Biocomputing 2025

The following sections are included:

• CONTENTS
• PACIFIC SYMPOSIUM ON BIOCOMPUTING 2025

The 2025 Pacific Symposium on Biocomputing (PSB) represents a remarkable milestone, as it is the thirtieth anniversary of PSB. We use this opportunity to analyze the bibliometric output of 30 years of PSB publications in a wide range of analyses with a focus on various eras that represent important disruptive breakpoints in the field of bioinformatics and biocomputing. These include an analysis of paper topics and keywords, flight emissions produced by travel to PSB by authors, citation and co-authorship networks and metrics, and a broad assessment of diversity and representation in PSB authors. We use the results of these analyses to identify insights that we can carry forward to the upcoming decades of PSB.

Founded nearly 30 years ago, the Pacific Symposium on Biocomputing (PSB) has continually promoted collaborative research in computational biology, annually highlighting emergent themes that reflect the expanding interdisciplinary nature of the field. This study aimed to explore the collaborative and thematic dynamics at PSB using topic modeling and network analysis methods. We identified 14 central topics that have characterized the discourse at PSB over the past three decades. Our findings demonstrate significant trends in topic relevance, with a growing emphasis on machine learning and integrative analyses. We observed not only an expanding nexus of collaboration but also PSB’s crucial role in fostering interdisciplinary collaborations. It remains unclear, however, whether the shift towards interdisciplinarity was driven by the conference itself, external academic trends, or broader societal shifts towards integrated research approaches. Future applications of next-generation analytical methods may offer deeper insights into these dynamics. Additionally, we have developed a web application that leverages retrieval augmented generation and large language models, enabling users to efficiently explore past PSB proceedings.

Artificial Intelligence (AI) technologies are increasingly capable of processing complex and multilayered datasets. Innovations in generative AI and deep learning have notably enhanced the extraction of insights from both unstructured texts, images, and structured data alike. These breakthroughs in AI technology have spurred a wave of research in the medical field, leading to the creation of a variety of tools aimed at improving clinical decision-making, patient monitoring, image analysis, and emergency response systems. However, thorough research is essential to fully understand the broader impact and potential consequences of deploying AI within the healthcare sector.

Human involvement remains critical in most instances of clinical decision-making. Recent advances in AI and machine learning opened the door for designing, implementing, and translating interactive AI systems to support clinicians in decision-making. Assessing the impact and implications of such systems on patient care and clinical workflows requires in-depth studies. Conducting evaluation studies of AI-supported interactive systems to support decision-making in clinical settings is challenging and time-consuming. These studies involve carefully collecting, analyzing, and interpreting quantitative and qualitative data to assess the performance of the underlying AI-supported system, its impact on the human decision-making process, and the implications for patient care. We have previously developed a toolkit for designing and implementing clinical AI software so that it can be subjected to an application-based evaluation. Here, we present a visual analytics frame-work for analyzing and interpreting the data collected during such an evaluation process. Our framework supports identifying subgroups of users and patients based on their characteristics, detecting outliers among them, and providing evidence to ensure adherence to regulatory guidelines. We used early-stage clinical AI regulatory guidelines to drive the system design, implemented multiple-factor analysis and hierarchical clustering as exemplary analysis tools, and provided interactive visualizations to explore and interpret results. We demonstrate the effectiveness of our framework through a case study to evaluate a prototype AI-based clinical decision-support system for diagnosing pediatric brain tumors.

The United States Medical Licensing Examination (USMLE) is a critical step in assessing the competence of future physicians, yet the process of creating exam questions and study materials is both time-consuming and costly. While Large Language Models (LLMs), such as OpenAI’s GPT-4, have demonstrated proficiency in answering medical exam questions, their potential in generating such questions remains underexplored. This study presents QUEST-AI, a novel system that utilizes LLMs to (1) generate USMLE-style questions, (2) identify and flag incorrect questions, and (3) correct errors in the flagged questions. We evaluated this system’s output by constructing a test set of 50 LLM-generated questions mixed with 50 human-generated questions and conducting a two-part assessment with three physicians and two medical students. The assessors attempted to distinguish between LLM and human-generated questions and evaluated the validity of the LLM-generated content. A majority of exam questions generated by QUEST-AI were deemed valid by a panel of three clinicians, with strong correlations between performance on LLM-generated and human-generated questions. This pioneering application of LLMs in medical education could significantly increase the ease and efficiency of developing USMLE-style medical exam content, offering a cost-effective and accessible alternative for exam preparation.

Accurately interpreting medical images and writing radiology reports is a critical but challenging task in healthcare. Both human-written and AI-generated reports can contain errors, ranging from clinical inaccuracies to linguistic mistakes. To address this, we introduce ReXErr, a methodology that leverages Large Language Models to generate representative errors within chest X-ray reports. Working with board-certified radiologists, we developed error categories that capture common mistakes in both human and AI-generated reports. Our approach uses a novel sampling scheme to inject diverse errors while maintaining clinical plausibility. ReXErr demonstrates consistency across error categories and produces errors that closely mimic those found in real-world scenarios. This method has the potential to aid in the development and evaluation of report correction algorithms, potentially enhancing the quality and reliability of radiology reporting.

Over the past decade, wearable technology has dramatically changed how patients manage chronic diseases. The widespread availability of on-body sensors, such as heart rate monitors and continuous glucose monitoring (CGM) sensors, has allowed patients to have real-time data about their health. Most of these data are readily available on patients’ smartphone applications, where patients can view their current and retrospective data. For patients with diabetes, CGM has transformed how their disease is managed. Many sensor devices interface with smartphones to display charts, metrics, and alerts. However, these metrics and plots may be challenging for some patients to interpret. In this work, we explore how large language models (LLMs) can be used to answer questions about CGM data. We produce an open-source benchmark of time-series question-answering tasks for CGM data in diabetes management. We evaluate different LLM frameworks to provide a performance benchmark. Lastly, we highlight the need for more research on how to optimize LLM frameworks to best handle questions about wearable data. Our benchmark is publicly available for future use and development. While this benchmark is specifically designed for diabetes care, our model implementation and several of the statistical tasks can be extended to other wearable device domains.

This study investigates the potential of using synthetic text to augment training data for Natural Language Processing (NLP) models, specifically within the context of peer support tools. We surveyed 22 participants—13 professional peer supporters and 9 AI-proficient individuals—tasked with distinguishing between AI-generated and human-written sentences. Using signal detection theory and confidence-based metrics, we evaluated the accuracy and confidence levels of both groups. The results show no significant differences in rater agreement between the two groups (p = 0.116), with overall classification accuracy falling below chance levels (mean accuracy = 43.10%, p < 0.001). Both groups exhibited a tendency to misclassify low-fidelity sentences as AI-generated, with peer supporters showing a significant bias (p = 0.007). Further analysis revealed a significant negative correlation between errors and confidence among AI-proficient raters (r = -0.429, p < 0.001), suggesting that as their confidence increased, their error rates decreased. Our findings support the feasibility of using synthetic text to mimic human communication, with important implications for improving the fidelity of peer support interventions through NLP model development.

We present a comparative study on the performance of two popular open-source large language models for early prediction of sepsis: Llama-3 8B and Mixtral 8x7B. The primary goal was to determine whether a smaller model could achieve comparable predictive accuracy to a significantly larger model in the context of sepsis prediction using clinical data.

Our proposed LLM-based sepsis prediction system, COMPOSER-LLM, enhances the previously published COMPOSER model, which utilizes structured EHR data to generate hourly sepsis risk scores. The new system incorporates an LLM-based approach to extract sepsis-related clinical signs and symptoms from unstructured clinical notes. For scores falling within high-uncertainty prediction regions, particularly those near the decision threshold, the system uses the LLM to draw additional clinical context from patient notes; thereby enhancing the model’s predictive accuracy in challenging diagnostic scenarios.

A total of 2,074 patient encounters admitted to the Emergency Department at two hospitals within the University of California San Diego Health system were used for model evaluation in this study. Our findings reveal that the Llama-3 8B model based system (COMPOSER-LLM_Llama) achieved a sensitivity of 70.3%, positive predictive value (PPV) of 32.5%, F-1 score of 44.4% and false alarms per patient hour (FAPH) of 0.0194, closely matching the performance of the larger Mixtral 8x7B model based system (COMPOSER-LLM_mixtral) which achieved a sensitivity of 72.1%, PPV of 31.9%, F-1 score of 44.2% and FAPH of 0.020. When prospectively evaluated, COMPOSER-LLM_Llama demonstrated similar performance to the COMPOSER-LLM_mixtral pipeline, with a sensitivity of 68.7%, PPV of 36.6%, F-1 score of 47.7% and FAPH of 0.019 vs. sensitivity of 70.5%, PPV of 36.3%, F-1 score of 47.9% and FAPH of 0.020. This result indicates that, for extraction of clinical signs and symptoms from unstructured clinical notes to enable early prediction of sepsis, the Llama-3 generation of smaller language models can perform as effectively and more efficiently than larger models. This finding has significant implications for healthcare settings with limited resources.

The primary challenge in reporting cancer cases lies in the labor-intensive and time-consuming process of manually reviewing numerous reports. Current methods predominantly rely on rule-based approaches or custom-supervised learning models, which predict diagnostic codes based on a single pathology report per patient. Although these methods show promising evaluation results, their biased outcomes in controlled settings may hinder adaption to real-world reporting workflows. In this feasibility study, we focused on lung cancer as a test case and developed an agentic retrieval-augmented generation (RAG) system to evaluate the potential of publicly available large language models (LLMs) for cancer registry coding. Our findings demonstrate that: (1) directly applying publicly available LLMs without fine-tuning is feasible for cancer registry coding; and (2) prompt engineering can significantly enhance the capability of pre-trained LLMs in cancer registry coding. The off-the-shelf LLM, combined with our proposed system architecture and basic prompts, achieved a macro-averaged F-score of 0.637 when evaluated on testing data consisting of patients’ medical reports spanning 1.5 years since their first visit. By employing chain of thought (CoT) reasoning and our proposed coding item grouping, the system outperformed the baseline by 0.187 in terms of the macro-averaged F-score. These findings demonstrate the great potential of leveraging LLMs with prompt engineering for cancer registry coding. Our system could offer cancer registrars a promising reference tool to enhance their daily workflow, improving efficiency and accuracy in cancer case reporting.

Background: Ensuring antibiotics are prescribed only when necessary is crucial for maintaining their effectiveness and is a key focus of public health initiatives worldwide. In cases of sinusitis, among the most common reasons for antibiotic prescriptions in children, healthcare providers must distinguish between bacterial and viral causes based on clinical signs and symptoms. However, due to the overlap between symptoms of acute sinusitis and viral upper respiratory infections, antibiotics are often over-prescribed.

Objectives: Currently, there are no electronic health record (EHR)-based methods, such as lab tests or ICD-10 codes, to retroactively assess the appropriateness of prescriptions for sinusitis, making manual chart reviews the only available method for evaluation, which is time-intensive and not feasible at a large scale. In this study, we propose using natural language processing to automate this assessment.

Methods: We developed, trained, and evaluated generative models to classify the appropriateness of antibiotic prescriptions in 300 clinical notes from pediatric patients with sinusitis seen at a primary care practice in the Children’s Hospital of Philadelphia network. We utilized standard prompt engineering techniques, including few-shot learning and chain-of-thought prompting, to refine an initial prompt. Additionally, we employed Parameter-Efficient Fine-Tuning to train a medium-sized generative model Llama 3 70B-instruct.

Results: While parameter-efficient fine-tuning did not enhance performance, the combination of few-shot learning and chain-of-thought prompting proved beneficial. Our best results were achieved using the largest generative model publicly available to date, the Llama 3.1 405B-instruct. On our evaluation set, the model correctly identified 94.7% of the 152 notes where antibiotic prescription was appropriate and 66.2% of the 83 notes where it was not appropriate. However, 15 notes that were insufficiently, vaguely, or ambiguously documented by physicians posed a challenge to our model, as none were accurately classified.

Conclusion: Our generative model demonstrated good performance in the challenging task of chart review. This level of performance may be sufficient for deploying the model within the EHR, where it can assist physicians in real-time to prescribe antibiotics in concordance with the guidelines, or for monitoring antibiotic stewardship on a large scale.

The rapid advancement of artificial intelligence and machine learning (AI/ML) technologies in healthcare presents significant opportunities for enhancing patient care through innovative diagnostic tools, monitoring systems, and personalized treatment plans. However, these innovative advancements might result in regulatory challenges given recent Supreme Court decisions that impact the authority of regulatory agencies like the Food and Drug Administration (FDA). This paper explores the implications of regulatory uncertainty for the healthcare industry related to balancing innovation in biotechnology and biocomputing with ensuring regulatory uniformity and patient safety. We examine key Supreme Court cases, including Loper Bright Enterprises v. Raimondo, Relentless, Inc. v. Department of Commerce, and Corner Post, Inc. v. Board of Governors of the Federal Reserve System, and their impact on the Chevron doctrine. We also discuss other relevant cases to highlight shifts in judicial approaches to agency deference and regulatory authority that might affect how science is handled in regulatory spaces, including how biocomputing and other health sciences are governed, how scientific facts are applied in policymaking, and how scientific expertise guides decision making. Through a detailed analysis, we assess the potential impact of regulatory uncertainty in healthcare. Additionally, we provide recommendations for the medical community on navigating these challenges.

Accurate prediction of suicide risk is crucial for identifying patients with elevated risk burden, helping ensure these patients receive targeted care. The US Department of Veteran Affairs’ suicide prediction model primarily leverages structured electronic health records (EHR) data. This approach largely overlooks unstructured EHR, a data format that could be utilized to enhance predictive accuracy. This study aims to enhance suicide risk models’ predictive accuracy by developing a model that incorporates both structured EHR predictors and semantic NLP-derived variables from unstructured EHR. XGBoost models were fit to predict suicide risk– the interactions identified by the model were extracted using SHAP, validated using logistic regression models, added to a ridge regression model, which was subsequently compared to a ridge regression approach without the use of interactions. By introducing a selection parameter, α, to balance the influence of structured (α=1) and unstructured (α=0) data, we found that intermediate α values achieved optimal performance across various risk strata, improved model performance of the ridge regression approach and uncovered significant cross-modal interactions between psychosocial constructs and patient characteristics. These interactions highlight how psychosocial risk factors are influenced by individual patient contexts, potentially informing improved risk prediction methods and personalized interventions. Our findings underscore the importance of incorporating nuanced narrative data into predictive models and set the stage for future research that will expand the use of advanced machine learning techniques, including deep learning, to further refine suicide risk prediction methods.

Given the rapidly expanding capabilities of generative AI models for radiology, there is a need for robust metrics that can accurately measure the quality of AI-generated radiology reports across diverse hospitals. We develop ReXamine-Global, a LLM-powered, multi-site framework that tests metrics across different writing styles and patient populations, exposing gaps in their generalization. First, our method tests whether a metric is undesirably sensitive to reporting style, providing different scores depending on whether AI-generated reports are stylistically similar to ground-truth reports or not. Second, our method measures whether a metric reliably agrees with experts, or whether metric and expert scores of AI-generated report quality diverge for some sites. Using 240 reports from 6 hospitals around the world, we apply ReXamine-Global to 7 established report evaluation metrics and uncover serious gaps in their generalizability. Developers can apply ReXamine-Global when designing new report evaluation metrics, ensuring their robustness across sites. Additionally, our analysis of existing metrics can guide users of those metrics towards evaluation procedures that work reliably at their sites of interest.

The emergent abilities of large language models (LLMs) have demonstrated great potential in solving medical questions. They can possess considerable medical knowledge, but may still hallucinate and are inflexible in the knowledge updates. While Retrieval-Augmented Generation (RAG) has been proposed to enhance the medical question-answering capabilities of LLMs with external knowledge bases, it may still fail in complex cases where multiple rounds of information-seeking are required. To address such an issue, we propose iterative RAG for medicine (i-MedRAG), where LLMs can iteratively ask follow-up queries based on previous information-seeking attempts. In each iteration of i-MedRAG, the follow-up queries will be answered by a vanilla RAG system and they will be further used to guide the query generation in the next iteration. Our experiments show the improved performance of various LLMs brought by i-MedRAG compared with vanilla RAG on complex questions from clinical vignettes in the United States Medical Licensing Examination (USMLE), as well as various knowledge tests in the Massive Multitask Language Understanding (MMLU) dataset. Notably, our zero-shot i-MedRAG outperforms all existing prompt engineering and fine-tuning methods on GPT-3.5, achieving an accuracy of 69.68% on the MedQA dataset. In addition, we characterize the scaling properties of i-MedRAG with different iterations of follow-up queries and different numbers of queries per iteration. Our case studies show that i-MedRAG can flexibly ask follow-up queries to form reasoning chains, providing an in-depth analysis of medical questions. To the best of our knowledge, this is the first-of-its-kind study on incorporating follow-up queries into medical RAG.

Artificial Intelligence (AI) algorithms showcase the potential to steer a paradigm shift in clinical medicine, especially medical imaging. Concerns associated with model generalizability and biases necessitate rigorous external validation of AI algorithms prior to their adoption into clinical workflows. To address the barriers associated with patient privacy, intellectual property, and diverse model requirements, we introduce ClinValAI, a framework for establishing robust cloud-based infrastructures to clinically validate AI algorithms in medical imaging. By featuring dedicated workflows for data ingestion, algorithm scoring, and output processing, we propose an easily customizable method to assess AI models and investigate biases. Our novel orchestration mechanism facilitates utilizing the complete potential of the cloud computing environment. ClinValAI’s input auditing and standardization mechanisms ensure that inputs consistent with model prerequisites are provided to the algorithm for a streamlined validation. The scoring workflow comprises multiple steps to facilitate consistent inferencing and systematic troubleshooting. The output processing workflow helps identify and analyze samples with missing results and aggregates final outputs for downstream analysis. We demonstrate the usability of our work by evaluating a state-of-the-art breast cancer risk prediction algorithm on a large and diverse dataset of 2D screening mammograms. We perform comprehensive statistical analysis to study model calibration and evaluate performance on important factors, including breast density, age, and race, to identify latent biases. ClinValAI provides a holistic framework to validate medical imaging models and has the potential to advance the development of generalizable AI models in clinical medicine and promote health equity.

Pharmacogenetics represents one of the most promising areas of precision medicine, with several guidelines for genetics-guided treatment ready for clinical use. Despite this, implementation has been slow, with few health systems incorporating the technology into their standard of care. One major barrier to uptake is the lack of education and awareness of pharmacogenetics among clinicians and patients. The introduction of large language models (LLMs) like GPT-4 has raised the possibility of medical chatbots that deliver timely information to clinicians, patients, and researchers with a simple interface. Although state-of-the-art LLMs have shown impressive performance at advanced tasks like medical licensing exams, in practice they still often provide false information, which is particularly hazardous in a clinical context. To quantify the extent of this issue, we developed a series of automated and expert-scored tests to evaluate the performance of chatbots in answering pharmacogenetics questions from the perspective of clinicians, patients, and researchers. We applied this benchmark to state-of-the-art LLMs and found that newer models like GPT-4o greatly outperform their predecessors, but still fall short of the standards required for clinical use. Our benchmark will be a valuable public resource for subsequent developments in this space as we work towards better clinical AI for pharmacogenetics.

The following sections are included:

• Overview
• Advancing multi-ancestry genetic research
• Integrating social determinants of health to enhance genetic risk models
• Methods to detect and mitigate disparities
• Addressing Disparities in Adverse Drug Reactions
• Conclusion
• Acknowledgments
• References

Genome-wide association studies (GWAS) are an important tool for the study of complex disease genetics. Decisions regarding the quality control (QC) procedures employed as part of a GWAS can have important implications on the results and their biological interpretation. Many GWAS have been conducted predominantly in cohorts of European ancestry, but many initiatives aim to increase the representation of diverse ancestries in genetic studies. The question of how these data should be combined and the consequences that genetic variation across ancestry groups might have on GWAS results warrants further investigation. In this study, we focus on several commonly used methods for combining genetic data across diverse ancestry groups and the impact these decisions have on the outcome of GWAS summary statistics. We ran GWAS on two binary phenotypes using ancestry-specific, multi-ancestry mega-analysis, and meta-analysis approaches. We found that while multi-ancestry mega-analysis and meta-analysis approaches can aid in identifying signals shared across ancestries, they can diminish the signal of ancestry-specific associations and modify their effect sizes. These results demonstrate the potential impact on downstream post-GWAS analyses and follow-up studies. Decisions regarding how the genetic data are combined has the potential to mask important findings that might serve individuals of ancestries that have been historically underrepresented in genetic studies. New methods that consider ancestry-specific variants in conjunction with the shared variants need to be developed.

Uterine leiomyomata, or fibroids, are common gynecological tumors causing pelvic and menstrual symptoms that can negatively affect quality of life and child-bearing desires. As fibroids grow, symptoms can intensify and lead to invasive treatments that are less likely to preserve fertility. Identifying individuals at highest risk for fibroids can aid in access to earlier diagnoses. Polygenic risk scores (PRS) quantify genetic risk to identify those at highest risk for disease. Utilizing the PRS software PRS-CSx and publicly available genome-wide association study (GWAS) summary statistics from FinnGen and Biobank Japan, we constructed a multi-ancestry (META) PRS for fibroids. We validated the META PRS in two cross-ancestry cohorts. In the cross-ancestry Electronic Medical Record and Genomics (eMERGE) Network cohort, the META PRS was significantly associated with fibroid status and exhibited 1.11 greater odds for fibroids per standard deviation increase in PRS (95% confidence interval [CI]: 1.05 – 1.17, p = 5.21x10⁻⁵). The META PRS was validated in two BioVU cohorts: one using ICD9/ICD10 codes and one requiring imaging confirmation of fibroid status. In the ICD cohort, a standard deviation increase in the META PRS increased the odds of fibroids by 1.23 (95% CI: 1.15 – 1.32, p = 9.68x10⁻⁹), while in the imaging cohort, the odds increased by 1.26 (95% CI: 1.18 – 1.35, p = 2.40x10⁻¹¹). We subsequently constructed single ancestry PRS for FinnGen (European ancestry [EUR]) and Biobank Japan (East Asian ancestry [EAS]) using PRS-CS and discovered a nominally significant association in the eMERGE cohort within fibroids and EAS PRS but not EUR PRS (95% CI: 1.09 – 1.20, p = 1.64x10⁻⁷). These findings highlight the strong predictive power of multi-ancestry PRS over single ancestry PRS. This study underscores the necessity of diverse population inclusion in genetic research to ensure precision medicine benefits all individuals equitably.

Uterine leiomyomata (fibroids, UFs) are common, benign tumors in females, having an estimated prevalence of up to 80%. They are fibrous masses growing within the myometrium leading to chronic symptoms like dysmenorrhea, abnormal uterine bleeding, anemia, severe pelvic pain, and infertility. Hypertension (HTN) is a common risk factor for UFs, though less prevalent in premenopausal individuals. While observational studies have indicated strong associations between UFs and HTN, the biological mechanisms linking the two conditions remain unclear. Understanding the relationship between HTN and UFs is crucial because UFs and HTN lead to substantial comorbidities adversely impacting female health. Identifying the common underlying biological mechanisms can improve treatment strategies for both conditions. To clarify the genetic and causal relationships between UFs and BP, we conducted a bidirectional, two-sample Mendelian randomization (MR) analysis and evaluated the genetic correlations across BP traits and UFs. We used data from a multi-ancestry genome-wide association study (GWAS) meta-analysis of UFs (44,205 cases and 356,552 controls), and data from a cross-ancestry GWAS meta-analysis of BP phenotypes (diastolic BP [DBP], systolic BP [SBP], and pulse pressure [PP], N=447,758). We evaluated genetic correlation of BP phenotypes and UFs with linkage disequilibrium score regression (LDSC). LDSC results indicated a positive genetic correlation between DBP and UFs (Rg=0.132, p<5.0x10^-5), and SBP and UFs (Rg=0.063, p<2.5x10^-2). MR using UFs as the exposure and BP traits as outcomes indicated a relationship where UFs increases DBP (odds ratio [OR]=1.20, p<2.7x10^-3). Having BP traits as exposures and UFs as the outcome showed that DBP and SBP increase risk for UFs (OR =1.04, p<2.2x10^-3; OR=1.00, p<4.0x10^-2; respectively). Our results provide evidence of shared genetic architecture and pleiotropy between HTN and UFs, suggesting common biological pathways driving their etiologies. Based on these findings, DBP appears to be a stronger risk factor for UFs compared to SBP and PP.

Women’s health conditions are influenced by both genetic and environmental factors. Understanding these factors individually and their interactions is crucial for implementing preventative, personalized medicine. However, since genetics and environmental exposures, particularly social determinants of health (SDoH), are correlated with race and ancestry, risk models without careful consideration of these measures can exacerbate health disparities. We focused on seven women’s health disorders in the All of Us Research Program: breast cancer, cervical cancer, endometriosis, ovarian cancer, preeclampsia, uterine cancer, and uterine fibroids. We computed polygenic risk scores (PRSs) from publicly available weights and tested the effect of the PRSs on their respective phenotypes as well as any effects of genetic risk on age at diagnosis. We next tested the effects of environmental risk factors (BMI, lifestyle measures, and SDoH) on age at diagnosis. Finally, we examined the impact of environmental exposures in modulating genetic risk by stratified logistic regressions for different tertiles of the environment variables, comparing the effect size of the PRS. Of the twelve sets of weights for the seven conditions, nine were significantly and positively associated with their respective phenotypes. None of the PRSs was associated with different ages at diagnoses in the time-to-event analyses. The highest environmental risk group tended to be diagnosed earlier than the low and medium-risk groups. For example, the cases of breast cancer, ovarian cancer, uterine cancer, and uterine fibroids in highest BMI tertile were diagnosed significantly earlier than the low and medium BMI groups, respectively). PRS regression coefficients were often the largest in the highest environment risk groups, showing increased susceptibility to genetic risk. This study’s strengths include the diversity of the All of Us study cohort, the consideration of SDoH themes, and the examination of key risk factors and their interrelationships. These elements collectively underscore the importance of integrating genetic and environmental data to develop more precise risk models, enhance personalized medicine, and ultimately reduce health disparities.

We leveraged electronic health record (EHR) data from the Accelerating Data Value Across a National Community Health Center Network (ADVANCE) Clinical Research Network (CRN) to identify social risk factor clusters, assess their association with obstructive sleep apnea (OSA), and determine relevant clinical predictors of cardiovascular (CV) outcomes among those experiencing OSA. Geographically informed social indicators were used to define social risk factor clusters via latent class analysis. EHR-wide diagnoses were used as predictors of 5-year incidence of major adverse CV events (MACE) using STREAMLINE, an end-to-end rigorous and interpretable automated machine learning pipeline. Analyses among over 1.4 million individuals revealed three major social risk factor clusters: lowest (35.7%), average (43.6%) and highest (22.7%) social burden. In adjusted analyses, those experiencing highest social burden were less likely to have received a diagnosis of OSA when compared to those experiencing lowest social burden (OR [95%CI]=0.85[0.82-0.88]). Among those with OSA and free of prior CV diseases (N=4,405), performance of predicting incident MACE reached a ROC-AUC of 0.70 [0.03] overall but varied when assessed within each social risk factor cluster. Feature importance also revealed that different clinical factors might explain predictions among each cluster. Results suggest relevant health disparities in the diagnosis of OSA and across clinical predictors of CV diseases among those with OSA, across social risk factor clusters, indicating that tailored interventions geared toward minimizing these disparities are warranted.

Clinical decisions to treat and diagnose patients are affected by implicit biases formed by racism, ableism, sexism, and other stereotypes. These biases reflect broader systemic discrimination in healthcare and risk marginalizing already disadvantaged groups. Existing methods for measuring implicit biases require controlled randomized testing and only capture individual attitudes rather than outcomes. However, the ”big-data” revolution has led to the availability of large observational medical datasets, like EHRs and biobanks, that provide the opportunity to investigate discrepancies in patient health outcomes. In this work, we propose a causal inference approach to detect the effect of clinician implicit biases on patient outcomes in large-scale medical data. Specifically, our method uses proximal mediation to disentangle pathway-specific effects of a patient’s sociodemographic attribute on a clinician’s diagnosis decision. We test our method on real-world data from the UK Biobank. Our work can serve as a tool that initiates conversation and brings awareness to unequal health outcomes caused by implicit biases.

In the Carolina Breast Cancer Study (CBCS), clustering census tracts based on spatial location, demographic variables, and socioeconomic status is crucial for understanding how these factors influence health outcomes and cancer risk. This task, known as spatial clustering, involves identifying clusters of similar locations by considering both geographic and characteristic patterns. While standard clustering methods such as K-means, spectral clustering, and hierarchical clustering are well-studied, spatial clustering is less explored due to the inherent differences between spatial domains and their corresponding covariates. In this paper, we introduce a spatial clustering algorithm called Gaussian Process Spatial Clustering (GPSC). GPSC leverages the flexibility of Gaussian Processes to cluster unobserved functions between different domains, extending traditional clustering techniques to effectively handle geospatial data. We provide theoretical guarantees for GPSC’s performance and demonstrate its capability to recover true clusters through several empirical studies. Specifically, we identify clusters of census tracts in North Carolina based on socioeconomic and environmental indicators associated with health and cancer risk.

Patients experiencing adverse drug events (ADE) from polypharmaceutical regimens present a huge challenge to modern healthcare. While computational efforts may reduce the incidence of these ADEs, current strategies are typically non-generalizable for standard healthcare systems. To address this, we carried out a retrospective study aimed at developing a statistical approach to detect and quantify potential ADEs. The data foundation comprised of almost 2 million patients from two health regions in Denmark and their drug and laboratory data during the years 2011 to 2016. We developed a series of multistate Cox models to compute hazard ratios for changes in laboratory test results before and after drug exposure. By linking the results to data from a drug-drug interaction database, we found that the models showed potential for applications for medical safety agencies and improved efficiency for drug approval pipelines.

Precision medicine focuses on developing treatments and preventative strategies tailored to an individual’s genomic profile, lifestyle, and environmental context. The Precision Medicine sessions at the Pacific Symposium on Biocomputing (PSB) have consistently spotlighted progress in this domain. Our 2025 manuscript collection features algorithmic innovations that integrate data across scales and diverse data modalities, presenting novel techniques to derive clinically relevant insights from molecular datasets. These studies highlight recent advances in technology and analytics and their application toward realizing the potential of precision medicine to enhance human health outcomes and extend lifespan.

Genetic perturbation of T cell receptor (TCR) T cells is a promising method to unlock better TCR T cell performance to create more powerful cancer immunotherapies, but understanding the changes to T cell behavior induced by genetic perturbations remains a challenge. Prior studies have evaluated the effect of different genetic modifications with cytokine production and metabolic activity assays. Live-cell imaging is an inexpensive and robust approach to capture TCR T cell responses to cancer. Most methods to quantify T cell responses in live-cell imaging data use simple approaches to count T cells and cancer cells across time, effectively quantifying how much space in the 2D well each cell type covers, leaving actionable information unexplored. In this study, we characterize changes in TCR T cell’s interactions with cancer cells from live-cell imaging data using explainable artificial intelligence (AI). We train convolutional neural networks to distinguish behaviors in TCR T cell with CRISPR knock outs of CUL5, RASA2, and a safe harbor control knockout. We use explainable AI to identify specific interaction types that define different knock-out conditions. We find that T cell and cancer cell coverage is a strong marker of TCR T cell modification when comparing similar experimental time points, but differences in cell aggregation characterize CUL5KO and RASA2KO behavior across all time points. Our pipeline for discovery in live-cell imaging data can be used for characterizing complex behaviors in arbitrary live-cell imaging datasets, and we describe best practices for this goal.

Alzheimer’s disease (AD) is characterized by cognitive decline and memory loss due to the abnormal accumulation of amyloid-beta (Aβ) plaques and tau tangles in the brain; its onset and progression also depend on genetic factors such as the apolipoprotein E (APOE) genotype. Understanding how these factors affect the brain’s neural pathways is important for early diagnostics and interventions. Tractometry is an advanced technique for 3D quantitative assessment of white matter tracts, localizing microstructural abnormalities in diseased populations in vivo. In this work, we applied BUAN (Bundle Analytics) tractometry to 3D diffusion MRI data from 730 participants in ADNI3 (phase 3 of the Alzheimer’s Disease Neuroimaging Initiative; age range: 55-95 years, 349M/381F, 214 with mild cognitive impairment, 69 with AD, and 447 cognitively healthy controls). Using along-tract statistical analysis, we assessed the localized impact of amyloid, tau, and APOE genetic variants on the brain’s neural pathways. BUAN quantifies microstructural properties of white matter tracts, supporting along-tract statistical analyses that identify factors associated with brain microstructure. We visualize the 3D profile of white matter tract associations with tau and amyloid burden in Alzheimer’s disease; strong associations near the cortex may support models of disease propagation along neural pathways. Relative to the neutral genotype, APOE ϵ3/ϵ3, carriers of the AD-risk conferring APOE ϵ4 genotype show microstructural abnormalities, while carriers of the protective ϵ2 genotype also show subtle differences. Of all the microstructural metrics, mean diffusivity (MD) generally shows the strongest associations with AD pathology, followed by axial diffusivity (AxD) and radial diffusivity (RD), while fractional anisotropy (FA) is typically the least sensitive metric. Along-tract microstructural metrics are sensitive to tau and amyloid accumulation, showing the potential of diffusion MRI to track AD pathology and map its impact on neural pathways.

Down syndrome (DS), caused by the triplication of chromosome 21 (T21), is a prevalent genetic disorder with a higher incidence of obesity. Traditional approaches have struggled to differentiate T21-specific molecular dysregulation from general obesity-related processes. This study introduces the omni-PLIER framework, combining the Pathway-Level Information ExtractoR (PLIER) with the omnigenic model, to uncover molecular mechanisms underlying obesity in DS. The PLIER framework aligns gene expression data with biological pathways, facilitating the identification of relevant molecular patterns. Using RNA sequencing data from the Human Trisome Project, omni-PLIER identified latent variables (LVs) significantly associated with both T21 and body mass index (BMI). Elastic net regression and causal mediation analysis revealed LVs mediating the effect of karyotype on BMI. Notably, LVs involving glutathione peroxidase-1 (GPX1) and MCL1 apoptosis regulator, BCL2 family members emerged as crucial mediators. These findings provide insights into the molecular interplay between DS and obesity. The omni-PLIER model offers a robust methodological advancement for dissecting complex genetic disorders, with implications for understanding obesity-related processes in both DS and the general population.

Alzheimer’s disease (AD), the predominant form of dementia, is influenced by several risk factors, including type 2 diabetes (T2D), a metabolic disorder characterized by the dysregulation of blood sugar levels. Despite mouse and human studies reporting this connection between T2D and AD, the mechanism by which T2D contributes to AD pathobiology is not well understood. A challenge in understanding mechanistic links between these conditions is that evidence between mouse and human experimental models must be synthesized, but translating between these systems is difficult due to evolutionary distance, physiological differences, and human heterogeneity. To address this, we employed a computational framework called translatable components regression (TransComp-R) to overcome discrepancies between pre-clinical and clinical studies using omics data. Here, we developed a novel extension of TransComp-R for multi-disease modeling to analyze transcriptomic data from brain samples of mouse models of AD, T2D, and simultaneous occurrence of both disease (ADxT2D) and postmortem human brain data to identify enriched pathways predictive of human AD status. Our TransComp-R model identified inflammatory and estrogen signaling pathways encoded by mouse principal components derived from models of T2D and ADxT2D, but not AD alone, predicted with human AD outcomes. The same mouse PCs predictive of human AD outcomes were able to capture sex-dependent differences in human AD biology, including significant effects unique to female patients, despite the TransComp-R being derived from data from only male mice. We demonstrated that our approach identifies biological pathways of interest at the intersection of the complex etiologies of AD and T2D which may guide future studies into pathogenesis and therapeutic development for patients with T2D-associated AD.

Given the complexity and multifactorial nature of Alzheimer’s disease, investigating potential drug-gene targets is imperative for developing effective therapies and advancing our understanding of the underlying mechanisms driving the disease. We present an explainable ML model that integrates the role and impact of gene interactions to drive the genomic variant feature selection. The model leverages both the Alzheimer’s knowledge base and the Drug-Gene interaction database (DGIdb) to identify a list of biologically plausible novel gene-drug targets for further investigation. Model validation is performed on an ethnically diverse study sample obtained from the Alzheimer’s Disease Sequencing Project (ADSP), a multi-ancestry multi-cohort genomic study. To mitigate population stratification and spurious associations from ML analysis, we implemented novel data curation methods. The study outcomes include a set of possible gene targets for further functional follow-up and drug repurposing.

Adverse drug responses (ADRs) result in over 7,000 deaths annually. Pharmacogenomic studies have shown that many ADRs are partially attributable to genetics. However, emerging data suggest that epigenetic mechanisms, such as DNA methylation (DNAm) also contribute to this variance. Understanding the impact of DNA methylation on drug response may minimize ADRs and improve the personalization of drug regimens. In this work, we identify DNA methylation sites that likely impact drug response phenotypes for anticoagulant and cardiometabolic drugs. We use instrumental variable analysis to integrate genome-wide association study (GWAS) summary statistics derived from electronic health records (EHRs) within the U.K. Biobank (UKBB) with methylation quantitative trait loci (mQTL) data from the Genetics of DNA Methylation Consortium (GoDMC). This approach allows us to achieve a robust sample size using the largest publicly available pharmacogenomic GWAS. For warfarin, we find 71 DNAm sites. Of those, 8 are near the gene VKORC1 and 48 are on chromosome 6 near the human leukocyte antigen (HLA) gene family. We also find 2 warfarin DNAm sites near the genes CYP2C9 and CYP2C19. For statins, we identify 17 DNAm sites. Eight are near the APOB gene, which encodes a carrier protein for low-density lipoprotein cholesterol (LDL-C). We find no novel significant epigenetic results for metformin.

CDK4/6 inhibitors such as palbociclib block cell cycle progression and improve outcomes for many ER+/HER2- breast cancer patients. Unfortunately, many patients are initially resistant to the drug or develop resistance over time in part due to heterogeneity among individual tumor cells. To better understand these mechanisms of resistance, we used multiplex, single-cell imaging to profile cell cycle proteins in ER+ breast tumor cells under increasing palbociclib concentrations. We then applied spherical principal component analysis (SPCA), a dimensionality reduction method that leverages the inherently cyclical nature of the high-dimensional imaging data, to look for changes in cell cycle behavior in resistant cells. SPCA characterizes data as a hypersphere and provides a framework for visualizing and quantifying differences in cell cycles across treatment-induced perturbations. The hypersphere representations revealed shifts in the mean cell state and population heterogeneity. SPCA validated expected trends of CDK4/6 inhibitor response such as decreased expression of proliferation markers (Ki67, pRB), but also revealed potential mechanisms of resistance including increased expression of cyclin D1 and CDK2. Understanding the molecular mechanisms that allow treated tumor cells to evade arrest is critical for identifying targets of future therapies. Ultimately, we seek to further SPCA as a tool of precision medicine, targeting treatments by individual tumors, and extending this computational framework to interpret other cyclical biological processes represented by high-dimensional data.

Alzheimer’s disease (AD) is a polygenic disorder with a prolonged prodromal phase, complicating early diagnosis. Recent research indicates that increased astrocyte reactivity is associated with a higher risk of pathogenic tau accumulation, particularly in amyloid-positive individuals. However, few clinical tools are available to predict which individuals are likely to exhibit elevated astrocyte activation and, consequently, be susceptible to hyperphosphorylated tau-induced neurodegeneration. Polygenic risk scores (PRS) aggregate the effects of multiple genetic loci to provide a single, continuous metric representing an individual’s genetic risk for a specific phenotype. We hypothesized that an astrocyte activation PRS could aid in the early detection of faster clinical decline. Therefore, we constructed an astrocyte activation PRS and assessed its predictive value for cognitive decline and AD biomarkers (i.e., cerebrospinal fluid [CSF] levels of Aβ1-42, total tau, and p-tau181) in a cohort of 791 elderly individuals. The astrocyte activation PRS showed significant main effects on cross-sectional memory (β = -0.07, p = 0.03) and longitudinal executive function (β = -0.01, p = 0.03). Additionally, the PRS interacted with amyloid positivity (p.intx = 0.02), whereby indicating that amyloid burden modifies the association between the PRS and annual rate of language decline. Furthermore, the PRS was negatively associated with CSF Aβ1-42 levels (β = -3.4, p = 0.07) and interacted with amyloid status, such that amyloid burden modifies the association between the PRS and CSF phosphorylated tau levels (p.intx = 0.08). These findings suggest that an astrocyte activation PRS could be a valuable tool for early disease risk prediction, potentially enabling intervention during the interval between pathogenic amyloid and tau accumulation.

Heart failure (HF) is one of the most common, complex, heterogeneous diseases in the world, with over 1-3% of the global population living with the condition. Progression of HF can be tracked via MRI measures of structural and functional changes to the heart, namely left ventricle (LV), including ejection fraction, mass, end-diastolic volume, and LV end-systolic volume. Moreover, while genome-wide association studies (GWAS) have been a useful tool to identify candidate variants involved in HF risk, they lack crucial tissue-specific and mechanistic information which can be gained from incorporating additional data modalities. This study addresses this gap by incorporating transcriptome-wide and proteome-wide association studies (TWAS and PWAS) to gain insights into genetically-regulated changes in gene expression and protein abundance in precursors to HF measured using MRI-derived cardiac measures as well as full-stage all-cause HF. We identified several gene and protein overlaps between LV ejection fraction and end-systolic volume measures. Many of the overlaps identified in MRI-derived measurements through TWAS and PWAS appear to be shared with all-cause HF. We implicate many putative pathways relevant in HF associated with these genes and proteins via gene-set enrichment and protein-protein interaction network approaches. The results of this study (1) highlight the benefit of using multi-omics to better understand genetics and (2) provide novel insights as to how changes in heart structure and function may relate to HF.

Inflammatory bowel disease (IBD), encompassing Crohn’s disease (CD) and ulcerative colitis (UC), has a significant genetic component and is increasingly prevalent due to environmental factors. Current polygenic risk scores (PRS) have limited predictive power and cannot inform time of symptom onset. Circulating proteomics profiling offers a novel, non-invasive approach for understanding the inflammatory state of complex diseases, enabling the creation of proteomic risk scores (ProRS). This study utilizes data from 51,772 individuals in the UK Biobank to evaluate the unique and combined contributions of PRS and ProRS to IBD risk prediction. We developed ProRS models for CD and UC, assessed their predictive performance over time, and examined the benefits of integrating PRS and ProRS for enhanced risk stratification. Our findings are the first to demonstrate that combining genetic and proteomic data improves IBD incidence prediction, with ProRS providing time-sensitive predictions and PRS offering additional long-term predictive value. We also show that the ProRS achieves better predictive performance among individuals with high PRS. This integrated approach highlights the potential for multi-omic data in precision medicine for IBD.

Gene-environment interaction (GxE) studies provide insights into the interplay between genetics and the environment but often overlook multiple environmental factors’ synergistic effects. This study encompasses the use of environment by environment interaction (ExE) studies to explore interactions among environmental factors affecting lipid phenotypes (e.g., HDL, LDL, and total cholesterol, and triglycerides), which are crucial for disease risk assessment. We developed a novel curated knowledge base, GE.db, integrating genomic and exposomic interactions. In this study, we filtered NHANES exposure variables (available 1999-2018) to identify significant ExE using GE.db. From 101,316 participants and 77 exposures, we identified 263 statistically significant interactions (FDR p < 0.1) in discovery and replication datasets, with 21 interactions significant for HDL-C (Bonferroni p < 0.05). Notable interactions included docosapentaenoic acid (22:5n-3) (DPA) - arachidic acid (20:0), stearic acid (18:0) - arachidic acid (20:0), and blood 2,5-dimethyfuran - blood benzene associated with HDL-C levels. These findings underscore GE.db’s role in enhancing -omics research efficiency and highlight the complex impact of environmental exposures on lipid metabolism, informing future health strategies.

The incidence of chronic kidney disease (CKD) is increasing worldwide, but there is no specific treatment available. Therefore, understanding and controlling the risk factors for CKD are essential for preventing disease occurrence. Salt intake raises blood pressure by increasing fluid volume and contributes to the deterioration of kidney function by enhancing the renin-angiotensin system and sympathetic tone. Thus, a low-salt diet is important to reduce blood pressure and prevent kidney diseases. With recent advancements in genetic research, our understanding of the etiology and genetic background of CKD has deepened, enabling the identification of populations with a high genetic predisposition to CKD. It is thought that the impact of lifestyle or environmental factors on disease occurrence or prevention may vary based on genetic factors. This study aims to investigate whether frequency of adding salt has different effects depending on genetic risk for CKD. CKD polygenic risk scores (PRS) were generated using CKDGen Consortium GWAS (N= 765,348) summary statics. Then we applied the CKD PRS to UK Biobank subjects. A total of 331,318 European individuals aged 40-69 without CKD were enrolled in the study between 2006-2010. The average age at enrollment of the participants in this study was 56.69, and 46% were male. Over an average follow-up period of 8 years, 12,279 CKD cases were identified. The group that developed CKD had a higher percentage of individuals who added salt (46.37% vs. 43.04%) and higher CKD high-risk PRS values compared to the group that did not develop CKD (23.53% vs. 19.86%). We classified the individuals into four groups based on PRS: low (0-19%), intermediate (20-79%), high (80-94%), very high (≥ 95%). Incidence of CKD increased incrementally according to CKD PRS even after adjusting for age, sex, race, Townsend deprivation index, body mass index, estimated glomerular filtration rate, smoking, alcohol, physical activity, diabetes mellitus, dyslipidemia, hypertension, coronary artery diseases, cerebrovascular diseases at baseline. Compared to the “never/rarely” frequency of adding salt group, “always” frequency of adding salt group had an increasing incidence of CKD proportionate to the degree of frequency of adding salt. However, the significant association of “always” group on incident CKD disappeared in the low PRS group. This study validated the signal from PRSs for CKD across a large cohort and confirmed that frequency of adding salt contributes to the occurrence of CKD. Additionally, it confirmed that the effect of frequency of “always” adding salt on CKD incidence is greater in those with more than intermediate CKD-PRS. This study suggests that increased salt intake is particularly concerning for individuals with genetic risk factors for CKD, underscoring the clinical importance of reducing salt intake for these individuals.

Precision medicine significantly enhances patients prognosis, offering personalized treatments. Particularly for metastatic cancer, incorporating primary tumor location into the diagnostic process greatly improves survival rates. However, traditional methods rely on human expertise, requiring substantial time and financial resources. To address this challenge, Machine Learning (ML) and Deep Learning (DL) have proven particularly effective. Yet, their application to medical data, especially genomic data, must consider and encompass privacy due to the highly sensitive nature of data. In this paper, we propose OGHE, a convolutional neural network-based approach for privacy-preserving cancer classification designed to exploit spatial patterns in genomic data, while maintaining confidentiality by means of Homomorphic Encryption (HE). This encryption scheme allows the processing directly on encrypted data, guaranteeing its confidentiality during the entire computation. The design of OGHE is specific for privacy-preserving applications, taking into account HE limitations from the outset, and introducing an efficient packing mechanism to minimize the computational overhead introduced by HE. Additionally, OGHE relies on a novel feature selection method, VarScout, designed to extract the most significant features through clustering and occurrence analysis, while preserving inherent spatial patterns. Coupled with VarScout, OGHE has been compared with existing privacy-preserving solutions for encrypted cancer classification on the iDash 2020 dataset, demonstrating their effectiveness in providing accurate privacy-preserving cancer classification, and reducing latency thanks to our packing mechanism. The code is released to the scientific community.

Multimodal models have become increasingly important as they surpass single-modality approaches on diverse tasks ranging from question-answering to disease diagnosis. Despite the importance of multimodal learning, existing efforts focus on vision-language applications, where the number of modalities rarely exceeds four (images, text, audio, video). However, data in healthcare domain, may include many more modalities like X-rays, PET scans, MRIs, genetic screening, genomic data, and clinical notes, creating a need for both efficient and accurate data integration. Many state-of-the-art multimodal models rely on cross-attention or self-attention for effective data integration, which do not scale well for applications with more than two modalities. The complexity per layer of computing attention in either paradigm is, at best, quadratic with respect to the number of modalities, posing a computational bottleneck that impedes broad adoption. To address this, we propose a new attention mechanism, One-Versus-Others (OvO) attention, that scales linearly with the number of modalities, thus offering a significant reduction in computational complexity compared to existing multimodal attention methods. Using three clinical datasets with multiple diverse modalities, we show that our method decreases computation costs while maintaining or increasing performance compared to popular integration techniques. Across all clinical datasets, OvO reduced the number of required floating point operations (FLOPs) by at least 91.98%, demonstrating its significant impact on efficiency and enabling multi-modal predictions in healthcare.

This PSB 2025 session is focused on opportunities, challenges and solutions for translating Big Data Imaging Genomic findings toward powering decision making in personalized medicine and guiding individual clinical decisions. It combines many of the scientific directions that are of interest to PSB members including Big Data analyses, pattern recognition, machine learning and AI, electronic health records and others.

Malnutrition poses risks regarding cognitive, behavioral, and physical well-being. The aim of this study was to investigate the prevalent health issues associated with malnutrition by utilizing electronic health records (EHR) data. The IBM Watson Health, Explorys platform was used to access the EHR data. Two cohorts were created by two queries; patients with a history of malnutrition (n=5180) and patients without a history of malnutrition diagnosis (n= 413890). The log odds ratio and χ2 statistic were used to identify the statistically significant differences between these two cohorts. We found that there were 35 terms that were more common among the cohort with the malnutrition diagnosis. These terms were categorized under developmental anomalies, infectious agents, respiratory system issues, digestive system issues, pregnancy/prenatal problems, mental, behavioral, or neurodevelopmental disorders, diseases of the ear or mastoid process, diseases of the visual system, and chromosomal anomalies. The management of malnutrition in children is a complex problem that can be addressed with a multifactorial approach. Based on the key themes emerging from among the commonly prevalent terms identified in our study, infection prevention, education in appropriate nutritional solutions for digestive health issues, supportive services to address neurodevelopmental needs, and quality prenatal healthcare would constitute beneficial prevention efforts. Improving our understanding of malnutrition is necessary to develop new interventions for prevention and treatment.

Autism Spectrum Disorder (ASD) encompasses a range of developmental disabilities marked by differences in social functioning, cognition, and behavior. Both genetic and environmental factors are known to contribute to ASD, yet the exact etiological factors remain unclear. Developing integrative models to explore the effects of gene expression on behavioral and cognitive traits attributed to ASD can uncover environmental and genetic interactions. A notable aspect of ASD research is the sex-wise diagnostic disparity: males are diagnosed more frequently than females, which suggests potential sex-specific biological influences. Investigating neuronal microstructure, particularly axonal conduction velocity offers insights into the neural basis of ASD. Developing robust models that evaluate the vast multidimensional datasets generated from genetic and microstructural processing poses significant challenges. Traditional feature selection techniques have limitations; thus, this research aims to integrate principal component analysis (PCA) with supervised machine learning algorithms to navigate the complex data space. By leveraging various neuroimaging techniques and transcriptomics data analysis methods, this methodology builds on traditional implementations of PCA to better contextualize the complex genetic and phenotypic heterogeneity linked to sex differences in ASD and pave the way for tailored interventions.

Alzheimer’s Disease and Related Dementias (ADRD) afflict almost 7 million people in the USA alone. The majority of research in ADRD is conducted using post-mortem samples of brain tissue or carefully recruited clinical trial patients. While these resources are excellent, they suffer from lack of sex/gender, and racial/ethnic inclusiveness. Electronic Health Records (EHR) data has the potential to bridge this gap by including real-world ADRD patients treated during routine clinical care. In this study, we utilize EHR data from a cohort of 70,420 ADRD patients diagnosed and treated at Penn Medicine. Our goal is to uncover important risk features leading to three types of Neuro-Degenerative Disorders (NDD), including Alzheimer’s Disease (AD), Parkinson’s Disease (PD) and Other Dementias (OD). We employ a variety of Machine Learning (ML) Methods, including uni-variate and multivariate ML approaches and compare accuracies across the ML methods. We also investigate the types of features identified by each method, the overlapping features and the unique features to highlight important advantages and disadvantages of each approach specific for certain NDD types. Our study is important for those interested in studying ADRD and NDD in EHRs as it highlights the strengths and limitations of popular approaches employed in the ML community. We found that the uni-variate approach was able to uncover features that were important and rare for specific types of NDD (AD, PD, OD), which is important from a clinical perspective. Features that were found across all methods represent features that are the most robust.

Illness related brain effects of neuropsychiatric disorders are not regionally uniform, with some regions showing large pathological effects while others are relatively spared. Presently, Big Data meta-analytic studies tabulate these effects using structural and/or functional brain atlases that are based on the anatomical boundaries, landmarks and connectivity patterns in healthy brains. These patterns are then translated to individual level predictors using approaches such as Regional Vulnerability Index (RVI), which quantifies the agreement between individual brain patterns and the canonical pattern found in the illness. However, the atlases from healthy brains are unlikely to align with deficit pattern expressed in specific disorders such as Major Depressive Disorder (MDD), thus reducing the statistical power for individualized predictions. Here, we evaluated a novel approach, where disorder specific templates are constructed using the Kullback-Leibler (KL) distance to balance granularity, signal-to-noise ratio and the contrast between regional effect sizes to maximize translatability of the population-wide illness pattern at the level of the individual. We used regional homogeneity (ReHo) maps extracted from resting state functional MRI for N = 2, 289 MDD sample (mean age ± s.d.: 63.2 ± 7.2 years) and N = 6104 control subjects (mean age ± s.d.: 62.9 ± 7.2 years) who were free of MDD and any other mental condition. The cortical effects of MDD were analyzed on the 3D spherical surfaces representing cerebral hemispheres. KL-distance was used to organize the cortical surface into 28 regions of interest based on effect sizes, connectivity and signal-to-noise ratio. The RVI values calculated using this novel approach showed significantly higher effect size of the illness than these calculated using standard Desikan brain atlas.

Alzheimer’s disease (AD) is a neurodegenerative disorder that results in progressive cognitive decline but without any clinically validated cures so far. Understanding the progression of AD is critical for early detection and risk assessment for AD in aging individuals, thereby enabling initiation of timely intervention and improved chance of success in AD trials. Recent pseudotime approach turns cross-sectional data into “faux” longitudinal data to understand how a complex process evolves over time. This is critical for Alzheimer, which unfolds over the course of decades, but the collected data offers only a snapshot. In this study, we tested several state-of-the-art pseudotime approaches to model the full spectrum of AD progression. Subsequently, we evaluated and compared the pseudotime progression score derived from individual imaging modalities and multi-modalities in the ADNI cohort. Our results showed that most existing pseudotime analysis tools do not generalize well to the imaging data, with either flipped progression score or poor separation of diagnosis groups. This is likely due to the underlying assumptions that only stand for single cell data. From the only tool with promising results, it was observed that all pseudotime, derived from either single imaging modalities or multi-modalities, captures the progressiveness of diagnosis groups. Pseudotime from multi-modality, but not the single modalities, confirmed the hypothetical temporal order of imaging phenotypes. In addition, we found that multi-modal pseudotime is mostly driven by amyloid and tau imaging, suggesting their continuous changes along the full spectrum of AD progression.

Alzheimer’s disease (AD) is a neurocognitive disorder that deteriorates memory and impairs cognitive functions. Mild Cognitive Impairment (MCI) is generally considered as an intermediate phase between normal cognitive aging and more severe conditions such as AD. Although not all individuals with MCI will develop AD, they are at an increased risk of developing AD. Diagnosing AD once strong symptoms are already present is of limited value, as AD leads to irreversible cognitive decline and brain damage. Thus, it is crucial to develop methods for the early prediction of AD in individuals with MCI. Recurrent Neural Networks (RNN)-based methods have been effectively used to predict the progression from MCI to AD by analyzing electronic health records (EHR). However, despite their widespread use, existing RNN-based tools may introduce increased model complexity and often face difficulties in capturing long-term dependencies. In this study, we introduced a novel Dynamic deep learning model for Early Prediction of AD (DyEPAD) to predict MCI subjects’ progression to AD utilizing EHR data. In the first phase of DyEPAD, embeddings for each time step or visit are captured through Graph Convolutional Networks (GCN) and aggregation functions. In the final phase, DyEPAD employs tensor algebraic operations for frequency domain analysis of these embeddings, capturing the full scope of evolutionary patterns across all time steps. Our experiments on the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and National Alzheimer’s Coordinating Center (NACC) datasets demonstrate that our proposed model outperforms or is in par with the state-of-the-art and baseline methods.

There is a disconnect between data practices in biomedicine and public understanding of those data practices, and this disconnect is expanding rapidly every day (with the emergence of synthetic data and digital twins and more widely adopted Artificial Intelligence (AI)/Machine Learning tools). Transparency alone is insufficient to bridge this gap. Concurrently, there is an increasingly complex landscape of laws, regulations, and institutional/ programmatic policies to navigate when engaged in biocomputing and digital health research, which makes it increasingly difficult for those wanting to “get it right” or “do the right thing.” Mandatory data protection obligations vary widely, sometimes focused on the type of data (and nuanced definition and scope parameters), the actor/entity involved, or the residency of the data subjects. Additional challenges come from attempts to celebrate biocomputing discoveries and digital health innovations, which frequently transform fair and accurate communications into exaggerated hype (e.g., to secure financial investment in future projects or lead to more favorable tenure and promotion decisions). Trust in scientists and scientific expertise can be quickly eroded if, for example, synthetic data is perceived by the public as “fake data” or if digital twins are perceived as “imaginary” patients. Researchers appear increasingly aware of the scientific and moral imperative to strengthen their work and facilitate its sustainability through increased diversity and community engagement. Moreover, there is a growing appreciation for the “data work” necessary to have scientific data become meaningful, actionable information, knowledge, and wisdom–not only for scientists but also for the individuals from whom those data were derived or to whom those data relate. Equity in the process of biocomputing and equity in the distribution of benefits and burdens of biocomputing both demand ongoing development, implementation, and refinement of embedded Ethical, Legal and Social Implications (ELSI) research practices. This workshop is intended to nurture interdisciplinary discussion of these issues and to highlight the skills and competencies all too often considered “soft skills” peripheral to other skills prioritized in traditional training and professional development programs. Data scientists attending this workshop will become better equipped to embed ELSI practices into their research.

Biobanks hold immense potential for genomic research, but fragmented data and incompatible tools slow progress. This workshop equipped participants with Nextflow, a powerful workflow language to streamline bioinformatic analyses across biobanks. We taught participants to write code in their preferred language and demonstrated how Nextflow handles the complexities, ensuring consistent, reproducible results across different platforms. This interactive session was ideal for beginner-to-intermediate researchers who want to (1) Leverage biobank data for genomic discoveries, (2) Build portable and scalable analysis pipelines, (3) Ensure reproducibility in their findings, (4) Gain hands-on experience through presentations, demonstrations, tutorials, and discussions with bioinformatics experts.

Large Language Models (LLMs) have shown significant promise across a wide array of fields, including biomedical research, but face notable limitations in their current applications. While they offer a new paradigm for data analysis and hypothesis generation, their efficacy in computational biology trails other applications such as natural language processing. This workshop addresses the state of the art in LLMs, discussing their challenges and the potential for future development tailored to computational biology. Key issues include difficulties in validating LLM outputs, proprietary model limitations, and the need for expertise in critical evaluation of model failure modes.

Large language models (LLMs) and biomedical annotations have a symbiotic relationship. LLMs rely on high-quality annotations for training and/or fine-tuning for specific biomedical tasks. These annotations are traditionally generated through expensive and time-consuming human curation. Meanwhile LLMs can also be used to accelerate the process of curation, thus simplifying the process, and potentially creating a virtuous feedback loop. However, their use also introduces new limitations and risks, which are as important to consider as the opportunities they offer. In this workshop, we will review the process that has led to the current rise of LLMs in several fields, and in particular in biomedicine, and discuss specifically the opportunities and pitfalls when they are applied to biomedical annotation and curation.

The following sections are included:

• Overview
• Background & key terms
• Earth Friendly Computation 574: Indigenous Data Sovereignty, Circular Systems, and Solarpunk Solutions for a Sustainable Future
• AI in Point-of-Care: A Sustainable Healthcare Revolution at the Edge
• Conclusion: The Future of Earth Friendly Computation
• Acknowledgments
• References

Recent advancements in Artificial Intelligence (AI) and data center infrastructure have brought the global cloud computing market to the forefront of conversations about sustainability and energy use. Current policy and infrastructure for data centers prioritize economic gain and resource extraction, inherently unsustainable models which generate massive amounts of energy and heat waste. Our team proposes the formation of policy around earth-friendly computation practices rooted in Indigenous models of circular systems of sustainability. By looking to alternative systems of sustainability rooted in Indigenous values of aloha ‘āina, or love for the land, we find examples of traditional ecological knowledge (TEK) that can be imagined alongside Solarpunk visions for a more sustainable future. One in which technology works with the environment, reusing electronic waste (e-waste) and improving data life cycles.

This paper examines the integration of artificial intelligence (AI) in point-of-care testing (POCT) to enhance diagnostic speed, accuracy, and accessibility, particularly in underserved regions. AI-driven POCT is shown to optimize clinical decision-making, reduce diagnostic times, and offer personalized healthcare solutions, with applications in genome sequencing and infectious disease management. The paper highlights the environmental challenges of AI, including high energy consumption and electronic waste, and proposes solutions such as energy-efficient algorithms and edge computing. It also addresses ethical concerns, emphasizing the reduction of algorithmic bias and the need for equitable access to AI technologies. While AI in POCT can improve healthcare and promote sustainability, collaboration within the POCT ecosystem—among researchers, healthcare providers, and policymakers—is essential to overcome the ethical, environmental, and technological challenges.

Polygenic risk scores (PRS) have predominantly been derived from genome-wide association studies (GWAS) conducted in European ancestry (EUR) individuals. In this study, we present an in-depth evaluation of PRS based on multi-ancestry GWAS for five cardiometabolic phenotypes in the Penn Medicine BioBank (PMBB) followed by a phenome-wide association study (PheWAS). We examine the PRS performance across all individuals and separately in African ancestry (AFR) and EUR ancestry groups. For AFR individuals, PRS derived using the multi-ancestry LD panel showed a higher effect size for four out of five PRSs (DBP, SBP, T2D, and BMI) than those derived from the AFR LD panel. In contrast, for EUR individuals, the multi-ancestry LD panel PRS demonstrated a higher effect size for two out of five PRSs (SBP and T2D) compared to the EUR LD panel. These findings underscore the potential benefits of utilizing a multi-ancestry LD panel for PRS derivation in diverse genetic backgrounds and demonstrate overall robustness in all individuals. Our results also revealed significant associations between PRS and various phenotypic categories. For instance, CAD PRS was linked with 18 phenotypes in AFR and 82 in EUR, while T2D PRS correlated with 84 phenotypes in AFR and 78 in EUR. Notably, associations like hyperlipidemia, renal failure, atrial fibrillation, coronary atherosclerosis, obesity, and hypertension were observed across different PRSs in both AFR and EUR groups, with varying effect sizes and significance levels. However, in AFR individuals, the strength and number of PRS associations with other phenotypes were generally reduced compared to EUR individuals. Our study underscores the need for future research to prioritize 1) conducting GWAS in diverse ancestry groups and 2) creating a cosmopolitan PRS methodology that is universally applicable across all genetic backgrounds. Such advances will foster a more equitable and personalized approach to precision medicine.