Duke University

In a recent paper, we proposed a Personalized Patient Preference Predictor (P4), building on earlier work by Rid, Wendler, and colleagues. The P4 is a hypothetical computer program that would, in the context of surrogate decision-making (e.g., following a substituted judgment standard), use generative AI models to infer a patient’s underlying values and preferences and, on that basis, predict which treatment option they would choose in the current situation. Such AI models, we suggested, could be “fine-tuned” on various preexisting data or materials produced by, or otherwise pertaining to (e.g., including information about), the patient considered as a unique individual.

The discovery of broadly protective antibodies to the influenza virus neuraminidase (NA) has raised interest in NA as a vaccine target. However, recombinant, solubilized tetrameric NA ectodomains are often challenging to express and isolate, hindering the study of anti-NA humoral responses. To address this obstacle, we established a panel of 22 non-adherent cell lines stably expressing native, historical N1, N2, N3, N9, and NB NAs anchored on the cell surface. The cell lines are barcoded with fluorescent proteins, enabling high-throughput, 16-plex analyses of antibody binding with commonly available flow cytometers. The cell lines were at least as efficient as a Luminex multiplex binding assay at identifying NA antibodies from a library of unselected clonal IgGs derived from human memory B cells. The cell lines were also useful for measuring the magnitude and breadth of the serum antibody response elicited by experimental infection of rhesus macaques with influenza virus. The membrane-anchored NAs are catalytically active and are compatible with established sialidase activity assays. NA-expressing K530 cell lines therefore represent a useful tool for studying NA immunity and evaluating influenza vaccine efficacy.

The connectome, a map of the structural and/or functional connections in the brain, provides a complex representation of the neurobiological phenotypes on which it supervenes. This information-rich data modality has the potential to transform our understanding of the relationship between patterns in brain connectivity and neurological processes, disorders, and diseases. However, existing computational techniques used to analyze connectomes are often insufficient for interrogating multi-subject connectomics datasets: many current methods are either solely designed to analyze single connectomes or leverage heuristic graph statistics that are unable to capture the complete topology of multiscale connections between brain regions. To enable more rigorous connectomics analysis, we introduce a set of robust and interpretable statistical hypothesis tests motivated by recent theoretical advances in random graph models. These tests facilitate simultaneous analysis of multiple connectomes across different scales of network topology, enabling the robust and reproducible discovery of hierarchical brain structures that vary in relation to phenotypic profiles. In addition to explaining the theoretical foundations and guarantees of our algorithms, we demonstrate their superiority over current state-of-the-art connectomics methods through extensive simulation studies and real-data experiments. Using a set of high-resolution connectomes obtained from genetically distinct mouse strains (including the BTBR mouse—a standard model of autism—and three behavioral wild-types), we illustrate how our methods successfully uncover latent information in multi-subject connectomics data and yield valuable insights into the connective correlates of neurological phenotypes that other methods do not capture. The data and code necessary to reproduce the analyses, simulations, and figures presented in this work are available at https://github.com/neurodata/MCC.

Qualitative research, rooted in interpretivism, is valuable for studying immigrant populations and understanding cultural influences on health behaviors. However, few studies have explored the methodological challenges of researching older Korean U.S. immigrants, particularly on sensitive topics like end-of-life care, which requires deep cultural understanding. This paper examines the challenges encountered during a pilot study on end-of-life care among older Korean U.S. immigrants. In addition to identifying key methodological obstacles, we highlight strategies to improve future research on sensitive topics within immigrant communities. Our study, informed by existing literature, faced unexpected challenges at every stage—recruitment, data collection, and analysis—each requiring careful adaptation. First, recruitment posed significant challenges. Many participants were hesitant to discuss end-of-life care due to cultural stigma, fearing it might invite misfortune. Some resisted signing consent forms, unfamiliar with Western research protocols and concerned about potential consequences. Others expected structured surveys rather than open-ended interviews, making engagement difficult. Second, conducting interviews brought additional hurdles. The setting needed to feel neutral, as medical or religious environments influenced responses. Language proficiency varied, requiring interpreters and adjusted phrasing. Discussions about end-of-life care sometimes triggered emotional distress, necessitating sensitivity and frequent check-ins. Third, data analysis required careful consideration. Translating nuanced Korean expressions into English was challenging, as some terms lacked direct equivalents. Case vignettes also needed thoughtful adaptation to ensure cultural relevance and avoid bias. Finally, participant feedback led to important revisions, reinforcing the value of involving participants throughout the research process. Engaging them early and reflecting on challenges afterward can improve study design and data quality. By addressing these methodological hurdles, this study provides practical insights for strengthening qualitative research on immigrant populations, particularly when exploring sensitive topics.

Dry biomaterial scaffolds (Drydux) are a novel platform that enhances viral transduction ~10-fold while simplifying therapeutic cell production. Here, we systematically evaluate the biomaterial properties governing transduction enhancement.

Background Previous efforts to translate vagus nerve stimulation (VNS) therapies from preclinical studies to human clinical applications (e.g., for stroke, heart failure, and inflammatory diseases) did not account for individual- or species-specific differences in nerve responses when selecting stimulation parameters. Lack of explicit consideration for producing equivalent nerve responses could contribute to clinical outcomes not replicating promising results from preclinical animal studies. Methods We used models of VNS built with ASCENT (Musselman, PLoS Comput Biol 17:e1009285, 2021) to quantify nerve responses across species and simulate translation of VNS therapies via either recycling or linear scaling of stimulation parameters. For humans ( n = 9) and pigs ( n = 12), we used previously validated computational models with the standard clinical helical cuff electrode on individual-specific nerve morphologies (Musselman, J Neural Eng 20:acda64, 2023b). We also modeled rat VNS ( n = 9) with the Micro-Leads Neuro bipolar cuff. We calculated thresholds for fiber activation (A-, B-, and C-fibers) with biphasic rectangular pulses (0.13, 0.25, 0.5 ms). We defined “K” as the ratio of activation thresholds between a pair of individuals. We used a mixed model ANOVA on the natural logarithm of K to test for differences in inter-species Ks across fiber types and pulse widths. Lastly, using the same nerve morphologies and application-specific device design (cuff and waveform), we developed models to predict nerve responses in chronic human and rat VNS studies for treatment of stroke, inflammation, and heart failure. Results Depending on the individual and species, the activation amplitude required to produce a given nerve response varied widely. Thus, applying the same VNS parameters across individuals within a species produced a large range of nerve responses. Further, applying the same or linearly scaled stimulation amplitudes across species also produced highly variable responses. Ks were greater for B fibers than A fibers ( p < 0.0001) and decreased with longer pulse widths ( p < 0.0001 between consecutive pairs). Conclusions The results highlight the need for systematic approaches to select stimulation parameters that account for individual- and species-specific differences in nerve responses to stimulation. Such parameter tuning may lead to higher response rates and greater therapeutic benefits from VNS therapies.

In Zanzibar City - the capital of the Zanzibar archipelago in Tanzania - the incidence of malaria has decreased over the past few decades due to standardized treatment protocols and public health interventions targeting adult mosquitoes. However, the incidence remains between 1–2%, and case numbers have increased over the past few years because of a continued influx of Plasmodium spp. from other malaria-endemic areas (including mainland Tanzania). Larviciding is a powerful tool to target mosquito populations and reduce the incidence of malaria. However, larvicidal strategies rely on knowledge of the breeding patterns of malaria vector mosquitoes. In Zanzibar City, no larval surveys have been done in the last few years. Our aim was to characterize Anopheles spp. breeding sites in Zanzibar City during the rainy season. We first conducted systematic larval surveys across 16 semi-permanent/permanent water bodies and 30 temporary water bodies. Then, we used principal component analysis and logistic regression to model the effects of physical/chemical parameters and rainfall on Anopheles presence. We found that Anopheles spp. prefer concrete, semi-permanent breeding sites with high levels of dissolved oxygen but are also found in natural sites after heavy rains. Our logistic regression model successfully predicted the presence of Anopheles larvae, achieving a positive predictive power of 65.7% and a negative predictive power of 88.8%. The data from our study suggest that Anopheles spp. have not yet adapted to more polluted breeding sites in Zanzibar City (as they have in some mainland locations). These results can inform targeted larvicidal strategies in Zanzibar City.

The susceptibility of a protein to aggregation upon exposure to copper ions (Cu) has been recognized as a contributor to Cu‐induced cellular dysfunction and toxicity. Different cell types succumb to Cu to varying degrees, indicating innate differences between species in the mechanisms used to tolerate exposure to Cu in excess of their biological needs. Investigated here are properties associated with metal‐induced protein precipitation (MiPP) compared across cell lysates generated from three cell lines from three different species: Escherichia coli, Candida albicans, and the human prostate cancer cell line 22Rv1. The human cell line was the most sensitive to Cu‐induced protein precipitation, while C. albicans was the most tolerant. This trend aligns with the relative susceptibilities of these cells to Cu‐induced cytotoxicity. The unique susceptibilities of these proteomes to precipitation by Cu were examined to identify factors that influence a protein's relative sensitivity to this effect. Identified were intrinsic factors such as frequency and solvent accessibility of known metal‐binding amino acids, as well as external factors related to the molecular composition of their native cell lysates. Overall, our findings help to elucidate the biomolecular basis underpinning the unique capacity of adventitious Cu to have differential effects on eukaryotic and prokaryotic organisms and the level of Cu needed to induce protein precipitation.

Transcription factors carry long intrinsically disordered regions often containing multiple activation domains. Despite numerous recent high‐throughput identifications and characterizations of activation domains, the interplay between sequence motifs, activation domains, and regulator binding in intrinsically disordered transcription factor regions remains unresolved. Here, we map sequence motifs and activation domains in an Arabidopsis thaliana NAC transcription factor clade, revealing that although sequence motifs and activation domains often coincide, no systematic overlap exists. Biophysical analyses using NMR spectroscopy show that the long intrinsically disordered region of senescence‐associated transcription factor ANAC046 is devoid of residual structure. We identify two activation domain/sequence motif regions, one at each end that both bind a panel of six positive and negative regulator domains from biologically relevant regulators promiscuously. Binding affinities measured using isothermal titration calorimetry reveal a hierarchy for regulator binding of the two ANAC046 activation domain/sequence motif regions defining these as regulatory hotspots. Despite extensive dynamic intramolecular contacts along the disordered chain revealed using paramagnetic relaxation enhancement experiments and simulations, the regions remain uncoupled in binding. Together, the results imply rheostatic regulation by ANAC046 through concentration‐dependent regulator competition, a mechanism likely mirrored in other transcription factors with distantly located activation domains.

Background Emerging research suggests that lesbian, gay, bisexual, and queer (LGBQ) women face barriers to breast cancer screening. The authors sought to quantify sexual identity disparities in mammography screening, health care access, and lifestyle‐related risk factors using two national surveys. Methods Data from the 2018, 2019, and 2021 National Health Interview Survey (NHIS) and the 2018, 2020, and 2022 Behavioral Risk Factor Surveillance System (BRFSS) survey were analyzed. The authors performed meta‐analyses to determine the relative risks (RRs) of self‐reported, up‐to‐date mammography for women identifying as LGBQ versus those identifying as straight. Differences in health care access and lifestyle‐related breast cancer risk factors were also assessed by sexual identity. Results LGBQ women reported lower up‐to‐date mammography (pooled RR [pRR], 0.95; 95% confidence interval [CI], 0.92–0.98) versus straight women, driven by differences among bisexual/queer women (pRR, 0.91; 95% CI, 0.87–0.95) and those entering screen‐eligibility at ages 40–49 years (pRR, 0.86; 95% CI, 0.80–0.91) and 50–59 years (pRR, 0.93; 95% CI, 0.88–0.98). LGBQ women were more likely than straight women to be uninsured (BRFSS survey [8.6%; 95% CI, 6.5%–11.2%] vs. NHIS [5.1%; 95% CI, 4.8%–5.4%]) and to experience financial barriers to care (BRFSS survey [13.8%; 95% CI, 11.6%–16.3%] vs. NHIS [8.9%; 8.5%–9.2%]). Lifestyle‐related breast cancer risk factors were more common among LGBQ women versus straight women, including current smoking (BRFSS survey [19.0%; 17.1%–21.2%] vs. NHIS [13.9%; 13.6%–14.3%]). Conclusions LGBQ women were more likely than straight women to be exposed to breast cancer risk factors, which were compounded by lower screening and facing health care access barriers. It is crucial to identify interventions for screening and risk reduction that are accessible and effective for LGBQ women, particularly bisexual/queer women and those aging into screen‐eligibility.

Due to the high volatility of financial markets and the prevalence of financial fraud, real-time stock market forecasting for listed companies remains a challenging task. To address these challenges, this study proposes TCNAttention-RAG, a hybrid deep learning framework integrating Temporal Convolutional Network (TCN), Multi-Layer Perceptron (MLP), Attention Mechanism, and Retrieval-Augmented Generation (RAG) for enhanced stock price forecasting. The model leverages TCN for temporal feature extraction, MLP for nonlinear representation, and Attention for feature weighting, while RAG dynamically retrieves key financial insights from corporate reports to improve predictive accuracy. Using NASDAQ-listed stock price data (20142020), combined with corporate financial reports, market transaction data, and macroeconomic indicators, a multi-dimensional dataset is constructed. Experimental results demonstrate that TCNAttention-RAG outperforms traditional models in accuracy and recall, effectively capturing stock price fluctuations. Despite its limitations in handling extreme market events, the model exhibits high reliability and predictive robustness. This study introduces a multi-modal data-driven approach to financial forecasting, offering insights into intelligent financial analysis and enhancing decision-making in volatile markets.

Background Sequential multivalent immunizations are used to counter diversity in rapidly mutating viruses. Here, we evaluated the effect of HIV-1 immunogen formats on the binding profile of memory B-cells elicited in two independent Rhesus macaque trials. Methods In one trial, female Rhesus macaques were immunized with a multiclade HIV-1 gp120 envelope glycoprotein (Env) cocktail and bled two weeks post final immunization. In another trial, male and female Rhesus macaques were sequentially immunized with clonally-related Env glycoproteins: Four immunogens were administered as non-stabilized gp140 Envs and the fifth as a specially stabilized gp140 Env trimer (SOSIP); animals were bled before and after SOSIP immunization. Immunogen-binding peripheral memory B-cells were sorted and cultured at limiting dilution. Culture supernatants were assessed by ELISA for binding to individual immunogens. Results In the first trial, 81% (591/734) of B-cells cross-react with multiple Envs and most bind to all immunogens. In the second trial, 81% (331/410) of B-cells isolated before SOSIP administration react with all non-stabilized gp140 Env immunogens and 27% also cross-react with the yet-to-be-administered SOSIP-stabilized Env. However, after SOSIP administration, SOSIP-stabilized trimer-reactive B-cells increase to 86% (219/256) but most (82%) do not cross-react with the preceding immunogens. Conclusions Multiclade and sequential regimens before SOSIP-stabilized Env immunization elicited B-cells that converge on shared epitopes. A change in immunogen format results in a divergent B-cell response that vastly fails to engage prior responses. Critically, B-cell priming with non-stabilized Env cannot modify the effect of the epitope immunodominance hierarchy in a SOSIP trimer. These results suggest that a change in immunogen format may cause off-target B-cell engagement, but also that B-cell repriming is possible despite pre-existing immunity.

PURPOSE Genomic testing for prostate cancer (PCa) clinical management and hereditary cancer assessment has grown in clinical impact; however, challenges remain regarding optimal implementation and end-user confidence. The Decision-making, Experience, and Confidence In Determining Genomic Evaluation (DECIDE) survey was designed to collect information regarding utility and understanding of genomic testing from PCa health care providers, researchers, and stakeholders. METHODS The DECIDE survey was administered online from October 2022 to January 2023 with 18 multiple-response questions. Survey domains included self-confidence with ordering and interpreting germline and somatic genomic tests, process of testing and use of results, decision-making factors, and barriers to testing. Data were summarized by evaluating counts and percentages of responses, and the results were presented by descriptive statistics. RESULTS One hundred twenty-two participants completed the survey. The majority were medical oncologists (70%) and at academic medical centers (89%). Self-confidence was high in knowing indications for genomic testing (82% respondents) but lower in interpretation of results, especially from circulating tumor DNA (52%). Confidence varied in interpreting pathogenic variants (65% high confidence), variants of unknown significance (47%), and incidental findings from genomic tests (35%). Common barriers to testing were difficulty obtaining tissue (71%) and cost (35%). Testing utility was sometimes limited by inability to obtain the recommended treatment (33%). Most of the respondents (55%) agreed that lack of education and training of health care professionals regarding genomic testing is impeding clinical translation. CONCLUSION The DECIDE survey provided critical insights into challenges with genomic testing, from provider confidence in interpretating results to testing and practice barriers. The results inform next steps to further educate PCa providers and to collectively improve testing and result reporting for enhanced implementation of PCa genomic testing.

A leading HIV vaccine strategy requires a priming immunogen to induce broadly neutralizing antibody (bnAb) precursors, followed by a series of heterologous boosters to elicit somatic hypermutation (SHM) and produce bnAbs. In two randomized, open-label phase 1 human clinical trials, IAVI-G002 in the United States and IAVI-G003 in Rwanda and South Africa, we evaluated the safety and immunogenicity of mRNA-encoded nanoparticles as priming immunogens (both trials) and first-boosting immunogens (IAVI-G002). The vaccines were generally safe and well tolerated, except 18% of IAVI-G002 participants experienced skin reactions. Priming induced bnAb precursors with substantial frequencies and SHM, and heterologous boosting elicited increased SHM, affinity, and neutralization activity toward bnAb development. The results establish clinical proof of concept that heterologous boosting can advance bnAb-precursor maturation and demonstrate bnAb priming in Africa where the HIV burden is highest.

Industrial statistics grew up in an era when manufacturing was the primary engine of commerce. Today, the driver is information technology. This paper discusses how statisticians need to adapt to contribute to this new business model, with particular emphasis upon computational advertising, autonomous vehicles, operations management, and large language models. Remarkably, many of our old tools are still relevant, even as the new problem space poses fresh research challenges for our employment and educational systems.