Adolescents encounter large volumes of non-core (i.e., nutritionally poor and unhealthy) food messages on social media. The relationship between these messages and eating is a product of both the volume of exposure to the message (e.g., frequent exposure) and the various sources (i.e., messengers) relaying each of these messages. This study investigates and compares how adolescents' perceived volume of exposure to food messages from different social media messenger categories (peers, influencers, celebrities, food brands, and health organizations) is associated with their eating outcomes. A cross-sectional survey of 1,002 adolescents (aged 11-19 years) showed that adolescents report significantly higher exposure to non-core food messages posted by peers, celebrities, influencers, and brands compared to core food messages, and that greater exposure to such messages is significantly associated with higher non-core food liking, norms and/or intake. Only exposure to core food messages posted by health organizations was significantly associated with core food consumption. These findings highlight the importance of food marketing regulations particularly those addressing peer and influencer driven content. They also highlight the significance of peers and health organizations, as messenger categories, in social media food communication to adolescents.
Perceived stimulus intensity is a core feature of sensory experience, yet how it emerges in the human olfactory system remains unknown. Here, we demonstrate that oscillatory dynamics in the human olfactory bulb (OB) and piriform cortex (PC) primarily encode subjective perceived intensity rather than physical concentration. Using noninvasive electrobulbogram recordings, we show that early gamma-band activity in the OB reflects bottom-up transmission of perceived intensity to the PC, which in turn sends top-down beta-band feedback that modulates OB activity via phase-amplitude coupling and transient beta bursts. This bidirectional communication supports a dynamic updating mechanism that maintains perceptual constancy across varying environmental odor concentrations. Our findings reveal a previously uncharacterized oscillatory framework for intensity coding in the human olfactory system, highlighting the primacy of perception over stimulus properties and offering a mechanistic basis for predictive processing in early sensory circuits.
High-throughput screening workflows often rank materials with a sequence of filters, but a single sequence can hide how strongly the final ranking depends on the chosen physical approximations. Here a multiscale cascade means an ordered workflow in which outputs from electronic, lattice, microstructural, uncertainty, and device-level models are passed from one layer to the next; the layers are theory or surrogate-model steps, not layers of LLMs. We use a large language model (LLM) as a workflow-design assistant to assemble candidate model stacks from the literature, after which the equations, code, and physical handoffs are inspected and implemented by the author. The test case is thermoelectric screening, where the dimensionless figure of merit ZT = S2σT/(κe + κL) combines the Seebeck coefficient S, electrical conductivity σ, electronic thermal conductivity κe, and lattice thermal conductivity κL. Two independently assembled eight-layer cascades are applied to the same 314-compound vacancy-containing chalcogenide library. LLM1 uses Fan-Migdal band gap renormalization, a Kubo-DMFT transport surrogate, and a literature-trained Gaussian process for κL. LLM2 replaces those three early layers with Bose-Einstein band gap renormalization, acoustic-phonon Boltzmann transport, and a Debye-Callaway integral. The common four-stage screen sends 15 compounds to full evaluation in each cascade: LLM1 selects tellurides headed by CuAlTe2 (ZTpeak = 8.61), whereas LLM2 selects nonoverlapping sulfides headed by CuPb2S4 (ZTpeak = 0.325). The absolute LLM1 values are not claimed as validated performance forecasts: CuAlTe2 is literature-supported as a promising bulk thermoelectric, but reported and expected values are closer to ZT ≲ 2 than to 8.6. This study underscores the importance of transparently comparing multiple workflow designs to understand the sensitivity and reliability of high-throughput screening outcomes in materials discovery.
Deep brain stimulation (DBS) of the globus pallidus internus (GPi) is an established therapy for medically refractory childhood dystonia. Achieving optimal outcomes requires repeated programming by experienced multidisciplinary teams in specialized centers. Geographic distance, travel burden, and access to expertise pose major barriers to families long-term care. This review summarizes literature and clinical experience on remote DBS programming for pediatric dystonia through March 2026 identified using targeted searches of PubMed and Embase. Telemedicine-enabled platforms allow clinicians to adjust stimulation parameters through secure interfaces while conducting real-time video assessments. Feasibility studies and early clinical applications indicate that remote programming is technically achievable, safe, and capable of delivering meaningful therapeutic adjustments without in-person visits. For children, who often require early programming, remote care can reduce travel, minimize disruption to school and family life, and facilitate therapy optimization. Challenges remain, including regulatory requirements, infrastructure needs, and development of standardized pediatric protocols. Remote DBS programming is a promising strategy to expand access to specialized care for children with dystonia. As telemedicine technologies, digital motor assessments, and neuromodulation platforms advance, remote programming has potential to become an integral component of long-term pediatric DBS management, improving outcomes and easing families' burden. Deep brain stimulation (DBS) is a treatment used for children with severe dystonia, a movement disorder that causes abnormal postures and involuntary movements. In DBS, small electrodes are implanted in the brain and connected to a device that sends electrical signals to help control symptoms. After surgery, doctors must adjust the device settings many times over months to achieve the best results. These adjustments are usually done at a small number of specialized centers, making it difficult for families who live far away to access care.Remote DBS programming is a newer approach that allows doctors to adjust the device from a distance using secure video and internet-based systems. During these virtual visits, clinicians can observe the child’s movements, change stimulation settings, and monitor for side effects in real-time. This approach may reduce the need for frequent travel, lower costs, and make it easier for families to access expert care.Studies in adults with movement disorders have shown that remote DBS is safe and works as well as in-person visits. Early experience suggests it is also feasible in children, although more research is needed. There are still challenges, including ensuring reliable internet access, protecting patient data, and developing guidelines specifically for children.In the future, remote DBS may become a routine part of care, combined with occasional in-person visits. New technologies, such as wearable devices that track movement, may further improve how treatment is monitored and adjusted over time.
The rapid development of wearable health tools has made it possible to continuously monitor physiological conditions for preventive care. However, stringent privacy laws, including HIPAA and GDPR, require decentralized methods such as federated learning (FL) to safeguard personal patient information. Nonetheless, empirical profiling in this paper finds that typical FL implementations are plagued by a serious performance trilemma; a naive federated model attains a 35.3 percent energy savings (3.84 vs. 5.93 kJ in the centralized models), but at the cost of a disastrous performance penalty of 13.87 percentage points (84.94 vs. 98.81 percent in centralized models). The failure in research is largely due to the on-device computational load of 4.24 MFLOPs per training sample, resulting in a "straggler" bottleneck that increases the total training duration to 1,066.26 s, almost 70 times longer than centralized training. As a result, the introduction of the hybrid hierarchical federated split learning (H-FedSL) architecture helps in strategically splitting the neural network at a cut layer to divide the workload between wearable and nearby edge servers. The methodology provides a new framework that offloads the heavy and deep-layer computations to the edge server, leaving the shallow feature extraction to the point of operation, and sends only privacy-sensitive abstractions of the smashed data, rather than raw signals. The integration of asynchronous protocols will help manage device heterogeneity and resource-aware client selection, thereby achieving the aim of H-FedSL to restore the gold-standard accuracy of 98.81% with the state-of-the-art 35.3% energy efficiency of the federated model. Thus, a technically and economically feasible pathway will be provided for deploying medical-grade AI on resource-constrained Internet of Medical Things (IoMT) devices.
Adaptive learning systems are a significant area of research in personalized education, especially for students with dyslexia, as structured and responsive instructional support can greatly enhance learning outcomes. Many conventional rule-based methodologies are not readily adaptable for personalized instruction or real-time modification. To overcome this limitation, this paper introduces a lightweight adaptive learning system that utilizes a machine learning model to produce instructional recommendations. The system employs a Decision Tree Classifier trained on structured quiz-response features to assess a learner's proficiency and recommend the most appropriate learning stages. The platform is not meant to be a diagnostic tool; instead, it is meant to be a post-identification instructional support system. It has three main parts: an interactive quiz interface, a classification module that groups learners into Beginner, Intermediate, or Advanced levels, and a Flask-based backend that makes predictions and sends them out. The model is still computationally efficient, easy to understand, and good for real-time adaptive learning environments because it only uses quiz performance indicators. When tested on a controlled synthetic dataset, the results were very good, with an overall accuracy of 98.25%. Advanced learners had a class-specific [Formula: see text] score of 99.15%, beginner learners had a score of 95.87%, and intermediate learners had a score of 96.27%. The macro-average precision, recall, and [Formula: see text]-score were 96.93%, 97.32%, and 97.10%, respectively. The weighted averages were all close to 97.00%. These results show that the suggested method is a good and useful way to recommend quiz-based adaptive learning in technology-supported special education settings.
In humans and nonhuman primates, the anterior cingulate cortex (ACC) is an interface between "interoceptive" and "exteroceptive" domains. The ACC contains discrete subdivisions that are distinct in cytoarchitecture, connectivity, and function. The subgenual ACC (sgACC) is a key area for arousal state modulation. Importantly, the sgACC is dysregulated in major depression and a target for neuromodulation therapies, including deep brain stimulation. In contrast, the perigenual ACC (pgACC) is important for cognitive functions, including social decision-making. Understanding the major sources of afferent input to the sgACC and pgACC is essential for elucidating functional modulation, including in major depression. We took a mesoscopic, 'connectomic' approach to examine the balance of projections to ACC subdivisions from two sources of glutamatergic input: the prefrontal cortex (PFC) and insula, and the thalamus (n=6 macaques). Using retrograde tracer injections into ACC subdivisions in male macaques, and unbiased statistical clustering, we revealed that the ACC subdivisions are under the influence of strikingly different "heavily-weighted" (HW) inputs from PFC and thalamus. Only one cortical region, area 10m, has HW projections to both sgACC and pgACC, suggesting an integrative role. Additionally, the pgACC sends "top-down" inputs to the sgACC, without significant "bottom-up" return input. Finally, ACC and thalamus-ACC circuits are hierarchically organized, governed by cortical granularity and thalamo-cortical connectivity. Agranular cortices and their associated thalamic nuclei formed most inputs to the sgACC. In contrast, pgACC receives a balanced set of afferents from agranular, dysgranular, and granular cortices, coupled with inputs from broader thalamic regions associated with these cortices.Significance statement The ACC contains discrete subdivisions based on cytoarchitecture and connectivity, which serve unique functional roles. The sgACC and pgACC subdivisions receive many similar inputs based on neuroimaging work. Here, we leverage higher resolution retrograde tract-tracing in macaques to examine the relative weights and relationships of multiple cortical and thalamic afferents to each region. Using unbiased analyses of labeled cells, we conclude that the balance of afferent inputs shifts from a connectome dominated by agranular cortices and their thalamic partners in sgACC, to a more balanced afferent connectome in pgACC, represented by agranular, dysgranular, and granular cortices and their broader thalamic partners. These results facilitate interpretation of functional studies, and bridge understanding of the connectional basis of psychiatric disorders.
Chemosensory perception is crucial for evaluating our environment. Unlike physical sensory modalities, which deal with constant stimuli (such as light and sound), the olfactory environment is continuously changing. This dynamic nature requires the olfactory system to adapt to various situations, achieved through a vast array of olfactory receptors that facilitate combinatorial odor recognition, enabling the perception of nearly an infinite variety of odorants. Each odorant activates a specific pattern of olfactory receptors, creating a unique chord that is transmitted to the olfactory bulb. In contrast to other sensory systems, the olfactory pathway sends signals directly to the piriform cortex and amygdala, bypassing the thalamus. These regions project to the orbitofrontal cortex, interacting with the amygdala to influence emotions and play a key role in associative learning. This distinct architecture of the olfactory system contributes to its unique connections with emotions and sleep, which are discussed in this review. Хемосенсорное восприятие — одна из важнейших систем оценки окружающей среды. В отличие от физических сенсорных модальностей, для которых характер стимулов постоянен (свет, звук), постоянно меняющаяся обонятельная среда требует способности обонятельной системы адаптироваться к различным ситуациям. Это достигается за счет большого количества обонятельных рецепторов, необходимых для комбинаторного распознавания запахов, позволяющего воспринимать практически бесконечное количество одорантов. Каждый одорант возбуждает определенный паттерн обонятельных рецепторов, и сигнал проецируется на обонятельную луковицу, откуда в отличие от других сенсорных систем, минуя таламус, направляется непосредственно в грушевидную кору и миндалину. Грушевидная кора и миндалина формируют проекции в орбитофронтальную кору, которая вместе с миндалиной влияет на эмоции и принимает участие в ассоциативном обучении. Такое строение обонятельной системы обусловливает ее особую связь с эмоциями и сном.
Epilepsy remains a major global health concern, particularly in regions where continuous medical monitoring is difficult to implement. This study introduces a wearable system powered by edge-based artificial intelligence, designed to detect epileptic seizures in real time. The device integrates multiple sensors-accelerometers for motion tracking, photoplethysmography (PPG) for cardiovascular monitoring, and GPS for location detection-to enhance reliability through sensor fusion. Multiple machine learning models, including support vector machines (SVM), neural networks (NN), and random forest classifiers, were deployed and assessed directly on the device. Among these, the optimized random forest algorithm achieved the highest accuracy and fastest response time. The fusion of sensor data significantly improved specificity and maintained a very low false alarm rate. When a seizure is detected, the system instantly sends SMS alerts with precise location details to assigned caregivers, enabling prompt medical intervention. Simulation results demonstrate that this cost-effective and self-contained platform offers strong potential for improving patient safety and facilitating rapid response in low-resource settings where conventional monitoring tools are unavailable.
This paper examines how Italian teenagers construct meanings, norms and boundaries around sexting within their everyday digital cultures. Drawing on six vignette-based focus groups with 49 participants aged 16-18 years, the study used a media practice approach to analyse how teenagers collectively interpret a hypothetical sexting vignette. Findings show that participants overwhelmingly rely on heteronormative scripts, reproducing the dominant "girl-sends-nude-to-boy" narrative and positioning boys as initiators, while girls bear the potential risks. Sexting was consistently framed through risk discourses-particularly the threat of non-consensual dissemination-leading girls to internalise fear, self-policing, and responsibility for managing potential harm. Peer gossip further reinforced these risk narratives, circulating cautionary tales that blurred the boundaries between consensual and non-consensual practices and sustained moral regulation among peers. In response, teenagers constructed informal boundaries distinguishing "appropriate" from "unsafe" sexting, often limiting acceptable practices to long-term relationships and less explicit content. These findings highlight how risk discourses, peer surveillance and entrenched gender norms constrain teenagers' digital intimacies, particularly in an Italian context marked by limited sexuality education and persistent gender inequalities. The paper argues for research and pedagogical approaches that move beyond risk prevention to support teenagers' rights to agentic digital intimacies.
It is estimated that millions of species exist in the deep-sea environment, such as hydrothermal vents, cold seeps, abyssal plains and seamounts, which have yet to be described. Non-invasive biodiversity assessment methods can be applied using deep-sea environmental DNA (eDNA) metabarcoding, but it can sometimes be limited by the fact that deep-sea taxa are not well represented in curated reference databases. This study proposes a proof-of-concept artificial intelligence (AI)-driven framework for deep-sea eDNA analysis consisting of three different and complementary parts. To learn taxonomically informative sequence representations that cannot be obtained from conventional reference matching, a hybrid CNN Transformer model is presented and trained. The suggested methodology confirmed the practicality of confidence-guided novel taxon discovery with a macro F1 score of 0.847 using simulated data sets drawn from CMLRE expedition metadata. These findings provide the framework for future validation with actual deep-sea sequencing datasets. Second, a module for discovering potential, confidence-based novel taxa has been implemented, which sends low-confidence predictions to an unsupervised workflow that includes UMAP and HDBSCAN. In the simulated evaluation scenario, this module retrieved 91.5% of novel taxa as new taxa, and with 88.2% precision. Third, taxonomic abundance estimations that are suitable for downstream alpha and beta diversity analysis are provided by normalization and ecological profiling modules. The proposed framework consumed one million sequencing reads in 2.1 h across four NVIDIA A100 GPUs, significantly shorter than the processing times used in the different workflows evaluated in this study. Collectively, these findings indicate that deep learning-assisted eDNA analysis could help in biodiversity assessments where there is a lack of extensive reference databases. Due to all quantitative assessments conducted on simulated datasets, the results obtained should be regarded as proof-of-concept results and further testing with real data from deep-sea expeditions needs to be carried out to exercise the operational performance.
The Stieltjes moment problem is studied in a new framework within the general Gelfand-Shilov spaces defined via weight sequences. The novelty consists of allowing for a naturally larger target space for the moment mapping, which sends a function to its sequence of Stieltjes moments. The motivation comes from a recent version of the Borel-Ritt theorem, concerning the surjectivity of the Borel mapping in Carleman-Roumieu ultraholomorphic classes in sectors, whose defining weight sequence is subject to the condition, weaker than derivation closedness, of having shifted moments. The injectivity and surjectivity of the moment mapping in this new setting is studied and, in some cases, characterized. Finally, results are provided for general weight sequences of fast and regular enough growth when the condition of shifted moments fails to hold.
Valence detection in complex environment is critical for natural behaviors like foraging. Previous studies have explored valence processing in brain regions like lateral horn (LH) and mushroom body (MB) using simple synthetic stimuli in Drosophila. However, the neural basis for valence detection of natural objects in complex contexts remains unclear. Here, by brain-wide connectome analysis, we identified the evolutionarily conserved superior protocerebrum (SP) that integrates brain-wide multimodal inputs mainly via LH and MB, and sends widespread outputs particularly to the central complex (CX). This forms a convergence-divergence circuit resembling an autoencoder architecture, with SP as the bottleneck integrating multimodal information into low-dimensional valence signals. Specifically, SP input LH neurons integrate ethologically related innate valences for robust valence detection in natural environments, and the integration can be unimodal, such as that of diverse odors signaling food, or multimodal, such as that of wind and temperature signaling lousy weather. Opponent valences of attraction and aversion are further integrated into SP for complex valence detection. MB learned valences are also integrated into SP to update LH innate valences with recent experience for flexible valence detection. Attractive and aversive valences, either innate or learned, are integrated via excitatory and inhibitory synapses, respectively to form complex valence signals in a single SP neuron. Organized synaptic compartments support dendritic computation, with SP neurons exhibiting opposite synaptic organizations for opponent valences, indicating dendritic integration for complex valence detection. Our study highlights the importance of SP in multimodal opponent valence integration and suggests generalizable network and dendritic structures for complex valence processing.
To examine how remittance behavior and pandemic-related disruptions to remitting affect household food security among Dominicans in the United States, highlighting the nutritional implications of transnational economic obligations for immigrant families. Cross-sectional data from the 2021 CUNY Dominican Health Survey. A probabilistic survey of Dominican adults during the COVID-19 pandemic residing in seven U.S. states that collectively encompass most of the Dominican population in the U.S. (New York, New Jersey, Florida, Massachusetts, Pennsylvania, Rhode Island, and Connecticut). A total of 785 adults of Dominican origin, with household food insecurity (past month) and food pantry use (past 16 months) as outcomes, remittance frequency (none, other, low, moderate, high) as the key predictor, and COVID-19-related disruptions to remitting (yes/no) as a moderator. Overall, 28% of participants (n = 227) reported low household food security, and 42.1% (n = 352) reported pantry use. Survey-weighted logistic regression models showed that compared with non-remitters, weekly and moderate remitters had significantly higher odds of household low food security and pantry use. Stratified models further revealed that COVID-related disruptions to remittance behavior magnified the associations with remitting frequency, low food security, and pantry use. Transnational financial responsibilities may impose economic strain on sending households, increasing their vulnerability to food insecurity. Simultaneously, remittance behaviors reflect enduring social ties and resilience among Dominican families. These findings highlight the need for nutrition policies that account for transnational economic commitments in immigrant populations.
Transnational education (TNE) represents an evolutionary development in internationalisation, involving the movement of institutions or their programmes rather than student mobility. Influenced by the ambitions and strategies of sending and receiving countries, TNE offers distinct values and benefits to all stakeholders. Most previous research has focused on the successes and challenges of TNE programmes from the perspectives of programme directors, administrators, and educators, but less is known about graduates' experiences. This study explored graduates' perceptions of a UK-affiliated transnational medical programme in Egypt and its impact on their personal and professional development and careers. We conducted a qualitative study with graduates from a long-established transnational medical programme delivered under a UK university's licence in Egypt. We purposively sampled graduates to maximise variation in age, gender, nationality, professional status, current job, and country of work. We conducted semi-structured online interviews via Zoom, audio-recorded them, and transcribed the recordings. We adopted a constructivist/interpretivist approach and conducted reflexive thematic analysis, ensuring reflexivity was maintained throughout the study. Twenty-three graduates were interviewed for an average of 45 min; 14 were female and 9 were male, with ages ranging from 23 to 33 years. We identified six overarching and interconnected themes: (1) Personal and professional transformation through generic skills and critical thinking; (2) An integrated and well-organised curriculum delivered through case-based discussions; (3) Clinical preparedness fostered by small class sizes and early clinical exposure; (4) International career readiness supported by English language proficiency, exchange programmes, relevant examination formats, and multinationalism; (5) Supportive teaching and leadership staff from both institutions; (6) Immersive research experience. We explored insightful perspectives and experiences of the graduates. We found that an integrated, internationally focused curriculum, delivered and overseen by effective and supportive staff, prepares graduates to serve locally and internationally with notable resilience and confidence. These firsthand insights contribute to the limited literature on graduates' perspectives in TNE, particularly in health professions education (HPE), and offer practical, achievable recommendations for curriculum, assessment, student support, and the learning environment. Further research exploring staff and institutional viewpoints is necessary to gain a more comprehensive understanding.
Objectives Patient portal use has steadily increased across most populations. Prior, now dated, studies indicated lower adoption rates among Spanish- vs. English- speaking patients. This study compared patient portal activation and use patterns between Spanish- and English-speaking patients. Methods This retrospective cohort study was conducted at three North Texas health systems using the MyChart patient portal (Epic Systems Co.) and included patients ≥18 years with ≥1 completed clinician encounter between 4/5/2021 and 4/4/2022. The primary activation outcome was the baseline MyChart account activation rate. The secondary activation outcome was the MyChart account activation rate within the following year among patients without an account at baseline. The primary use outcome was the rate of patients logging in. Secondary use outcomes included rates of results review, notes review, and message initiation in the following year. We also evaluated the rates of proxy account use and mobile app use. We fit multivariable logistic regression models adjusting for health system, age, sex, comorbidity count, and the number of prior-year encounters. Results Spanish speakers represented 128,338 of 1,550,220 (8.3%) patients. Spanish speakers had lower odds of having an activated account at baseline (aOR 0.39 [0.39 - 0.40]) or activating one in the next year (aOR 0.68 [0.65 - 0.71]). Spanish speakers also had lower odds of logging in (aOR 0.62 [0.61 - 0.63]), reviewing results (aOR 0.79 [0.76 - 0.81]), reviewing notes (aOR 0.87 [0.84 - 0.89]), or sending messages (aOR 0.41 [0.40 - 0.42]). More Spanish than English speakers used the mobile app (59% vs 50%). There were inter-site differences in the rate of proxy account use. Conclusions Given lower levels of portal activation and use among Spanish-speaking patients, strategies are needed to identify and address barriers to activation and use. Qualitative studies could delineate these barriers and potential mitigating strategies.
Measurement of androstenedione (4-AD) and 17α-hydroxyprogesterone (17-OHP) in blood is crucial for diagnosing and monitoring endocrine disorders such as congenital adrenal hyperplasia (CAH), which results from enzymatic defects in the steroidogenesis pathway. CAH patients require lifelong hormone replacement therapy to manage adrenal insufficiency and control androgen excess. Microsampling offers a less invasive alternative to frequent venous blood sampling, particularly in paediatric CAH patients, and enables at-home monitoring. Including 11-ketotestosterone (11-KT), an emerging biomarker in CAH, further enhances the clinical relevance of this approach. Liquid chromatography-tandem mass spectrometry-based methods were established and fully validated for the determination of 4-AD, 17-OHP and 11-KT in plasma, blood and 20 μL dried blood microsamples collected by volumetric absorptive microsampling (VAMS). For VAMS samples, we paid particular attention to robustness regarding storage and hematocrit to support future home sampling applications. The method's applicability was evaluated by assessing the agreement with plasma as the reference matrix for steroid hormone determination. A robust extraction procedure was developed, ensuring consistent recoveries across storage conditions and hematocrits (0.20 - 0.60 L/L). The method was successfully validated for 4-AD, 17-OHP and 11-KT in VAMS, blood and plasma samples, achieving lower limits of quantification of 30, 144 and 81 pg/mL in VAMS samples, respectively. Accuracy and precision were within 6.6% and 14.0% for all matrices. VAMS samples remained stable for up to one week at room temperature and after postal sending. A proof-of-concept study showed that plasma concentrations can reliably be derived from VAMS concentrations, when taking into account the hematocrit.
Cellular organization is driven by recurrent, coordinated interactions between multiple cell types, each sending and receiving multiple signals. Existing computational methods for spatial profiling data consider only individual ligand-receptor interactions and fail to capture the higher-order interactions governing the tissue microenvironment. To address this gap, we developed ALARMIST (Assessment of Ligand And Receptor Motifs And Impacts in Spatial Transcriptomics), a probabilistic framework that infers interpretable multicellular communication patterns from spatial data. ALARMIST decomposes neighborhood-level signaling patterns into motifs: recurrent communication subnetworks involving multiple cell types and sets of enriched ligand-receptor interactions. For each cell, ALARMIST identifies its active motifs and estimates the downstream phenotypic effects of each motif on active cells. We applied alarmist to spatial datasets of lung adenocarcinoma (LUAD) and glioblastoma (GBM) to identify microenvironmental drivers of tumor progression. In paired LUAD and adenocarcinoma-in-situ (AIS) samples, ALARMIST identified an immune-active vascular motif at the tumor-normal boundary and implicated motif-active plasmacytoid dendritic cells as drivers of inflammation in early carcinogenesis. In matched low- and high-grade glioma samples, ALARMIST identified a hub-and-spoke motif centered on a malignant macrophage subpopulation, implicating a GRN-SORT1 signaling axis with a downstream impact gene set predictive of survival in low-grade glioma patients. Code for ALARMIST is available at https://github.com/tansey-lab/alarmist.
Background Diabetic retinopathy is one of the leading causes of preventable blindness worldwide, yet it can be stopped through early detection. AI models are increasingly being used as a key enabler to automate this screening, and the results in research settings look very promising. The real challenge is not building more sophisticated models but deploying one that works safely in real clinics. Clinical safety standards require the system to catch nearly every true case of disease, even if that means sending many healthy patients for unnecessary follow-up. This trade-off between keeping patients safe and avoiding a flood of false alarms is the core problem this paper addresses. Given that clinics typically fix a minimum sensitivity target in advance, this study compares decision-making strategies held to the same sensitivity requirement to determine which produces the fewest unnecessary referrals.  Methods We evaluate five decision strategies under identical conditions on the public EyePACS dataset of 5,270 retinal fundus images, with 1,366 labeled as having diabetic retinopathy. The strategies range from a single AI model making every referral decision independently, to ensemble methods that combine the probability scores of multiple models into one unified output, to a two-tiered method in which all images are first screened by a primary model, and a group of secondary models can overturn a referral when their disagreement is high enough. Each strategy is evaluated under two clinically grounded sensitivity targets, a strict 95% requirement and a more moderate 90% requirement, so the results reflect realistic deployment conditions rather than unconstrained optimal performance. Results When the system is required to achieve 95% sensitivity, all strategies produce high false-positive rates, with the best single model reaching only 17.5% specificity. Ensembles offer only marginal improvement at this threshold, while majority voting consistently performs worst across both sensitivity levels. Reducing the sensitivity target from 95% to 90% alone decreases false positives by about 17%. When this lower threshold is combined with an ensemble strategy, unnecessary referrals drop by nearly 25% compared with a single model at 95%. Neither adjustment alone produces this level of improvement; the benefit appears only when both are applied together. Conclusions This study shows that two decisions matter most in AI-based diabetic retinopathy screening: the sensitivity target a clinic sets and the decision strategy it pairs with that target. Although the sensitivity target had a greater influence on referral burden, the best outcomes occurred only when both the target and the strategy were carefully chosen. Pairing a 90% sensitivity target with a weighted ensemble reduced unnecessary referrals by nearly 25% compared with a single model at the stricter 95% target, while majority voting produced the highest false-positive burden at both thresholds. These findings suggest that clinically grounded threshold selection is just as important as the decision strategy itself and that seemingly intuitive approaches such as majority voting may underperform when evaluated under the same safety constraints.
Thousands of scholars across Europe faced discrimination for their heritage, especially Jewish heritage, throughout the 1930s. Hundreds of these persecuted academics worked in science and healthcare, often as primary researchers within a medical sub-specialty. In response, in 1933 British professors founded the Society for the Protection of Science and Learning (SPSL) with two objectives: raising funds to provide maintenance grants for displaced scholars and connecting scholars to academic placements worldwide. In this article, I analyze the stories of nineteen physicians fleeing fascist persecution in Europe through their 712 leaves of correspondence within the archive of the SPSL. I selected two medical subspecialties, to compare whether the well-funded field of oncology would have greater success placing refugees compared to the less academically-elevated field of orthopedic surgery. Through the first analysis of two complete medical subspecialty records from the SPSL archives, I examine how the SPSL faced difficulties deciding which colleagues to help financially and triaged levels of support. I also situate the experiences of the analyzed applicants within the context of British interwar medicine and academia, including British skepticism of foreign research and overlaps between SPSL leadership and the British eugenics movement. When funding ran low, the SPSL still assisted scholars through different channels, whether referring to other aid organizations, sending support letters to potential employers, or simply staying in contact. The surge in applications in 1938, five years after the organization's foundation, hindered scholars' chances as persecution increased and funding tapered. Importantly, the British medical establishment distrusted the overall research quality of many fleeing scholars and felt even less confident admitting medical practitioners who were not also researchers. Ultimately, medical scholars from fascist Germany and Austria who applied to the SPSL had a similar experience to other aspiring refugees seeking British assistance before 1939: they could not rely on Britain to support their immigration and predominantly settled elsewhere.