. Building on these strengths, Artificial Intelligence (AI) agents, which combine LLMs with external resources like databases, tools and APIs, can automate tasks and workflows, mimicking human expert decision-making. Although they are widely used in industry, their potential in bioinformatics has only recently been explored. The overall objective of our project is to develop GENZ-AI (Gut ENZyme AI), a multi-agent AI system for automated curation and functional annotation of gut microbial enzymes. GENZ-AI will leverage LLM and advanced AI agents to autonomously delegate tasks, integrate diverse data sources, and deliver enriched annotations with relevant references. We will use advanced techniques, such as prompt optimization and imitation learning, to continuously refine its performance based on real-world annotation sample workflows and user feedback. The significance of GENZ-AI lies in leveraging these cutting-edge technologies to automate and enhance the data curation and workflow organization for enhanced enzyme annotation. This achievement will also improve gut microbiome-based diagnostics and therapeutics (e.g., dietary interventions, drug enhancement, immune modulation) while substantially reducing the time and effort required. The outcome will be a set of novel computational approaches implemented as user- friendly, reusable, open-source tools, including specialized applications for CAZymes, a class of glycan- metabolism enzymes critical to gut microbiome functions. The CAZyme annotation results and software tools will be integrated into dbCAN-PUL and dbCAN-sub databases. The key innovations of this project include a structure-informed protein language model for generalized EC number prediction, the application of CrewAI framework to build a multi-agent system optimized for enzyme annotation in microbiome, and the in-depth investigation of CAZyme and its glycan substrate utilization through GENZ-AI. The broader impact extends beyond the human gut microbiome, as GENZ-AI can be applied to any microbes, providing a scalable solution for diverse microbial ecosystems and pioneering the adaptation of LLM-powered AI agents in bioinformatics.

NLM - National Library of Medicine

PROJECT SUMMARY For individuals with chronic illnesses such as diabetes, heart failure, cancer, or obesity, early intervention prevents symptom escalation, acute care use, and mortality. The growing availability of longitudinal electronic health record (EHR), patient-reported outcome (PRO), and mobile health (mHealth) data, including passively collected accelerometer, smartphone, and sensor data, offers new opportunities for proactive intervention. However, the rapid expansion of routinely collected mHealth data has outpaced the research community's ability to interpret it effectively. In particular, the high dimensionality and irregular collection of longitudinal EHR, PRO, and mHealth data introduces key challenges for predictive modelling due to planned sparsity or unplanned missingness. Current methods fall short in three key areas: 1. Informative missingness: Data gaps often carry predictive signal, but are typically treated as nuisance, obscuring meaningful patterns in their timing and duration. 2. Loss of intra-day detail: Fine-grained mHealth data are often reduced to pre-specified daily or weekly summaries, discarding rich intra-day information with potential predictive value. 3. Population heterogeneity: Models trained on populations often perform poorly for underrepresented groups and fail to generalize to individuals, especially when only limited data are available per person. To address these gaps, we propose developing a robust methodological framework for predictive modelling using irregularly collected EHR, PRO, and mHealth data that improves upon imputation-based standard of care methods. In Aim 1, we will develop univariate and multivariate longitudinal models that account for delays between predictors and outcomes and incorporate detailed intra-day mHealth patterns using distributional learning with low-dimensional, near-lossless embeddings. In Aim 2, we address population heterogeneity by personalizing prediction through an embedding-based approach using landmark multidimensional scaling (MDS) and transfer learning, with reweighting of MDS landmarks to improve performance for underrepresented subgroups. In Aim 3, we validate these methods across diverse mHealth and EHR datasets, including NIH All of Us, UK Biobank, and other disease-agnostic and -specific retrospective and prospective datasets, using mixed-methods studies among clinicians to operationalize model outputs for clinical decision support. Though broadly applicable to multi-modal longitudinal data of all types and a range of disease settings, we focus on four chronic conditions with high clinical impact: cancer, congestive heart failure, diabetes, and obesity. The success of this project and its open-source tools will help close a critical methodological gap and enable effective use of multi-modal longitudinal data to improve clinical decision-making for chronic disease management.

NLM - National Library of Medicine

Emerging spatial omics atlases, including Human Cell Atlas, HubMAP, and Human Tumor Atlas, open new opportunities to investigate the relationships between cell spatial arrangement and tissue functions in various biological systems and diseased microenvironments. However, topological coordinating rules among different cell types, such as tissue spatial patterns in functional modules, are still under-investigated as the general unit across various tissues and diseases. Different from clustering cell type composition from classical top-down multicellular neighborhood analysis, bottom-up methods formulate the cell organization from Cellular Community (CC) motifs as discrete conservative patterns of recurring interconnections of various cell types. We hypothesize that CC motifs surrogating functional modules serve as precise modular markers, enhancing interpretability, generalizability, and sensitivity in comparative analyses and facilitating detailed mechanistic understanding in spatial omics studies, including spatial transcriptomics, proteomics, epigenomics, and histopathological images. In this project, we propose building a scalable and generalizable computational framework to comprehensively study conservative spatial organizations as functional modules, alongside an AI-innate web portal to quantitatively inquire, analyze, and curate the knowledge on functional modules in atlas-level spatial omics studies. We first analyze spatial multicellular neighborhoods by proposing a computationally efficient search algorithm CC-index identifying various conservative CC motifs as functional modules. In the triangulated tessellation space from spatial omics at the atlas level, the proposed approach is specially optimized to make precise multi-size CC motif identification computationally feasible (Aim 1). These functional modules facilitate further spatial omics research expanding across multiple markers, samples, modalities, longitudes, resolutions, and scales. They can be further generalized to compare different biological statuses within multiple markers in continuous transcriptomics and proteomics expression. These functional modules will be used to facilitate various methodological and biomedical applications in spatial omics research and different applications in spatial omics research (Aim 2). A large language model (LLM)-powered web portal, FMportal, will provide researchers with a user-friendly interface to query, analyze, and organize functional modules through a vector database across spatial omics atlases. This AI-innate system includes natural language interaction and empowered information summarization with RAG and reasoning (Aim 3). All the generated results and knowledge will be carefully evaluated at multiple levels and through validations. Enabled by fast accumulating data, this work is likely a game changer, transforming spatial omics data into functional modules that reveal the broad and fundamental roles of spatial organization in tissue differentiation, organ development, disease progression, and drug response across diverse biomedical and pathological contexts.

NLM - National Library of Medicine

Project Summary Foundation models (FMs), including large language models (LLMs), have created immense promise and excitement with their potential to change multiple aspects of how we pursue the science, practice and delivery of medicine. Given the potential, there is a need to create systematic ways to verify presumed benefits of using LLMs for medical tasks. The challenges in discerning hype from benefit are largely due to the absence of shared benchmarks that enable consistent evaluation (benchmarking), little guidance on trade-offs in performance for different post-training strategies (adaptation recipes), and a dearth of studies on the optimal mode of combining human expertise with models (human-AI teaming). We investigate methods to systematically evaluate LLMs in a setting where there is interaction between human experts to generate reusable guidance for performance benchmarking, adaptation recipes using human preferences, and implementing human-ai teaming in clinical settings. If successful, we will develop and implement concrete procedures to verify benefits of using LLMs in the context of the clinical tasks for which they are used. Overall, we seek to improve the creation and evaluation of foundation models in healthcare by rigorous benchmarking, developing systematic adaptation recipes, and guidance for optimal human-AI teaming setups.

NLM - National Library of Medicine

Project Summary Comparative effectiveness research is a critical component of medical research that relies on randomized experiments and observational study evidence to identify interventions that improve healthcare outcomes. In many cases, the analysis of the data obtained from such studies is complicated by truncation by death, time-varying confounders, and zero-inflated outcomes. A common approach to address these difficulties is to use models that rely on unrealistic, untestable assumptions. In many cases, the quality of the evidence is low, since it is hard to assess which assumptions are necessary to make inferences. To that end, we develop ablation methods for causal inference. The result is an ablation framework that allows analysts to understand which assumption or combination of assumptions lead to informative conclusions. For instance, analysts will be able to identify the minimum set of assumptions that are needed for either point identifi- cation or the sign of the effect. We develop three different causal ablation frameworks for three specific applications that are common in comparative effectiveness research: truncation by death, time-varying confounders, and zero-inflated outcomes. Under each aim, for each application, we develop (1) appropriate assumption sets that meet a compatibility criterion, (2) a grid search tai- lored to the assumption sets, (3) a corresponding set of point and partial identification estimators, and (4) a summary step that reports results by order of assumption credibility.