Objectives/Goals: The scientific workforce has seen a trending decrease in future clinicians pursuing research over the last several decades. This study assessed effectiveness of a mentored research program to increase research skills and future research interest among medical students. Methods/Study Population: The Medical Student Research Program (MSRP) at UCIis designed to provide research training to medical students beginning their first year of medical school until graduation. All students applying to enroll in the MSRP complete a baseline survey prior to program enrollment and an annual survey yearly afterward. Baseline surveys collected students’ self-rated confidence in research skills, along with their future intent to use research as a clinician and to pursue a future research career after medical school. Annual surveys tracked change over time in students’ research competencies and their future intent to use research. Results/Anticipated Results: Annual surveys (N=128) assessed the change from baseline to one year in research competencies and students’ intent to use research in clinical practice (mixed design ANOVA) and in their intent to conduct research in their career. Evaluations indicated improved medical student research skills, with a significant increase in student research competencies across time (F(1, 173) = 25.4, p<0.001) found for both non-enrolled (M=3.76) and enrolled (M=4.07) MSRP students, with MSRP students reporting higher mean research competencies at one year. Yet similar to national trends, there was a decrease in students’ intentions to use research in their clinical career at one year (F(1, 173) = 25.4, p<0.001). This trend was more pronounced in students no longer enrolled in MSRP (F(1, 173) = 4.7, p = 0.006). Discussion/Significance of Impact: Preliminary findings demonstrate enhanced student research competencies during the first year of the program. However, the decline in students’ intentions to use research in the future highlights a need for continued investigation into the factors driving this decline and solutions to prevent decline in the research workforce.

Objectives/Goals: Pediatric ocular hypertension (OHTN) carries a variable risk of glaucoma progression, yet its progression to true glaucoma is not well-defined. This study characterizes the clinical course of a cohort of pediatric eyes with OHTN and identifies predictors to glaucoma conversion. Methods/Study Population: A retrospective review of 119 eyes from 81 pediatric glaucoma suspects (IOP >21 mmHg on ≥2 visits) between 2005–2016. Conversion to glaucoma was defined by the Childhood Glaucoma Research Network criteria (progressive optic nerve cupping, visual field loss, or other structural changes associated with elevated IOP). Demographics, comorbidities, and tonometry methods were analyzed. Logistic regression identified independent risk factors for glaucoma conversion. Thus far, analyses include data from the 2005–2016 cohort, with additional data from the 2017–2023 patient cohort under collection and expected to be ready by the conference date. Results/Anticipated Results: Of 119eyes (age 9.4±4.7years; baseline IOP27.7±5.7mmHg), 47(39.5%) converted to glaucoma over average 1.9years.Among 72 non-converters, 23(31.9%) normalized without treatment, 17(23.6%) with treatment, 17(23.6%)remained elevated untreated, and 15(20.8%) with treatment. Conversion was most frequent with baseline IOP 24–28mmHg(16.0%)and less with <24mmHg(8.3%). Absence of ocular (11.8%vs50.6%) and systemic risk factors (36.0%vs48.5%) reduced conversion, lowering treatment need and improving normalization. On multivariable analysis, younger age (OR0.84/year, p=0.002), ocular risk factors (OR4.95,p=0.01), and time to therapy initiation (OR0.74/year, p<0.001) predicted conversion. Discussion/Significance of Impact: This study establishes baseline age and ocular risk factors as strong predictors of glaucoma conversion in pediatric ocular hypertensives. These findings are critical for clinicians, directly guiding an evidence-based approach to patient risk stratification, surveillance frequency, and treatment initiation.

Objectives/Goals: We advance multimodal modeling, including digital twins (DTs), within an AI-enabled Learning Health System (LHS) to turn healthcare data into insights for precision care and translational science. DTs are dynamic patient models integrating EHRs, neuroimaging, biomarkers, wearables, and other real-world data (RWD) tools. Methods/Study Population: Multiple sclerosis (MS) provides a compelling use case of heterogeneous manifestations including fatigue, mobility deficits, cognitive deficits, and sleep disturbance, requiring multimodal integration to capture complexity. DTs integrate EHR, neuroimaging, biomarkers, wearable, and other RWD via ETL pipelines using standards like FHIR, OMOP, and DICOM. Modeling uses ensemble learning, LSTM/DBNs, dimensionality reduction, and bias auditing. Explainable AI (XAI) provides transparency through counterfactual “what if” analyses, such as testing whether risk changes if a patient improves gait speed, adjusts therapy, increases sleep duration, or lowers biomarker levels. Continuous monitoring detects model drift, ensuring reliability per FDA V3: verification, validation, vigilance. Results/Anticipated Results: A robust extract–transform–load (ETL) pipeline maps these inputs into standards such as Fast Healthcare Interoperability Resources (FHIR), the Observational Medical Outcomes Partnership (OMOP) common data model, and the Digital Imaging and Communications in Medicine (DICOM) format, ensuring interoperability and reuse across sites. Development emphasizes robustness through parameter optimization, ensemble learning, prevention of data leakage, preprocessing, dimensionality reduction, and regularization. Key AI applications include MRI lesion and atrophy segmentation with nnU-Net, temporal deep learning for longitudinal forecasting with deep belief networks (DBNs) and long short-term memory (LSTM) models, ensemble modeling to improve generalizability, and bias auditing to promote fairness. Discussion/Significance of Impact: This initiative advances digital health, neuroinformatics, and precision medicine for MS and other complex diseases. Aligned with NIH ODSS Data COUNTS and FDA V3, the framework supports discovery, regulatory science, and generalization to other variable conditions.

Objectives/Goals: To identify drivers of publication impact and output in surgical AI across countries and assess whether cross-border collaboration, especially between developed and developing nations, is associated with higher citations and knowledge diffusion over 2000–2025. Methods/Study Population: We queried PubMed EDirect for surgical AI articles (2000–2025) and linked Crossref citations. Papers were grouped by collaboration type: single developed, single developing, developed–developed, developed–developing, and developing–developing. Success metrics included citations/paper, total papers, and collaboration rates. Country-level covariates from World Bank (infrastructure, economics, R&D, technology, education) were integrated. Analyses: chi-square(counts), Kruskal–Wallis (citation distributions), ANOVA (means), and Pearson correlations (predictors of citations and production). Results/Anticipated Results: Developed–developing collaborations had the highest citation impact (21.3 citations/paper) vs developed–developed (21.1), single developed (20.1), developing–developing (13.2), and single developing (8.7) (all p<0.001). In developing nations, citation impact correlated most with healthcare infrastructure – hospital beds (r=0.896), physicians/1,000 (r=0.891), and health expenditure (r=0.883) (all p<0.001). Paper production correlated most with economics and R&D—GDP (r=0.991), researchers in R&D (r=0.997), and R&D spend (r=0.989) (all p<0.001). Technology access (mobile subscriptions, r=0.874) and tertiary enrollment (r=0.871) also tracked with citation impact. Discussion/Significance of Impact: Cross-border collaboration, especially developed–developing, maximizes impact and advances equitable knowledge transfer in surgical AI. Policies should fund international partnerships, strengthen infrastructure in developing countries, and build networks to improve global diffusion of innovation.

Objectives/Goals: This project evaluates the acceptability to patients of offering free tax preparation and joint charity-care enrollment in a clinical setting, leveraging an Internal Revenue Service-supported free income tax preparation program. The goal is to reduce administrative barriers and promote continuous access to care. Methods/Study Population: In September 2025, three members of our study team conducted in-person surveys with patients and visitors in outpatient waiting areas at Parkland Health, the county safety-net health system serving Dallas County in North Texas. The 16-item bilingual (English/Spanish) survey was developed in partnership with a large, Internal Revenue Service-supported, local VITA program, Dallas Community Tax Centers (DCTC). The survey collected sociodemographic information, household income, tax-filing practices, current charity care program enrollment status, and interest in combined VITA–charity care enrollment/renewal services. Data were entered into REDCap and analyzed descriptively in Excel. Results/Anticipated Results: Of 300 approached patients, 232 completed the survey, yielding a response rate of 77% (28% Spanish-preferring). Most respondents (86%) estimated that their household income was below $67,000, which would qualify them for VITA tax preparation, yet 82% of those eligible reported that they paid a tax preparer last year. More than half (54%) of respondents were enrolled in the charity care program for the county health system, and 87%of these adults indicated that integrated tax-and-charity care enrollment/renewal services would be helpful. Awareness and reported uptake of the Earned Income Tax Credit (EITC) was low: only 11% of respondents reported applying for EITC during the recent tax season. Interest in shifting to free on-site tax prep services was high, with 72% of respondents indicating a favorable response. Discussion/Significance of Impact: Embedding tax preparation and streamlining charity-care assistance applications in clinical environments was well received and highly acceptable. Future steps include piloting joint VITA–charity-care applications services to assess impact on operational workflows, uptake of both programs, continuity of care, and health outcomes.

Objectives/Goals: The “Reducing Disability in Alzheimer’s Disease” (RDAD) intervention promotes physical function for people with dementia and reduces caregiver burden but has never been tested in adults with Down syndrome despite their 90% lifetime risk of Alzheimer’s disease. Our objective is to modify RDAD for the needs of this population using the ADAPT framework. Methods/Study Population: We worked collaboratively with families, self-advocates, and professionals to conduct a 2-phase adaptation process of the RDAD intervention materials, content, and mode of delivery. In Phase 1, we convened focus groups comprised of adults with Down syndrome, caregivers, community-based disability service professionals, and researchers to review and advise changes to the RDAD intervention and analyzed their feedback qualitatively using content and thematic analysis. In Phase 2, we conducted a 4-week usability pilot test of RDAD with 5 older adults with Down syndrome and their caregivers to gather additional feedback on intervention feasibility, usability, and acceptability using weekly surveys and a final interview. Results/Anticipated Results: In Phase 1, twelve stakeholders met for four 90-minute focus group sessions. Based on their feedback, we modified the mode of delivery (remote delivery) and changed from one-on-one to group classes to support social connectedness. In the revised materials, we addressed the broad resource needs around dementia diagnosis, healthcare, and caregiving. We updated intervention materials to make them more usable and attractive and integrated music into the live, remote exercise classes, which include a person with Down syndrome as an instructional assistant. In Phase 2, we tested the revised intervention for 4 weeks and collected preferences for the exercise classes, content for the caregiver training, and modes of delivery. We have integrated these findings into the revised intervention and renamed it CareFit-DSAD. Discussion/Significance of Impact: We adapted an evidence-based intervention to the needs of families with Down syndrome using input and evaluation by key stakeholders. The adapted intervention is now undergoing a 12-week pilot and feasibility test (n = 20 dyads) to further assess feasibility, acceptability, and preliminary changes in physical function and caregiver burden.

Objectives/Goals: To identify and characterize real-world AI error points in vascular imaging, and then develop and use curie8, a large language model (LLM)-based data pipeline, to curate a broad, multimodal dataset for COMPASS – an AI stress-testing framework to define safety boundaries for clinical deployment. Methods/Study Population: We analyzed a pilot dataset of CT pulmonary angiography (CTPA) studies processed by a commercial AI tool for pulmonary embolism detection. AI errors were categorized using a radiologist-defined clinical and technical taxonomy to identify factors which affect AI performance. In parallel, curie8 was developed as an institutional LLM pipeline to query the radiology report text database and automatically link corresponding DICOM medical images and metadata. Together, the pilot dataset and taxonomy will inform the design of a full curie8-curated stress-test CTPA real-world dataset across the health system’s nine imaging sites (Figure 1). 2025-10-20 ACTS Fig and Table.pptx [https://somumaryland-my.sharepoint.com/:p:/g/personal/fdoo_som_umaryland_edu/EXPxiwh3cxhHpnWYLglMxK4BaByWvFivJq5p9ho5Csm4GA?e=ZPCrAC] Results/Anticipated Results: Pilot analyses (n=5,923) showed high overall performance of the commercial AI PE detection tool (sensitivity 89.6%, specificity 98.9%, PPV 88.6%, NPV 99.0%, accuracy 98.1%). AI errors were a small fraction, but had characteristic features – such as small embolus false negatives and artifact-driven false positives – influenced by patient and scan characteristics (Table 1) and technical factors (Table 2). These findings will guide thecurie8 pipeline in building a full multimodal vascular imaging dataset for benchmarking AI models under real-world stressors. Expected deliverables include a harmonized stress-test framework for vascular imaging AI safety evaluation. 2025-10-20 ACTS Fig and Table.pptx [https://somumaryland-my.sharepoint.com/:p:/g/personal/fdoo_som_umaryland_edu/EXPxiwh3cxhHpnWYLglMxK4BaByWvFivJq5p9ho5Csm4GA?e=ZPCrAC] Discussion/Significance of Impact: COMPASS combines institutional LLM-driven data curation (curie8) with structured AI error analysis to build standardized stress-testing resources. This framework advances reproducible, regulatory-aligned evaluation of AI safety and reliability in clinical imaging settings.

Objectives/Goals: To observe mechanisms of natural supplements and their links and similar pathways to chemotherapy and hormone therapy in order to use natural therapies as a way to ease treatment side effects in breast cancer patients. Methods/Study Population: A systematic search was conducted using databases such as PubMed and Google Scholar, using search terms “breast cancer,” “treatment,” “pathways,” “targets,” and “cell lines” in combination with “natural therapies” as well as “hormone therapies” and “chemotherapies” to understand drug and supplement mechanisms. ClinicalTrials.gov was used to investigate studies using natural therapies used to treat side effects of cancer treatments. Promising results were discovered on the effectiveness of natural therapies and their correlation with treatment for breast cancer. Results/Anticipated Results: Preclinical evidence supports 23 natural remedies being used to ease side effects brought on by chemotherapy. These natural supplements are shown to affect at least one of the seven breast cancer cell lines observed. Fifty-two total clinical trials were used to verify the preclinical models of the 23 natural remedies, and the clinical trials would be patient symptom assessment-based, or determining tumor growth or shrinkage. Many of these clinical trials, however, did not supply results. This shows that there is a gap in getting participants for oncology-related clinical studies. With the trials that supplied results, 16 natural remedies have clinical data showing their ability to be used to ease common chemotherapy side effects, or even further leading to cancer death in related cell lines. Discussion/Significance of Impact: There are not many specific findings about young women and post-menopausal women being affected by breast cancer (despite being more sensitive groups). Looking into differences in biology, reactions to treatment and effectiveness of treatment, and clinical trials with supplements between these groups could address issues affecting them.

Objectives/Goals: Broad and shallow data from large national cohorts provide generalizable real-world insights, while deep and narrow data from smaller cohorts capture detailed multimodal measures. This project integrates both approaches to study cognitive decline in Parkinson’s disease, which affects up to 50% of patients. Methods/Study Population: This longitudinal study includes validated digital biomarkers of motor, non-motor, sleep, and driving activity in 150 participants, plus novel social connectedness measures in 150 PD – partner dyads. PD and cognitive status are assessed annually; 4 weeks of continuous real-world behavior via actigraphy, social connectedness, and driving sensors are collected biannually. Anchor variables from pilot PD, PPMI, and LongROAD enable harmonization and cross-cohort modeling. Model development focuses on generative time series models that can capture the joint distribution of multimodal time series data for predicting values for future time steps. Together combined with genetics, plasma, and neuroimaging, they enable digital twins to predict decline and identify modifiable risks. Results/Anticipated Results: This project is designed to make impactful contributions to scientific knowledge, technical capability, clinical practice, and healthcare equity. Addressing key gaps in understanding PD cognitive decline, developing clinically validated digital biomarkers, and advancing data modeling for public health and precision medicine, including the creation of Digital Twin models. By integrating real-world data and social context into PD research, we will enhance clinical trials, improve patient outcomes, and provide clinicians with valuable tools, leading to a deeper understanding of PD and ADRDs. The data generated will be mined and shared with colleagues for years, driving new research and insights. Discussion/Significance of Impact: This project enhances monitoring and prediction of cognitive decline in PD, developing clinically valid, interpretable tools for real-world health. This generalizable platform unifies digital, imaging, and biomarker data as a model for translational research, advancing trial readiness across neurological and chronic diseases.

Objectives/Goals: Colorectal cancer risk varies across populations. Bile acid (BA) composition and the gut microbiome can impact CRC risk, but their differences across populations is poorly understood. We characterized serum BA and gut microbiome profiles in non-Hispanic Black and White Americans and examined associations with diet and social determinants of health. Methods/Study Population: To characterize BA composition, participants from the Chicago Multiethnic Prevention and Surveillance Study with a banked serum sample (n=100) were selected, with a subset of participants having completed a diet recall survey (ASA24; n=23). Sixty-two BA (primary, secondary, and glyco/tauro-conjugated BA) from serum samples were analyzed using liquid chromatography-mass spectrometry. For microbiome profiling, normal colonic biopsy samples from non-cancer participants were obtained from the Digestive Disease Research Core Center’s Integrative Clinical and Biospecimens Core (University of Chicago; n=109). Following DNA extraction, the mucosally adherent microbial community was assessed by 16S rRNA sequencing. Results/Anticipated Results: Of the BA detected in over 80% of the samples, glycochenodeoxycholic acid (GCDCA) and deoxycholic (DCA) had the highest median concentrations (56.7 and 55.0 µg/mL, respectively). Moderate positive correlations between DCA/isoDCA and calories, fat, and cholesterol were observed. The linear regression model employed to understand the relationship between log-transformed DCA and race, age, sex, and area deprivation index highlighted age as a key predictor of DCA level when controlling for other covariates. Microbiome analysis revealed that diversity was not different when comparing sexes, races, or colonic anatomic locations. At the phylum level, taxa belonging to the Bacillota and Bacteroidota phyla were common across samples. Discussion/Significance of Impact: These data and subsequent analysis, including additional BA analysis, improvement of the regression model, and deeper analysis of microbiome community related to BA metabolism, will provide additional context to the role of social and biological factors in better understanding differences in CRC risk and pathogenesis across human populations.

Objectives/Goals: To develop and implement a new course to fill an identified gap in the present IU School of Medicine Clinical Research Staff Education program, enhance data accuracy, and minimize audit findings. The goal was to provide a robust and thorough training on study orientation, source documentation, and adhering to ALCOA+ principles to clinical research staff. Methods/Study Population: A 90-minute presentation titled Best Practices for Study Documentation, or Level 1.5 because it fills the gap between the currently offered Level 1 and 2 courses, was developed by monitoring team members with input from the Director of Research Staff Education. A pilot session was presented to senior research staff for feedback prior to launch. The 90-minute presentation was designed with interactive polls to assess attendees’ research and documentation experience at the start and to maintain engagement throughout the session. Attendees receive a post-course evaluation survey along with a certificate of attendance. The course is offered every two months. Completion of Level 1 training is a prerequisite and aimed at those who are three to six months of starting in research at Indiana University School of Medicine. Results/Anticipated Results: Results from the post-course evaluation survey described Best Practices in Study Documentation as engaging, informative, and valuable for new and experienced research staff. Many learners reported increased confidence, improved understanding of documentation and regulatory requirements, greater awareness of the tools and resources available and appreciation for real-world examples from the instructors. While attendees valued the presentation’s clarity and interactive polls, they suggested leaving more time for questions, supporting reference materials, and an ability to watch a recording of the session. Overall, feedback has been highly positive, with learners emphasizing the relevance, practicality, and effectiveness of the presentation. Discussion/Significance of Impact: Best Practices for Study Documentation has been delivered nine times to 235 attendees since its inception in Jun 2024, 63% of learners report they have < 2 years’ research experience. Evaluation feedback is positive; however, no quantitative method has been identified to measure improvement in documentation quality or reduction in audit findings.

Objectives/Goals: Our objective in this study was to develop a nanoparticle-based RNA delivery system that reprograms Müller glia toward neuronal fates in human relevant models, enabling localized and transient regeneration without viral vectors. Methods/Study Population: - Results/Anticipated Results: Fabricated nanoparticles encapsulating RNA cargoes were ~110–150 nm in size with polydispersity indices of ~0.2–0.3 and encapsulation efficiencies >95%. Nanoparticles successfully delivered GFP mRNA in both organoid and retinal explant models, showing significant expression of fluorescent reporters in both live cell imaging and in slides prepared using immunohistochemistry. We anticipate that our delivery system will lead to transcription factor expression after delivery of corresponding RNAs with evidence of proliferation and neuronal marker upregulation. We also hypothesize that this approach will enhance neurogenesis in retinal organoids and explants, potentially establishing a platform for non-viral retinal regeneration. Discussion/Significance of Impact: This work demonstrates a clinically translatable, virus-free strategy for retinal repair through targeted RNA delivery. By enabling transient expression of neurogenic factors, nanoparticle-mediated reprogramming presents a scalable and adaptable approach for restoring vision in degenerative retinal diseases.

Objectives/Goals: * To identify community priorities and compare alignment with research initiatives * To connect research(ers) to community groups or members with similar priorities * To identify areas requiring outreach and/or collaboration * To build capacity and to catalyze new meaningful CTR partnerships * To evaluate ongoing collaborations for sustainability. Methods/Study Population: Together with our Medical School and Hospital, our Institute for Clinical and Translational Research (ICTR) led the development of a Community Engagement Tracker (CE Tracker). This tool tracks organizations, events, programs, and individual contacts to provide insight into current and past collaborations and capacity with the goal of building capacity and catalyzing new research partnerships with community members and organizations. We implemented a 6-stage development process involving: 1) internal stakeholder engagement, 2) initial development, 3) collaborator feedback, 4) refinement, 5) ongoing use, and 6) further refinement. Results/Anticipated Results: In the process of internal stakeholder engagement, we engaged with key internal stakeholders including the Montefiore-Einstein Offices of Community and Population Health and Community Affairs and the Comprehensive Cancer Center’s Community Outreach team to align goals and development plans. With stakeholders, we created an initial set of linked databases for tracking community engagement. We then collected feedback from 36 collaborator groups including the Montefiore Care Management Organization. Our collaborators expressed concerns about privacy, so we refined our tracker to limit data accessibility. At present, there are 36 active users and 28 community and faith-based organizations are tracked. We are collecting data to improve usability of the tracker and plan to implement dashboards in the future. Discussion/Significance of Impact: Catalyzing and sustaining community engagement work in clinical and translational research are paramount but are limited by several logistical barriers. We hope the CE Tracker can decrease logistical barriers to working with communities by building on and tracking currently existing relationships.

Objectives/Goals: The objective of this study is to assemble a community advisory board (CAB) of interdisciplinary stakeholders to contextualize study results and co-create translational materials that can be used by local governments, community leaders, and community-based organizations to prevent electric bicycle (ebike) injuries. Methods/Study Population: This study merges community-engaged research and data science approaches. We are assembling an interdisciplinary CAB who are key stakeholders in informing ebike safety outside of the scientific research community. The CAB will advise on a retrospective observational data science study currently in progress that uses a California wide dataset to examine how features of neighborhoods and intersections impact ebike injuries. The CAB will provide context for study results based on their professional experiences and co-design translational materials (e.g., policy brief) that convey research findings for effective translation into policies and practices to prevent ebike injuries. We will track data on CAB activities, perceptions of participation from members, dissemination and reach of translational materials. Results/Anticipated Results: We assembled a CAB of five individuals who are policymakers, local government officials, first responders, school board members, and non-profit workers. We will report data on the frequency of meetings, themes emerging from each meeting based on discussions with CAB members, and perceptions of the experience by CAB members. We will also report on the type and content of the translational products created, the process dissemination of these translational materials, and the reach of the translational materials. Discussion/Significance of Impact: The translational aspect of this project is significant for disseminating findings to communities outside of the research realm who can implement changes that improve ebike safety. Our goal is to shorten the time for research to be translated into policies and interventions by leveraging community-engaged research and data science approaches.

Objectives/Goals: In patients with paraesophageal hernia (PEH), evaluate associations between CT radiomic features and clinical data with (1) undergoing elective repair and (2) volvulus/gastric outlet obstruction and develop an ML pipeline for automated feature extraction and prognosis estimation. Methods/Study Population: Retrospective case–control of adults (>18) at Penn Medicine (2017–2025) with hiatal hernia ICD-9/10 codes and CT chest or CT abdomen/pelvis from clinical care. Radiology reports, H&P, and operative notes will be used to identify those with PEH. We will use traditional statistics and classical ML methods (linear/logistic regression, decision trees, random forest) to test associations between radiomic features (hiatal defect diameter (HDD), hernia sac volume (HSV), herniated gastric volume (HGV)) and outcomes. We will explore CNN-based models – nnU-Net (e.g., TotalSegmentator) and ResNet-based models (e.g., Merlin) – and foundation models (MedSAM2, VISTA-3D) to build the pipeline. Results/Anticipated Results: We hypothesize that CT-derived radiomic features will be associated with (1) undergoing elective repair and (2) developing acute volvulus/GOO. Greater HDD, HSV, and HGV are expected to be associated with increased odds of both outcomes. The ML pipeline is expected to provide organ segmentation and reproducible automated feature extraction using state-of-the-art architectures, and to capture synergistic effects between radiographic and clinical variables to inform patient-level prognosis. Discussion/Significance of Impact: In asymptomatic/minimally symptomatic patients, PEH care often defaults to watchful waiting; candidacy and timing for elective repair are uncertain. Our ML pipeline integrates clinical and CT data to standardize preop radiomics, estimate prognosis, and inform if/when to operate.

Objectives/Goals: This project aims to develop a multimodal vision-language deep learning model that can integrate cell microscopy images and sequencing data in retinal cells to help identify novel pathways involved in diseases such as glaucoma and age-related macular degeneration. Methods/Study Population: Retinal cells, derived from human induced pluripotent stem cells, will be cultured in monolayers and divided into control and stressed groups. The cells will be exposed to oxidative and inflammatory stimuli, then sequenced and imaged at various timepoints. Confocal microscopy will capture morphology and marker staining, while single cell RNA sequencing will profile gene expression. A pretrained dual-encoder vision-language model will be fine-tuned with the multimodal data to embed images and omics data into a shared space via self-supervised learning. The model will then be further adapted for downstream tasks, such as predicting transcriptomic changes from cell imaging and vice versa. Results/Anticipated Results: We anticipate that the fine-tuned model will be able accurately align assign stress conditions of the cell from imaging features and transcriptome profiles, as well as predict cell morphology from gene expressions and vice versa. Additionally, we expect it to be able to generate candidate gene networks and predict gene perturbations related to cellular stress response. The performance of the model will be evaluated as a combination of machine learning metrics such as classification accuracy and area under the curve (AUC) and biological metrics such as single-cell clustering quality. Discussion/Significance of Impact: Unlike most other clinical AI applications that utilize single-modal data, this study pioneers a multimodal vision-language model capable of integrating imaging and sequencing data in ophthalmology. The model may uncover new retinal disease pathways and serve as a foundation for diverse downstream tools for the research community.

Objectives/Goals: Cholangiocarcinoma is an aggressive cancer with a poor prognosis. Fibroblast Growth Factor Receptor 2 fusions are therapeutic targets in ~20% of cholangiocarcinoma, which restricts their benefit to only a subset of patients. We investigate reverse phase protein array, as a novel assay for biomarker identification in cholangiocarcinoma. Methods/Study Population: Tumor samples were obtained via our registry (OSU-13053, NCT02090530). RPPA analysis was performed on tumor samples from 24 cholangiocarcinoma patients. Of these, 12 had FGFR2 fusions, 4 had FGFR2 SNVs, and 8 were FGFR wildtype. Tumor content was enriched using laser capture microdissection prior to analysis by reverse phase protein array. RPPA signal was measured by antibody binding to the phosphorylated tyrosine residue at location Y653.654 of FGFR2. Additional analysis was done for the following proteins: CD3, RSK, SGK1, STAT3, VEGFR2, 4EBP1, AKT, EGFR, eIF4G, ERK, FSR2, HLA, mTOR, p38 MAPK, p70S6K, p90RSK, PD-L1, PDGFR, RET, SHC, STAT, and ZAP70. Results/Anticipated Results: There was not a significant difference between FGFR2-phosphorylation in tumors with FGFR2 genomic alterations and tumors with wildtype FGFR. On the other hand, CD3 protein abundance was significantly lower in FGFR-altered tumors compared to wildtype, potentially indicating a cold tumor immune microenvironment in these patients. We plan to perform RPPA analysis on 61 more cases (17 FGFR-altered and 44 FGFR-wildtype) to generate a more robust sample size for subsequent analysis. Furthermore, we have access to 580 additional archival cholangiocarcinoma tumors for RPPA analysis (IRB approved protocol OSU-15030). We plan to sequence these tumors to determine FGFR status. We predict approximately 116 (20%) of these cases to be FGFR-altered. Discussion/Significance of Impact: Our results suggest that FGFR wildtype tumor may have increased FGFR pathway activation and could respond to FGFR inhibitors. Additionally, the cold tumor microenvironment in FGFR-altered tumors provide rationale for combination of FGFR inhibitors and immunotherapy. These results will be confirmed by RPPA analysis on additional cases.