Objectives/Goals: Predictive performance alone may not determine a model’s clinical utility. Neurobiological changes in obesity alter brain structures, but traditional voxel-based morphometry is limited to group-level analysis. We propose a probabilistic model with uncertainty heatmaps to improve interpretability and personalized prediction. Methods/Study Population: The data for this study are sourced from the Human Connectome Project (HCP), with approval from the Washington University in St. Louis Institutional Review Board. We preprocessed raw T1-weighted structural MRI scans from 525 patients using an automated pipeline. The dataset is divided into training (357 cases), calibration (63 cases), and testing (105 cases). Our probabilistic model is a convolutional neural network (CNN) with dropout regularization. It generates a prediction set containing high-probability correct predictions using conformal prediction techniques, which add an uncertainty layer to the CNN. Additionally, gradient-based localization mapping is employed to identify brain regions associated with low uncertainty cases. Results/Anticipated Results: The performance of the computational conformal model is evaluated using training and testing data with varying dropout rates from 0.1 to 0.5. The best results are achieved with a dropout rate of 0.5, yielding a fivefold cross-validated average precision of 72.19% and an F1-score of 70.66%. Additionally, the model provides probabilistic uncertainty quantification along with gradient-based localization maps that identify key brain regions, including the temporal lobe, putamen, caudate, and occipital lobe, relevant to obesity prediction. Comparisons with standard segmented brain atlases and existing literature highlight that our model’s uncertainty quantification mapping offers complementary evidence linking obesity to structural brain regions. Discussion/Significance of Impact: This research offers two significant advancements. First, it introduces a probabilistic model for predicting obesity from structural magnetic resonance imaging data, focusing on uncertainty quantification for reliable results. Second, it improves interpretability using localization maps to identify key brain regions linked to obesity.

The Students Participating as Ambassadors for Research in Kentucky (SPARK) program provides novel health equity research training and targeted mentorship for undergraduates, particularly those from groups underrepresented in the biomedical and behavioral research and workforce. SPARK aims to address inadequate diversity in the medical and scientific research fields by providing comprehensive research mentorship and skill-building. Unlike most existing research training programs that are brief, focus on laboratory research, or are limited to graduate students and junior faculty, SPARK delivers a 16-month intensive behavioral and population health science training, equipping students with needed tools to conceptualize, plan, execute, and analyze their own health equity research study. Trainees complete didactic coursework on health equity, study design and proposal development, data analysis, and ethics. Students receive a stipend and research expenses, and multiple mentors guide them in creating original research projects for which they serve as Principal Investigator. Students disseminate their findings annually at an academic research conference as a capstone. Evaluation data from the first three cohorts suggest SPARK has been pivotal in preparing students for graduate studies and research careers in health equity and behavioral and population health sciences, providing strong support for further investments in similar undergraduate research training models.

Objectives/Goals: Opioid use disorder (OUD) at delivery increased between 1999 and 2014. Clinical guidelines include medication for OUD (MOUD) for pregnant women with OUD and is associated with better fetal outcomes. Few large studies have compared prenatal MOUD outcomes to no MOUD. We evaluated the association of documented MOUD prescription during pregnancy with maternal outcomes. Methods/Study Population: We utilized aggregated electronic health records using the TriNetX platform to conduct a retrospective cohort study of females, aged 1249 years with a childbirth CPT code and documented opioid use via ICD-10 codes in the nine months before delivery between 2014 and 2020, comparing patients with MOUD prescription of buprenorphine or methadone during the nine months before delivery to demographically matched patients without any documented MOUD, using hazard ratios and 95% CIs for outcomes occurring one week to one or three years after childbirth. Results/Anticipated Results: MOUD cohort (n = 6,945, 85.33% White; 82.77% Non-Hispanic or Latino) was associated with significantly higher subsequent documented MOUD prescription (HR, 9.26 [95% CI, 7.98–10.76]; 6.21 [95% CI, 5.60–6.88]) and new remission codes (HR, 2.79 [95% CI, 2.15–3.62]; 2.85 [95% CI, 2.38–3.40]) at one and three years, lower ED visits at one year (HR, 0.88 [95% CI, 0.81–0.96]), no significant association of ED visits at three years (0.95 [95% CI, 0.89–1.02]), higher outpatient visits (HR, 1.26 [95% CI, 1.20–1.32]; HR, 1.27 [95% CI, 1.21–1.33], and no significant association of inpatient visits at one and three years (HR, 0.93 [95% CI, 0.813–1.06]; 1.06 [95% CI, 0.96–1.18]) than the never-MOUD cohort (n = 4,708, 76.11% White; 75.68% non-Hispanic or Latino). Discussion/Significance of Impact: A documented prescription for MOUD during pregnancy is associated with newly documented remission of OUD, increased outpatient visits, decreased ED visits, and additional documented MOUD prescriptions suggestive of increased access to continuity care. Efforts to increase MOUD use in pregnancy may improve maternal outcomes.

Objectives/Goals: Manual skin assessment in chronic graft-versus-host disease (cGVHD) can be time consuming and inconsistent (>20% affected area) even for experts. Building on previous work we explore methods to use unmarked photos to train artificial intelligence (AI) models, aiming to improve performance by expanding and diversifying the training data without additional burden on experts. Methods/Study Population: Common to many medical imaging projects, we have a small number of expert-marked patient photos (N = 36, n = 360), and many unmarked photos (N = 337, n = 25,842). Dark skin (Fitzpatrick type 4+) is underrepresented in both sets; 11% of patients in the marked set and 9% in the unmarked set. In addition, a set of 20 expert-marked photos from 20 patients were withheld from training to assess model performance, with 20% dark skin type. Our gold standard markings were manual contours around affected skin by a trained expert. Three AI training methods were tested. Our established baseline uses only the small number of marked photos (supervised method). The semi-supervised method uses a mix of marked and unmarked photos with human feedback. The self-supervised method uses only unmarked photos without any human feedback. Results/Anticipated Results: We evaluated performance by comparing predicted skin areas with expert markings. The error was given by the absolute difference between the percentage areas marked by the AI model and expert, where lower is better. Across all test patients, the median error was 19% (interquartile range 6 – 34) for the supervised method and 10% (5 – 23) for the semi-supervised method, which incorporated unmarked photos from 83 patients. On dark skin types, the median error was 36% (18 – 62) for supervised and 28% (14 – 52) for semi-supervised, compared to a median error on light skin of 18% (5 – 26) for supervised and 7% (4 – 17) for semi-supervised. Self-supervised, using all 337 unmarked patients, is expected to further improve performance and consistency due to increased data diversity. Full results will be presented at the meeting. Discussion/Significance of Impact: By automating skin assessment for cGVHD, AI could improve accuracy and consistency compared to manual methods. If translated to clinical use, this would ease clinical burden and scale to large patient cohorts. Future work will focus on ensuring equitable performance across all skin types, providing fair and accurate assessments for every patient.

Objectives/Goals: To address the manual, time-consuming processes of validating IRB compliance and ensuring the secure delivery of i2b2 data, this project automates compliance checks, streamlines Protected Health Information (PHI) access, and provides timely, secure data availability while reducing administrative burdens and non-compliance risks. Methods/Study Population: This project enhances the i2b2 application to automate compliance processes and facilitate secure data delivery through integration with REDCap. By linking i2b2 with the IRB system, the application performs automatic compliance checks for project requests, verifying GCP and HIPAA certifications, only allowing the release of IRB-approved PHI variables, safeguarding against unauthorized data access. Manual signatures confirm non-automated compliance processes. Once verified, the application automatically creates a REDCap project, assigns user access, and securely delivers data, ensuring compliance with HIPAA regulations. Results/Anticipated Results: The automated system successfully streamlined IRB compliance checks and data delivery for i2b2 requests. Validation of certifications like GCP and HIPAA, now occurs automatically, significantly reducing the risk of non-compliance. Personnel access to data is limited to IRB-approved PHI, ensuring data security and adherence to institutional standards. The integration with REDCap has reduced manual processes, cutting data request processing time to approximately 30 minutes. Researchers and administrative staff experienced a notable decrease in administrative burden, with faster, more efficient access to approved data while maintaining full compliance with IRB and HIPAA regulations. Discussion/Significance of Impact: The lessons learned can be adapted by institutions to improve compliance efficiency and reduce administrative overhead. Implementing similar automation of certification checks and data delivery, sites can enhance data security, minimize errors, and ensure faster, compliant access to research data.

Objectives/Goals: Endometrial cancer is one of the few cancers that has both a rising incidence and mortality rate. Molecular classification is becoming more important for the management of endometrial cancer but the ability to translate this into clinical practice remains constrained. Our goal is to use AI to predict the molecular subtype from histopathology slides. Methods/Study Population: We utilized the open source endometrial cancer datasets from The Cancer Genome Atlas (TCGA) (N = 387) and Cancer Proteomics Transcriptomic Tumor Analysis Consortium (CPTAC) (N = 135) to develop and train a vision transformer AI model. We used a proprietary cohort of patients (N = 548) for external validation. Whole slide images (WSI) and molecular subtype data were collected. Subtypes include POLE ultramutated (POLE), microsatellite instability (MSI-H), copy-number low (CNV-L), and copy-number high (CNV-H). WSI were preprocessed, and features were extracted. Modified STAMP protocol was used in training, utilizing a pretrained foundation transformer model (Virchow2). Cross-validation of the TCGA was used for initial training, followed by testing on the CPTAC dataset and validation on our proprietary cohort. Results/Anticipated Results: Fivefold cross-validation of the TCGA database (60% training, 20% testing, and 20% validation) developed a best overall model with a mean AUC of 0.74 (POLE 0.78, MSI-H 0.76, CNV-H 0.86, CNV-L 0.77). Overall precision 0.58, recall 0.55. CNV-H was the subtype with the most accurate prediction. CPTAC holdout testing revealed moderately high AUC (POLE 0.63, MSI-H 0.62, CNV-H 0.98, and CNV-L 0.76). Overall precision 0.54 and recall 0.58. Again, CNV-H was the most accurate prediction. Validation on our proprietary cohort revealed a drop in performance with overall mixed results by AUC (POLE 0.50, MSI-H 0.69, CNV-H 0.78, and CNV-L 0.61). Overall precision 0.57, recall 0.45. Again, CNV-H with the most accurate prediction but F1 score dropped from 0.77 in the CPTAC to 0.47 on validation. POLE was the least accurate prediction subtype. Discussion/Significance of Impact: The CNV-H subtype demonstrated robust performance, suggesting the model effectively captures the features associated with this subtype. CNV-L had moderate performance. MSI-H and POLE were notably lower. WSI-based AI models show translational potential for subtype prediction in the management of endometrial cancer but more work is necessary.

Extant literature reveals how patients of marginalized social identities, socioeconomic status (SES), and medical experiences – especially patients of color and older adults – are underrepresented in cancer clinical trials (CCTs). Emerging evidence increasingly indicates CCT underrepresentation among patients of lower SES or rural origin, sexual and gender minorities, and patients with comorbid disability. This review applies an intersectional perspective to characterizing CCT representativeness across race and ethnicity, age, sexual and gender identity, SES, and disability. Four databases were systematically queried for articles addressing CCT participation inequities across these marginalizing indicators, using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. One hundred one articles were included in a qualitative evaluation of CCT representativeness within each target population in the context of their intersectional impacts on participation. Findings corroborate strong evidence of CCT underrepresentation among patients of color, older age, lower SES, rural origin, and comorbid disabling conditions while highlighting systemic limitations in data available to characterize representativeness. Results emphasize how observed inequities interactively manifest through the compounding effects of minoritized social identity, inequitable economic conditions, and marginalizing medical experiences. Recommendations are discussed to more accurately quantify CCT participation inequities across underserved cancer populations and understand their underpinning mechanisms.

Objectives/Goals: This pilot study aims to assess the implementation of the DataDay app in memory clinics for patients with MCI or dementia, focusing on usability, user satisfaction, and impact on health outcomes. We seek to identify barriers and facilitators to implementation and evaluate its effect on reducing unnecessary hospital stays. Methods/Study Population: This mixed-methods study will involve 50 participants, 25 diads of patients with MCI or mild-to-moderate dementia and their caregivers from the community. Participants will use DataDay for 12 weeks, receiving reminders to log daily activities such as nutrition, mood, cognition, and physical activity. Baseline demographic data will be collected from self-reported surveys. Participants will receive training on app use, with follow-up interviews at 4, 8, and 12 weeks to gather feedback. Quantitative data analysis will include repeated measures analysis of variance to compare pre- and post-intervention outcomes, such as medication use and ER visits. Thematic analysis will be conducted on interview transcripts to understand user experiences. Results/Anticipated Results: We anticipate the study will demonstrate the feasibility of the DataDay app for self-management in individuals with MCI or dementia. Expected outcomes include improved medication adherence, reduced emergency room visits, and increased user engagement with daily health monitoring. Qualitative feedback is expected to highlight user satisfaction with the app’s reminders and ease of integration into daily routine. We also expect potential challenges to be identified such as initial learning difficulties and technology-related frustration. The data will help refine the app for better usability and inform strategies for widespread implementation in memory assessment clinics. Discussion/Significance of Impact: The study will provide insights into the practicality of implementing DataDay in memory clinics. The results will highlight necessary adjustments and provide key factors for successful adoption in other clinics. DataDay aims to allow individuals with MCI or dementia to manage their condition at home and enhance their quality of life.

Objectives/Goals: Ischemic stroke treatments assist in restoring blood flow, but do not guarantee good outcomes. Since extracellular vesicles (EVs) able to cross the blood brain barrier, total (nonspecific) and astrocyte enriched EVs (TEVs, AEVs, respectively) from plasma may emerge as plasma biomarkers for prognostication and targeted therapeutics. Methods/Study Population: “Blood and Clot Thrombectomy Registry and Collaboration” (BACTRAC; NCT03153683) is a human stroke biobank at the University of Kentucky that collects samples at the time of mechanical thrombectomy during emergent large vessel occlusions (ELVO; ischemic stroke). EVs were isolated, via size exclusion chromatography, from unbanked plasma and concentrated resulting in TEVs. AEVs were immunoprecipitated with anti-EAAT1 (GLAST), an astrocyte-specific transmembrane glycoprotein. Isolated protein was sent to Olink and ran on their metabolic panel. Demographics and medical histories of the subjects were exported from REDcap and investigators were blinded during EV analysis. Results/Anticipated Results: ELVO subjects (8 females/ 5 males) were an average age of 71.1 ± 11.7 years. Lower TEV enolase 2, a neuronal glycolysis enzyme, associated with increased stroke severity (NIHSS; rs =  -0.7819, p = 0.0476). Higher systemically TEV quinoid dihydropteridine reductase (QDPR), essential co-factor enzyme, was associated with more severe strokes (NIHSS; rs =  0.8486, p = 0.0123) and lower cognition (MoCA; r2 =  0.7515, p = 0.0254). Interestingly, higher intracranial AEVs QDPR was associated with lower infarct volumes (rs =  -0.7333, p = 0.0202), less severe strokes (NIHSS; rs =  -0.6095, p = 0.0388), and better cognition (MoCA; r2 =  0.6095, p = 0.0388). Increased AEV nicotinamide adenine dinucleotide kinase another essential co-factor enzyme, intracranially also correlated to higher cognition (MoCA; rs =  0.8356, p = 0.0298). Discussion/Significance of Impact: Plasma TEV and AEV metabolic proteins correlate with the progression of stroke outcomes and should be investigated as target therapies during MT to improve outcomes.

Objectives/Goals: Pharmacogenomic (PGx) testing identifies genetic variations affecting medication response but is not yet in routine clinical whole-genome sequencing (WGS) workflows. We aimed to establish a streamlined bioinformatics pipeline for incorporating PGx reporting into clinical WGS and to determine clinical implications for medication treatment. Methods/Study Population: A PGx profiling pipeline based on existing WGS data was developed, integrating three WGS-based PGx calling tools: Aldy, PyPGx, and Cyrius (CYP2D6 only), to provide genotype calls for 17 key pharmacogenes. The pipeline was validated using WGS data from 70 individuals with diverse backgrounds (36% European, 27% African, 27% Asian, and 10% admixed) from the Genetic Testing Reference Materials Coordination Program (GeT-RM). Results were manually reviewed against published data. The validated pipeline was then applied to 144 clinical patients previously screened for neurodevelopmental disorders or suspected hereditary diseases, followed by diplotype-to-phenotype translation and preemptive PGx-guided medication recommendations based on consensus guidelines and FDA labeling for commonly used medications. Results/Anticipated Results: Congruent phenotype call rates for GeT-RM samples were 100% for 13 genes (CFTR, CYP2B6, CYP2C19, CYP2C9, CYP3A4, CYP4F2, DPYD, G6PD, IFNL3, NAT2, NUDT15, TPMT, and VKORC1), 99% for three genes (CYP3A5, SLCO1B1, UGT1A1), and 97% for CYP2D6, indicating strong pipeline performance. Among 144 clinical patients, 99.3% had at least one clinically actionable PGx results relevant to 36 of top 300 medications in the USA across psychotropic, cardiovascular, musculoskeletal, gastrointestinal, and other therapeutic areas. The most prevalent drug–gene interactions involved sertraline and CYP2B6, affecting 49% patients: 41% were intermediate metabolizers who may require slower titration and lower maintenance doses, while 8% poor metabolizers may benefit from a lower starting dose or alternative antidepressants. Discussion/Significance of Impact: Our validated WGS-based PGx profiling pipeline successfully extracted actionable PGx data from clinical WGS. By aligning PGx profiles with guideline-recommended clinical actions, we demonstrated the clinical value of integrating PGx reporting in WGS workflows, improving personalized medication management.

Objectives/Goals: Many left hemisphere stroke survivors have a reading disorder (alexia), which is experienced as decreasing well-being. Therapies produce inconsistent results, demonstrating a need for treatment response predictors. We identify neural correlates of a computational model of reading, which may provide biomarkers to improve therapeutic outcomes. Methods/Study Population: Left hemisphere stroke survivors (LHSS) (n  =  52) performed an oral reading task and tests of semantic and phonological processing. Artificial neural network (ANN) models, mapping between orthography (visual word form), phonology (auditory word form), and semantics (word meaning), were trained to read single words at an adult reading level. Stroke was simulated by removing percentages (in 10% intervals) of the connections into and out of semantics, phonology, and the combination thereof. The lesioned model producing the smallest average Euclidean distance over word and pseudoword reading accuracy to each LHSS was selected as the matched model. Two voxelwise lesion-symptom mapping (VLSM) analyses identified the neural correlates of the percent of phonological and semantic links removed in the matched models. Results/Anticipated Results: Model reading was correlated with LHSS reading (high-frequency regular words, r(48) = 0.96; high-frequency irregular words, r(48) = 0.94; low-frequency regular words, r(48) = 0.97); low-frequency irregular words, r(48) = 0.85; all p’s Discussion/Significance of Impact: Our results show that ANN models of reading, when closely matched to LHSS reading performance, directly connect cognitive processes to the brain. Using matched models as a precision medicine framework to predict therapy response or to identify targets for neurostimulation provides a valuable route toward improving poststroke language outcomes.

Objectives/Goals: Diabetic kidney disease (DKD) affects 40% of diabetic patients, leading to renal failure, yet the molecular drivers remain elusive. MicroRNAs, noncoding regulators of gene expression, may hold the key. This study aims to identify key miRNAs in DKD, providing crucial insights for early intervention. Methods/Study Population: miRNA sequencing was conducted on kidneys from 8-week old male BTBR wild type and BTBR ob/ob mice. BTBR ob/ob mice lack the hormone leptin and spontaneously develop type 2 diabetes, with morphological renal lesions characteristic of human DKD. Total RNA was extracted from whole kidney sections and processed using the QIAseq miRNA library kit. Sequencing was performed on an Illumina NextSeq 550 platform. GeneGlobe analysis was used to identify differentially expressed miRNA functional pathways, while ingenuity pathway analysis (IPA) was employed to predict master regulators and causal networks involved in DKD. Results/Anticipated Results: miRNA sequencing identified significantly differentially expressed miRNAs (p < 0.05) between 8-week-old BTBR WT and BTBR ob/ob male mice, including miR-34a (-6.86 fold), miR-122 (-5.01 fold), miR-129 (-2.23 fold), miR-142a (+2.78 fold), miR-346 (+4.66 fold), miR-547 (-2.49 fold), miR-592 (+11.81 fold), miR-802 (-6.95 fold), and miR-6539 (-7.93 fold). Qiagen GeneGlobe analysis revealed biological processes potentially targeted by these miRNAs, including endocytosis, phagocytosis, hyperglycemia (p = 7.59e-3), and insulin-dependent diabetes (p = 4.32e-4). IPA predicted activation of RRAS, a small GTPase regulating cell growth and signaling (Z-score +2), with miR-34a and miR-122 targeting MYC, PI3K, and TGF-β in DKD progression in BTBR ob/ob mice. Discussion/Significance of Impact: We identified kidney miRNA expression in BTBR ob/ob mice at a pivotal disease stage. miR-34a, miR-122, and RRAS emerged as key drivers in DKD progression, showing remarkable early biomarker potential. These findings lay the groundwork for early detection and innovative therapies to halt DKD and improve patient outcomes.