Objectives/Goals: Rare disease patients often face lengthy delays in receiving accurate diagnoses or experience misdiagnoses due to a lack of available information. The NCATS Rare Disease Alert System (RDAS) is a public, comprehensive rare disease resource to collect and share accurate, up-to-date, and standardized data on rare diseases. Methods/Study Population: RDAS is composed of a frontend UI, Application Programming Interfaces, and backend Neo4j graph database. Each component of data collection, data annotation, data standardization, and data representation as steps were implemented during the process of each graph database creation. The UI allows users to search, browse, and subscribe to RDAS to receive the latest information and findings about their rare disease(s) of interest. The back-end data include four knowledge graphs built by integrating information from the NCATS Genetic and Rare Disease program, PubMed articles, clinical trials, and NIH grant funding. Ultimately, the integrative information pertinent to rare diseases from RDAS would advance rare diseases research. Results/Anticipated Results: Of 5001 rare diseases belonging to 32 distinct disease categories, we identified 1294 diseases that are mapped to 45,647 distinct, NIH-funded projects obtained from the NIH ExPORTER by implementing semantic annotation of project titles. To capture semantic relationships presenting among mapped research funding data, we defined a data model comprised of seven primary classes and corresponding object and data properties. A Neo4j knowledge graph based on this predefined data model has been developed, and we performed multiple case studies over this knowledge graph to demonstrate its use in directing and promoting rare disease research. Discussion/Significance of Impact: We developed an integrative knowledge graph with rare disease data and demonstrated its use as a source to identify and generate scientific evidence to support rare disease research. With the success of this study, we plan to implement advanced computation to analyze more funding related data and link to other types of data to perform further research.

Objectives/Goals: As a priority area in translational science, rural health research can benefit from informatics methods for conducting thematic and environment scans. This study demonstrates an efficient approach to gaining insights about the rural health research literature by automated bibliometrics analysis. Methods/Study Population: We developed an automated pipeline to retrieve the 1972–2023 PubMed publications indexed with the MeSH terms “Rural Health” and “United States”. The article metadata in XML format were downloaded and parsed, including title, year, journal, author institutions, and MeSH terms. Each institution address was augmented by Google Maps API to obtain the county and latitude/longitude coordinates. Summary statistics were computed for the publication years, journals, author departments, and locations. A topic network was generated from the frequent co-occurring MeSH terms. The institutions were linked to Rural-Urban Continuum Codes and labeled on a map to visualize their geographic distribution. Results/Anticipated Results: A total of 4564 articles on rural health were analyzed. Two salient peaks of publications were revealed, one around 1978 and the other around 1993. The top author departments include Family Medicine, Nursing, Pediatrics, and Epidemiology. The five leading institutions reside in Chapel Hill, Minneapolis, Iowa City, Seattle, and Atlanta. The geographic distribution shows few institutions that reside in deep rural areas are well published on rural health, although the most scholarly productive institutions do seem adjacent to some moderately rural pockets. The frequently identified topics pertain to age group, study design, and specific concepts such as health services accessibility. Discussion/Significance of Impact: The two publication peaks in history were likely linked to certain policy milestone or seminal publication. Primary care and epidemiology departments have been most active in rural health research. Of concern, the geographic distribution of authors suggests under-investment in rural institutions.

Objectives/Goals: We aim to identify how IDH mutant (IDHm) gliomas use exosomes to modulate the local and systemic immune system. We will do so by characterizing differential miRNA expression between IDHm and IDH wild type (IDHwt) exosomes and identifying the specific immune cell population targeted by exosomes in vivo. Methods/Study Population: Exosome RNA will be isolated from cultured patient glioma samples and perform small RNA sequencing to investigate differential expression of miRNA between IDHwt and IDHm exosomes. We will then utilize miRNA target databases in conjunction with bioinformatic pathway analysis to generate potential target regulatory pathways. To identify the in vivo effect of tumor exosomes, we will generate a novel glioma mouse model that has been genetically engineered to release labeled exosomes using the RCAS retroviral system. We will collect peripheral blood and tumor tissue for flow cytometric immune profiling and single-cell RNA sequencing. The transcriptomic data will be analyzed to identify subsets of immune populations that have taken up the labeled exosomes and assess the resulting expression changes in those cells. Results/Anticipated Results: From the small RNA sequencing and bioinformatics analysis, we expect to find several unique miRNA expressed in IDHm exosomes that induce immunosuppressive pathways in local and systemic immune cell populations when compared to IDHwt exosomes. Furthermore, using our novel murine model, we expect to be able to track endogenously released exosomes in the local tumor microenvironment and in the circulating blood. We hypothesize that IDHm exosomes specifically target precursor myeloid cells within the local and peripheral circulating immune populations and induce the expansion of monocytes, M2 macrophages, and mono-MDSCs. Discussion/Significance of Impact: Immunosuppression in IDHm glioma has hindered the development of adequate therapies to treat this fatal disease. Our study will illuminate the mechanism by which tumor exosomes can suppress immune surveillance. These results will help identify new therapeutic targets to sensitize the immune system against glioma cells.

Objectives/Goals: We aim to predict and rank conserved, immunogenic targets within key malaria proteins using computational tools. These tools incorporate parasite protein diversity and regional HLA allele frequencies to prioritize antigens for further validation and inclusion in a malaria vaccine targeting circulating strains. Methods/Study Population: We identified 42 conserved malaria proteins with nonredundant functions for P. falciparum invasion and transmission as vaccine targets. Protein sequence datasets were constructed from samples collected in highly endemic areas. We predicted targets most likely to be presented to CD4+ and CD8+ T cells. We designed and used heuristic-based and AI-weighting models that integrated predicted binding affinities to HLA alleles, HLA allele frequency data, and sequence conservation to score and rank targets. We validated our model by comparing predicted epitope distributions with published in vitro and in vivo immunogenicity data available in the Immune Epitope Database and Tools repository. Results/Anticipated Results: We successfully predicted and ranked targets within the vaccine candidate proteins, identifying conserved and HLA-nonspecific targets that correspond to positive immunogenicity data, validating our approach. We are currently analyzing model performance by comparing predictions to over 5,800 experimentally validated P. falciparum targets from clinical trials and immune assays. We will evaluate each models’ accuracy and ability to prioritize targets and compare their performances as measured quantitatively by precision and area under the curve metrics. We expect the AI-based model to significantly outperform the heuristic approach, improving the identification of effective vaccine targets. Discussion/Significance of Impact: By incorporating parasite diversity and regional HLA allele frequencies, our approach addresses the challenge of directing the human immune response against genetically diverse P. falciparum strains in highly endemic areas. This strategy could significantly enhance malaria vaccine efficacy and can be adapted for use against other pathogens.

Objectives/Goals: This study aimed to investigate the role of artificial intelligence (AI) in translational science, including personalization of interventions and drug development. Methods/Study Population: A comprehensive literature search was conducted via PubMed, the Cumulative Index for Nursing and Allied Health Literature (CINAHL), Cochrane Library, Medline, and Web of Science. The risk of bias in the eligible studies was assessed using the risk of bias in nonrandomized studies. Data were systematically extracted and analyzed. Results/Anticipated Results: The literature search yielded 2129 records, from which 20 studies that met the eligibility criteria were included. Meta-analysis demonstrated the high specificity of AI-based diagnostics, reassuring the reliability of AI. Furthermore, AI applications significantly improved biomarker identification through machine learning algorithms, enhancing prognostic accuracy and treatment personalization. Moreover, AI showed enhanced diagnostic precision with high sensitivity and specificity in cancer detection, further validating its role in healthcare. AI-driven risk stratification was used in chemotherapy decisions. Discussion/Significance of Impact: This study highlights the transformative power of AI in translational oncology research with applications in drug development and personalized patient care in cancer treatment and research.

Objectives/Goals: This study aims to evaluate the performance of a third-party artificial intelligence (AI) product in predicting diagnosis-related groups (DRGs) in a community healthcare system. We highlight a use case illustrating how clinical practice leverages AI-predicted information in unexpected yet advantageous ways and assess the AI predictions accuracy and practical application. Methods/Study Population: DRGs are crucial for hospital reimbursement under the prospective payment model. The Mayo Clinic Health System (MCHS), a network of clinics and hospitals serving a substantial rural population in Minnesota and Wisconsin, has recently adopted an AI algorithm developed by Xsolis (an AI-focused healthcare solution provider). This algorithm, a 1D convolutional neural network, predicts DRGs based on clinical documentation. To assess the accuracy of AI-generated DRG predictions for inpatient discharges, we analyzed data from 930 patients hospitalized at MCHS Mankato between March 2 and May 13, 2024. The Xsolis platform provided the top three DRG predictions for the first 48 hours of each inpatient stay. The accuracy of these predictions was then compared against the final billed DRG codes from the hospital’s records. Results/Anticipated Results: In our validation set, Xsolis achieved a top-3 DRG prediction accuracy of 71% at 24 hours and 81% at 48 hours, which is lower than the originally reported accuracy of 81.1% and 83.3%, respectively. Interestingly, discussions with clinical practice leaders revealed that the most valuable information derived from the AI predictions was the expected geometric mean length of stay (GMLOS), which Xsolis was perceived to predict accurately. In the Medicare system, each DRG is associated with an expected GMLOS, a critical factor for efficient hospital flow planning. A subsequent analysis comparing predicted GMLOS with the actual length of stay showed variances of -0.10 days on day 1 and 0.14 days on day 2, indicating a high degree of accuracy and aligning with clinical practice perceptions. Discussion/Significance of Impact: Our research underscores that clinical practice can leverage AI predictions in unexpected yet beneficial ways. While initially focused on DRG prediction, the associated GMLOS emerged as more significant. This suggests that AI algorithm validation should be tailored to specific clinical needs rather than relying solely on generalized benchmarks.

Objectives/Goals: This study aims to evaluate common features of mobile health (mHealth) apps and their role in helping caregivers make health decisions for children. Methods/Study Population: A scoping review of literature on caregivers’ use of mHealth apps (published since 2008) was conducted across 5 databases (i.e., Embase, PubMed, CINAHL, Clinicaltrials.gov, and IEEE Xplore). Selected papers were categorized based on app purposes, target users, and mHealth agile development phases. Common features were also identified and analyzed along with users’ pros and cons. Further, primary feature requests were summarized to inform future development. Results/Anticipated Results: This review included 62 studies. Most apps were about maternity and infant care and specific diseases. Major users were caregivers and pregnant women. Around 20% of papers covered multiple phases in the mHealth agile development lifecycle. The effectiveness/clinical trial (phase III) was the most common. E-learning, personalization and customization, and health tracking features were the three most common features of mHealth apps included in this review. More positive feedback was found regarding features than concerns. Caregivers perceived apps as helpful and empowered them to make informed decisions. Concerns were mainly over 1) technical issues, 2) inappropriate design, and 3) ambiguous terms. Requested new features included content comprehensiveness, user engagement, and usage flexibility. Discussion/Significance of Impact: To our knowledge, this is the first review to investigate the usability of mHealth app features in this area. The results offer feasible strategies for developers to improve the effectiveness of apps for caregiver decision-making.

Objectives/Goals: Oral health is an important and understudied part of overall health. Poor oral health is linked to many systemic conditions, but little has been done to explore these issues in large electronic health records data sources that include dental health records. Here we report on our exploration of data readiness and completeness of three of these data sources in the Clinical and Translational Science Awards (CTSA) network. Methods/Study Population: Three CTSAs from the Consortium of Rural States (CORES) with diverse geographies, demographics, and data ecosystems can integrate medical and dental records, but it is unknown if the target population having both dental and medical records have sufficient completeness and similarity to enable dental/medical health studies. Here we use descriptive analytics to characterize the demographics, and the “complete data” approach presented by Weber et al. to evaluate differences between the completeness of the general populations and the one having both dental/medical records. We accomplish this by identifying patients with dental records in commonly used research networks and performing empirical patient statistics in comparison to the entire population available at the three institutions. Results/Anticipated Results: This poster will present the results of using the Weber et al. approach to compare the completeness of records of the general patient population in the Iowa, Kentucky, and Utah medical/dental health care systems to those for which they have also dental records. The completeness of the records of these two subpopulations is also associated with different demographic characteristics, as it has been established that the populations served by the dental clinics is biased by dental insurance considerations. The work will show what retrospective studies can (or not) be done using these populations when taking into account that it is well established that studies of populations with different level of completeness can be inconsistent. Discussion/Significance of Impact: This study provides an informatics framework to assess similarity and completeness of patient records with and without dental records. Establishing the level of similarity and completeness in these patient populations is critical to justify the validity of studies that utilize a combined record.

Objectives/Goals: Identifying and indexing rare disease studies is labor intensive, especially in research centers with a large number of trials. To address this gap, we applied natural language processing (NLP) and visualization techniques to develop an efficient pipeline and user-friendly web interface. Our goal is to offer the rare disease study identification (RDSI) tool for adoption by other sites. Methods/Study Population: The RDSI retrieves study information (short and long titles, study abstract) from the IRB system. These descriptive fields are then processed by the MetaMap Lite NLP program for identifying disease terms and standardizing them to UMLS concepts. By terminology identifier mapping, the diseases intersecting with concepts in rare disease databases (Genetic and Rare Disease program and Orphanet) are further scored to pinpoint studies that focus on a rare disease. The web interface displays a scatter bubble chart as an overview of all the rare diseases, with each bubble size proportional to the number of studies for that disease. In addition to the visual navigation, users can search studies by disease name, PI, or IRB number. Search results contain detailed study information as well as the evidence used by algorithms of the pipeline. Results/Anticipated Results: The RDSI identification results and functions were verified manually and spot-checked by several study investigators. The web interface is a self-contained solution available to our staff for various use cases like reporting or environment scan. We have built in a versioning mechanism that logs the date of each major result in the process. Therefore, even as the rare disease data sources evolve over time, we will be able to preserve any historical context or perform updates as needed. The RDSI outputs are replicated to Mayo Clinic’s enterprise data warehouse daily, allowing tech-savvy users to leverage any useful intermediate results at the backend. We anticipate the performance of the rare disease identification to be further enhanced by employing the advancements in AI technology. Discussion/Significance of Impact: The RDSI represents an informatics solution that offers efficiency in identifying and navigating rare disease clinical studies. It features the use of public databases and open-source tools, manifesting return on investment from the broad translational science ecosystem. These considerations are informative and adoptable by other institutions.

Objectives/Goals: This work aims to identify functional brain networks that differentiate opioid use disorder (OUD) subjects from healthy controls (HC) using machine learning (ML) analysis of resting-state fMRI (rs-fMRI). We investigate the default mode network (DMN), salience network (SN), and executive control network (ECN), as well as demographic features. Methods/Study Population: This work uses high-resolution rs-fMRI data from a National Institute on Drug Abuse study (IRB #HM20023630) with 31 OUD and 45 HC subjects. We extract rs-fMRI blood oxygenation level-dependent (BOLD) features from the DMN, SN, and ECN. The Boruta ML algorithm identifies statistically significant features and brain activity mapping visualizes regions of heightened neural activity for OUD. We conduct fivefold cross-validation classification experiments (OUD vs. HC) to assess the discriminative power of functional network features with and without incorporating demographic features. Demographic features are ranked based on ML classification importance. Follow-up Boruta analysis is performed to study the medial prefrontal cortex (mPFC), posterior cingulate cortex, and temporoparietal junctions in the DMN. Results/Anticipated Results: Boruta ML analysis identifies the DMN as the most salient functional network for differentiating OUD from HC, with 33% of DMN features found significant (p < 0.05), compared to 10% and 0% for the SN and ECN, respectively. The Boruta ML algorithm identifies age and education as the most significant demographic features. Brain activity mapping shows heightened neural activity in the DMN for OUD. The DMN exhibits the greatest discriminative power, with a mean AUC of 69.74%, compared to 47.14% and 54.15% for the SN and ECN, respectively. Fusing DMN BOLD features with the most important demographic features improves the mean AUC to 80.91% and the F1 score to 73.97%. Follow-up Boruta analysis highlights the mPFC as the most important functional hub within the DMN, with 65% significant features. Discussion/Significance of Impact: Our study enhances the understanding of OUD neurobiology, identifying the DMN as the most significant network using ML rs-fMRI BOLD feature analysis. Ethnicity, education, and age rank are the most important demographic features and the mPFC emerges as a key functional hub for OUD. Future research can build on these findings to inform treatment of OUD.

Objectives/Goals: Aspiration causes or aggravates lung diseases. While bedside swallow evaluations are not sensitive/specific, gold standard tests for aspiration are invasive, uncomfortable, expose patients to radiation, and are resource intensive. We propose the development and validation of an AI model that analyzes voice to noninvasively predict aspiration. Methods/Study Population: Retrospectively recorded [i] phonations from 163 unique ENT patients were analyzed for acoustic features including jitter, shimmer, harmonic to noise ratio (HNR), etc. Patients were classified into three groups: aspirators (Penetration-Aspiration Scale, PAS 6–8), probable (PAS 3–5), and non-aspirators (PAS 1–2) based on video fluoroscopic swallow (VFSS) findings. Multivariate analysis evaluated patient demographics, history of head and neck surgery, radiation, neurological illness, obstructive sleep apnea, esophageal disease, body mass index, and vocal cord dysfunction. Supervised machine learning using five folds cross-validated neural additive network modelling (NAM) was performed on the phonations of aspirator versus non-aspirators. The model was then validated using an independent, external database. Results/Anticipated Results: Aspirators were found to have quantifiably worse quality of sound with higher jitter and shimmer but lower harmonics noise ratio. NAM modeling classified aspirators and non-aspirators as distinct groups (aspirator NAM risk score 0.528+0.2478 (mean + std) vs. non-aspirator (control) risk score of 0.252+0.241 (mean + std); p Discussion/Significance of Impact: We report the use of voice as a novel, noninvasive biomarker to detect aspiration risk using machine learning techniques. This tool has the potential to be used for the safe and early detection of aspiration in a variety of clinical settings including intensive care units, wards, outpatient clinics, and remote monitoring.

Objectives/Goals: Current popular scoring systems for evaluating facial nerve function are subjective and imprecise. This study aims to quantify speech and facial motor changes in patients suffering from facial palsy after cerebellopontine angle (CPA) tumor resection to lay the foundation for a scoring algorithm that is higher resolution and more objective. Methods/Study Population: We will obtain audio and video recordings from 20 adult patients prior to and after surgical resection of unilateral CPA tumors between October 2024 and February 2025. We will obtain preoperative recordings within two weeks prior to surgery and postoperative recordings following a preset schedule starting from the day after surgery up to one year. Audio recordings entail patient readings of standardized passages and phonations while video recordings entail patient performance of standardized facial expressions. We will analyze video data for key distance measurements, such as eye opening and wrinkle size, using DynaFace. We will process audio data using VoiceLab to extract metrics such as prominence and tonality. We will perform statistical tests such as t-tests and ANOVA to elucidate changes across time. Results/Anticipated Results: I expect to obtain 9 sets of audio and video recordings from each of the 20 participants. In terms of speech, I expect average speech duration to increase postoperatively. Similarly, I expect to find increases in time spent breathing, number of breaths taken, and mean breathing duration. In terms of facial movement, I expect nasolabial fold length to decrease postoperatively, as well as eye opening size and left-right symmetry at rest. For both audio and video, I expect these changes to revert towards their preoperative baseline as time passes. I also expect average House-Brackmann and Sunnybrook facial grading scores to increase postoperatively and then decrease with time, correlating strongly with the video and audio findings. I will use trajectory analysis and time point matching to handle any missing data. Discussion/Significance of Impact: This study will validate our analysis platform’s ability to automatically quantify measurable changes that occur to speech and facial movement which correlate strongly with existing scoring systems. Future work will synthesize these data streams to move towards establishing biomarkers for facial nerve function that aid clinical decision-making.

Objectives/Goals: Upon diagnosis, patients with acute myeloid leukemia (AML) have significant information needs. Given its recent increase in popularity, patients may use ChatGPT to access information about AML. We will examine the quality, reliability, and readability of information that ChatGPT provides in response to frequently asked questions (FAQs) about AML. Methods/Study Population: From FAQs on the top 3 patient-facing websites about AML, we derived 26 questions, written in lay terms, about AML diagnosis, treatment, prognosis, and functional impact. We queried ChatGPT-4o on 10/14/2024 using a new Google account with no prior history. We asked each question in a separate chat window once, verbatim, and without prompt engineering. After calibration, 5 oncologists independently reviewed ChatGPT responses. We assessed quality via the Global Quality Scale (GQS), scored from 1 (poor) to 5 (excellent) based on flow, topic coverage, and usefulness. For reliability, we assessed whether each response addresses the query and is factually accurate, elaborating on specific inaccuracies. For readability, we assessed Flesch-Kincaid Grade Level, Gunning Fog Index, and Simple Measure of Gobbledygook. Results/Anticipated Results: This will be a descriptive analysis of ChatGPT responses. For quality and reliability assessments, we will report Fleiss’ kappa for inter-rater reliability and expect substantial agreement or greater (≥0.61). Per prior studies in other domains, we hypothesize that ChatGPT responses will have good quality on average (i.e., GQS score near 4). We hypothesize that nearly all responses will address their query and will mostly be accurate; a minority of responses may have partial inaccuracies. Finally, we hypothesize that readability metrics will suggest that a higher educational level (e.g., college-level education) is required for comprehension. Overall, these findings will help elucidate strengths and limitations of ChatGPT for AML and guide discussion of factors patients should be aware of when using ChatGPT. Discussion/Significance of Impact: No prior study has examined the educational quality of ChatGPT for AML. Our study will detail whether patients are receiving trustworthy and meaningful information, identify misinformation, and provide guidance to oncologists when recommending information resources to patients or fielding questions that patients may raise after using ChatGPT.

Objectives/Goals: Translational researchers spend significant amounts of time finding available datasets and other research data resources for their purposes. Objectives of this program are develop and evaluate a multipronged approach to supporting researchers with existing data resources. Methods/Study Population: We established a dedicated service with expertise in data resources to increase awareness, understanding, and utilization of existing data resources. This program assists investigators and trainees discover appropriate data resources, formulate scientific problems in computable formats, advise on state-of-the-art data analytics, data management, build collaborations, mentor data users, and develop a service pipeline for streamlined data resource project management. This is accomplished through these essential functions: (1) Discover, catalog, document, and manage metadata resources, (2) train and present data resources to the research community, (3) provide individual consultations, and (4) explore and assess novel data resources. Results/Anticipated Results: In a phased approach, the data navigation program is performing outreach to the research community and integrating with existing data efforts on campus, presenting and demonstrating existing data resources, established a consultation service, and building core competencies into long-term usage and navigation of resources across campus. Evaluating the program monthly has shown an increase in various metrics for evaluating commitment and engagement including number of requests for access to data resource, consultations, publications and presentations, co-authorship, and proposals. Unawareness and inappropriate use of data resources leads to delays in performing research and potentially unnecessary duplications of efforts. Discussion/Significance of Impact: Our data navigation program has increased use of data resources in research. Next steps are to continue evaluation and further streamline informatics approaches to data discovery, abstraction, formulation, and analysis. Harmonized data resource programs are important translational science approach to foster the next generation of research.

Objectives/Goals: This study examines associations between childhood violence exposure, accelerated biological aging, and adolescent depression using DNA methylation-based epigenetic clocks. Findings aim to identify biomarkers for early detection, guide interventions, and address youth mental health disparities. Methods/Study Population:Data from the Future of Families and Child Wellbeing Study (N = 4,898), a diverse urban U.S. cohort, were analyzed. Childhood violence exposure, assessed using the Parent-Child Conflict Tactics Scale, included measures of physical, emotional, and psychological aggression and neglect. Biological aging at age 15 was evaluated using second-generation epigenetic clocks derived from saliva DNA methylation patterns, while depressive symptoms were measured with the CES-D scale. Multiple linear regression models tested associations between early violence exposure, epigenetic aging, and depressive symptoms, adjusting for socioeconomic status, caregiver mental health, and other key covariates. Results/Anticipated Results: Preliminary results suggest that early violence exposure may be linked to accelerated biological aging and depressive symptoms during adolescence, a critical developmental period. Epigenetic clocks offer an objective method for identifying high-risk youth, complementing mental health evaluations. With further validation and participatory action research, these findings could guide the development of biomarkers for longitudinal testing in school-based screenings and community health programs. These tools aim to be accessible, culturally relevant, and tailored to diverse populations, enhancing early detection, informing personalized interventions, and supporting scalable clinical applications. Discussion/Significance of Impact: This study explores links between early adversity, biological aging, and mental health, advancing understanding of adolescent depression. Epigenetic biomarkers could improve risk detection and guide tailored interventions in schools and community settings, enhancing access and reducing disparities.

Objectives/Goals: Understanding the interconnections among over 20,000 human diseases spanning organ systems could inform more precise diagnosis and treatment of diseases. Here, we examine whether the ability of large language models (LLMs) to learn universal representations of concepts can be leveraged to discover complex relationships across human diseases. Methods/Study Population: To address the challenge of computationally representing thousands diseases spanning multiple organ systems, we used internal representations of concepts by LLMs to encode diseases based on their descriptions from standard disease ontologies (ICD10 and Phecodes). To do this, we leveraged application programming interfaces (APIs) of three LLMs-GPT3.5, Mistral and Voyage to encode disease relationships. We then performed unsupervised clustering of the diseases using their encodings (embeddings) from each LLM to determine whether the resulting clusters reflect disease relationships. To enable deeper exploration of disease relationships, we developed interactive plots that provide a system level view of the relationships between thousands of diseases and their association with specific organ systems. Results/Anticipated Results: We found that unsupervised analysis of disease relationships using the LLM encodings reveal high similarities among diseases based on organ systems they affect. All the LLMs clustered diseases into groups largely defined by the organ systems they affect without being trained to specifically classify diseases into their corresponding organ system classification. An exception to this was tumors in which we observed that most tumors cluster together as a group irrespective of the organs they affect. Interestingly, we found that tumors affecting anatomically related organs show higher similarity to each other than to those affecting distantly related organs. In addition to anatomical relationships between diseases, we found that the LLM embeddings capture genetic relationships between diseases. Discussion/Significance of Impact: Overall, we found that the LLM-derived encodings uphold biologically and clinically significant relationships across organ systems and disease types. These results suggest that LLM encodings could provide a universal framework for representing diseases as computable phenotypes and enable the discovery of complex disease relationships.

Objectives/Goals: The standard care for malignant gliomas includes maximal tumor resection, but challenges arise near functional (speech) areas. Direct cortical stimulation (DCS) identifies functional (nonresectable) cortex. We aim to identify electrophysiologic (via subdural electrode recordings [ECOG]) biomarkers of DCS-positive (functional) areas. Methods/Study Population: Our lab maintains one of the largest datasets of electrophysiology analysis of glioma infiltrated brain cortex in the USA. Recordings of intraoperative brain mapping were analyzed to identify cortical sites that were found to be positive (functional) during DCS. DCS positive and negative (nonfunctional) sites were aligned to corresponding subdural electrodes. Future analysis: We plan to compare the temporal and spectral electrophysiologic variations associated with cortical sites found to be DCS positive versus negative during brain mapping. We plan to train machine learning classifiers that utilize these electrophysiologic biomarkers to discriminate between DCS positive and negative sites. Results/Anticipated Results: In total, our database comprised of 110 resections with brain mapping (DCS) and ECOG, including 4 patients who underwent a second procedure for resection. Eight patients were excluded as their resections were for brain metastases, not glioma. Our final cohort was comprised of 98 glioma resections, including 4 patients who underwent surgery twice for recurrence. During these resections, a total of 1393 sites were mapped via DCS for language function (including picture naming, word reading, and sentence syntax tasks). Of these 1393 sites, 100 sites were found to be DCS positive (7.1% positivity rate). (Currently in the process of conducting analysis comparing electrophysiologic features and biomarkers of DCS positive versus negative sites.) Discussion/Significance of Impact: This research is ongoing. Identifying electrophysiologic biomarkers of critical DCS-positive regions may provide a durable alternative to stimulation mapping. Due to its resource intensity, DCS has access barriers. Future neurosurgeons may use biomarkers from subdural electrode recordings to plan safer cortical resections.

Objectives/Goals: Neurodegenerative diseases involve progressive neuronal loss or dysfunction, often due to accumulated damage and impaired repair mechanisms. Our research evaluates the role of innate immune recognition proteins to provide insights into age-related neurodegeneration and cognitive decline. Methods/Study Population: We will utilize transcriptomic data from the Long-Life Family Study (LLFS), a cohort rich in genetic and phenotypic data related to aging and longevity. Our approach includes assessing a set of innate immune recognition proteins, also known as pattern recognition receptors (PRRs) expression across various age groups, focusing on potential correlations with cognitive performance. By analyzing serum transcriptomic profiles, we aim to map changes in expression and DNA repair genes over time, evaluating their connection to cognitive health and neurodegeneration in aging populations. Results/Anticipated Results: We anticipate that the expression of some PRRs will increase with age and correlate with cognitive decline, suggesting a role in age-related neurodegeneration. We also expect a decrease in DNA repair pathway gene expression in older age groups, contrasting with an increase in genes involved in endogenous DNA detection. These results will reveal how PRRs may function as neuroprotective factors and how their expression changes may relate to the decline in DNA repair processes with age, providing a better understanding of innate immune recognition in cognitive health. Discussion/Significance of Impact: This study will reveal the role of PRRs in aging and neurodegeneration, potentially establishing them as a key player in neuronal protection. Findings may guide future research into therapeutic strategies targeting them for Alzheimer’s and other age-related neurodegenerative diseases.

Objectives/Goals: Global surgical education is largely driven by high-income countries (HICs), with curricula not tailored to the needs of low- and middle-income countries (LMICs). This study assessed country-specific needs for global surgical curricula and used generative AI to develop tailored curricula. Methods/Study Population: A curriculum framework was developed using expert opinion. Using a focused needs assessment survey, we evaluated international medical students’ and trainees’ needs for structured global surgery curricula, covering research, education, data and develop tailored curriculum templates for each country, ensuring alignment with the distinct needs of respective LMIC and HIC respondents. The AI-generated curricula were then compared across countries to identify variations in content and focus areas. Results/Anticipated Results: A total of 145 respondents from 18 countries and 6 continents participated, with 94 from LMICs and 51 from HICs. Four countries [Uganda (n = 31), Nigeria (n = 34), the USA (n = 23), and the UK (n = 23)] had more than 10 respondents, with the creation of a country specific global surgery curriculum. Curricula developed by HIC trainees focused on access to resources and infrastructure, future directions of global surgical research, and the role of medical students and early career development with a decreased focus on the history of global surgery. LMIC country-based curriculum focused on introducing the concepts of global surgery, quantifying the burden and epidemiology of surgical disease and had a greater emphasis on case studies and use cases, with decreased focus on resources and collaboration. Discussion/Significance of Impact: The research introduces a “precision education” approach that could help close the surgical education access gap globally. Further pilot and qualitative studies are necessary to validate the feasibility of AI-generated needs-based curricula.

Objectives/Goals: We aim to enhance risk prediction in kidney transplantation outcomes by improving models of peptide antigen presentation of mismatched HLA molecules. HLA-derived peptides presented by HLA Class II to T-cells can activate an immune response, ultimately leading to graft failure. We aim to improve peptide prediction by modeling antigen processing. Methods/Study Population: T-cell epitope models for HLA mismatching struggle to predict which peptides are presented because antigen processing by proteases is not well modeled. We model antigen processing of HLA Class II proteins using 3D HLA structures (crystallography data) to create an HLA-specific antigen processing likelihood (APL) model. APL uses conformational stability measurements such as b-factor, COREX, solvent accessible surface area, and sequence entropy to predict cleavage sites from proteolysis. We will integrate APL into a T-cell epitope prediction tool for HLA-derived peptides based on donor and recipient HLA genotypes. Finally, we will associate the risk of graft failure with counts of these peptides derived from APL-integrated prediction models using a historical kidney transplant cohort from 2000 to 2023. Results/Anticipated Results: We expect that applying APL could reduce false-positive peptide binders influencing risk prediction scores. We anticipate improved peptide prediction accuracy compared to existing tools such as NetMHCIIPan, which assumes all possible peptides are equally likely to emerge from antigen processing. NetMHCIIPan is currently used by PIRCHE-II HLA mismatch risk algorithm. We expect that merging antigen processing (APL) and peptide-binding (NetMHCIIPan) models into a unified model would enhance risk stratification for graft failure. Current risk stratification still leads to poor outcomes post-transplant, especially for minority population groups. Our model can identify an alternative pool of well-matched donors and has the potential to improve equity for non-White minority candidates. Discussion/Significance of Impact: Improving the understanding of how HLA matching contributes to kidney transplant outcomes can better stratify risks for kidney transplant recipients, enable personalized treatment, and ultimately improve outcomes for those undergoing kidney transplantation to treat renal diseases.