Browse Articles

E-RespiNet: An LLM-ELECTRA driven triple-stream CNN with feature fusion for asthma classification

Respiratory disease diagnosis remains challenging in resource-constrained settings, where limited specialist expertise contributes to diagnostic uncertainties affecting over 300 million people worldwide. This study presents E-RespiNet, a novel multi-modal deep learning architecture that integrates ELECTRA’s discriminative pre-training with a triple-stream convolutional neural network framework, enhanced through Harmony Search with Opposition-Based Learning optimization for automated respiratory sound classification. The architecture simultaneously processes mel-frequency cepstral coefficients, discrete wavelet transforms, and mel-spectrograms through parallel CNN streams, with features integrated through hierarchical fusion and ELECTRA-based contextual enhancement. Comprehensive evaluations on two independent clinical datasets—the Asthma Detection Dataset Version 2 (1,211 recordings across five conditions) and King Abdullah University Hospital dataset (940 samples from 81 subjects across four conditions)—demonstrated exceptional performance with 98.9% and 98.8% accuracy respectively, representing 5.0% and 4.3% improvements over baseline configurations. Cross-institutional validation revealed 75.7% average accuracy with a 23.3% generalization gap, substantially better than typical medical AI cross-domain performance. Gradient-weighted class activation mapping provided clinically relevant interpretability, while the Harmony Search optimization framework enhanced both performance and cross-institutional robustness. These results demonstrate significant advances in automated respiratory sound analysis through discriminative language model integration and metaheuristic optimization, establishing important benchmarks for deployable respiratory diagnostic tools in diverse healthcare settings.

Can neuromorphic computing help reduce AI’s high energy cost?

Assessment of cyclists yielding to pedestrians at an unsignalized zebra crossing in Germany using drone video

Abstract Previous research has examined vehicle-pedestrian and vehicle-cyclist interactions, but there have been few studies that examined cyclist-pedestrian interactions at intersections. This study addresses this gap by analyzing cyclist-pedestrian interactions at an unsignalized intersection in Germany using publicly available drone data. The study presents a framework and proof of concept for analyzing cyclist behavior proactively, without relying on crash data. The primary objectives are to identify the variables influencing cyclist yielding behavior and obstructed travel time (OTT) within a predefined zone at a zebra crossing and to classify cyclist behaviors. Using logistic and linear regression models, several key predictors were identified, including cyclist speed, trajectory changes, pedestrian time-to-conflict-point, and interaction proximity, which significantly impacted yielding behavior. Speed reduction and pedestrian presence on the zebra crossing were found to improve yielding rates. Additionally, clustering analysis revealed two optimal and distinct cyclist behavior groups: one cluster exhibiting less yielding behavior, while the other demonstrated greater compliance with traffic laws. This proactive approach provides a valuable alternative in environments where crash data acquisition is complicated by privacy regulations. It offers critical insights for traffic management strategies aimed at enhancing pedestrian safety at unsignalized intersections, making it applicable to broader contexts with similar data challenges.

Identification and validation of stage-specific microRNAs and target genes for prostate cancer: Utilizing bioinformatics tools for diagnostic marker discovery

Given the urgent need for more specific, sensitive, and non-invasive markers for prostate cancer screening and differential diagnosis, circulating miRNAs have emerged as valuable candidates. Sixty seven prostate cancer subjects in different stages were included in this study. The participants were categorized into groups based on their pathological characteristics as local, biochemical relapse and metastatic. We retrieved eligible datasets from GEO database to identify stage-specific differentially expressed up/down-regulated genes. Cytohubba, built-in application of Cytoscape software, and Reactome pathway database were applied to select hub genes. To select upstream miRNAs, we utilized the MiRWalk and miRNet online tools. To construct the miRNA-mRNA regulatory networks, we employed rna22. Finally, three miRNAs and five target genes were validated in peripheral blood mononuclear cells of PCa patients compared with benign prostate hyperplasia. PSA level was also measured using ELISA. Our findings revealed the potential role of PRC1 and UBA52 to be used as biomarkers for the metastatic stage, RCC1 for both biochemical relapse, and metastatic subjects. Furthermore, elevated levels of miR-124-3p and downregulation of miR-133a-3p can be introduced as biochemical relapse stage identifier. We also identified the tumor suppressor role of miR-17-5p, which was associated with higher Gleason scores. We propose PRC1, UBA52, RCC1, miR-124-3p and miR133a-3p as stage-specific PCa identifiers.

Correction for Tapinova et al., Integrated Ising model with global inhibition for decision-making

Spatial prediction of Spodoptera frugiperda expansion in India using MaxEnt under shifting climate regimes

Photophysical image analysis for sCMOS cameras: Noise modelling and estimation of background parameters in fluorescence-microscopy images

Fluorescence microscopy is an effective tool for imaging biological samples, yet captured images often contain noises, including photon shot noise and camera read noise. To analyze biological samples accurately, separating background pixels from signal pixels is crucial. This would ideally be guided by the knowledge of a parameter called the Poisson parameter, λ bg , representing the mean number of photons collected in a background pixel (for the case when quantum efficiency = 1 and the dark current is negligible). This study introduces a method for estimating λ bg , from an image which contains both background and signal pixels, using probabilistic noise modeling for an sCMOS camera. The approach incorporates Poisson-distributed photon shot noise and sCMOS camera read noise modelled with a Tukey-Lambda distribution. We apply a chi-square test and a truncated fit technique to estimate λ bg directly from a general sCMOS image, with camera parameters determined through calibration experiments. We validate our method by comparing λ bg estimates in images captured by sCMOS and EMCCD cameras for the same field of view. Our analysis shows strong agreement for low to moderate exposure images, where estimated values for λ bg align well between the sCMOS and EMCCD images. Based on our estimated λ bg , we perform image thresholding and segmentation using our previously introduced procedure. Our publicly available software provides a platform for photophysical image analysis for sCMOS camera systems.

Correction for Jarvis et al., The impact of air pollution on petcare utilization

Kinetics study of sonotransesterification of low grade crude palm oil (CPO) using heterogeneous Na2O/activated natural mordenite catalyst

Assessing the socio-cognitive determinants of personal protective equipment uses among domestic waste collectors in the Ho municipality, Ghana: A cross-sectional study

Domestic waste collectors (DWCs) are exposed to occupational safety and health related morbidities and mortalities globally due to the non-use, improper use, and non-availability of personal protective equipment (PPE) in their jobs which endangers DWCs’ lives, safety, and well-being. The present study investigated the extent to which socio-cognitive determinants predicted PPE use among DWCs in the Ho municipality in the Volta Region in Ghana. A quantitative cross-sectional survey was conducted among DWCs (n = 344) in the Ho Municipality of Ghana to assess the socio-cognitive determinants of PPE use. The questionnaire consisted of 107 items that were informed by a literature review in previous qualitative research, and two theoretical frameworks explaining behavior (i.e., the Health Belief Model (HBM) and Reasoned Action Approach (RAA) and measured constructs such as perceived severity and susceptibility of work-related health risks, perceived benefits, and barriers, perceived norm, and self-efficacy towards PPE use. Partial least squares structural equation modeling (PLS-SEM) was used to evaluate the structural model describing the relationship between the socio-cognitive determinants and intention to use PPE, which was the main outcome measure. The integrated model explained 67% of the variance in PPE-use intention. Intention to use PPE was significantly positively and directly influenced by attitude (β = 0.174, p < 0.001), indicated cues to action (β = 0.500, p < 0.001), perceived rule enforcement by the management (β = 0.114, p < 0.05), and self-efficacy (β = 0.199, p < 0.01). The direct effect of subjective norms on intention to use PPE was not significant (β = 0.040, p = 0.396). Attitude in turn was significantly predicted by perceived severity (β = 0.244, p < 0.001), perceived benefits (β = 0.209, p < 0.01), and behavioral beliefs (β = 0.342, p < 0.001), whereas perceived barriers were significantly associated with self-efficacy (β = 0.377, p < 0.001). In conclusion, the current study successfully expanded the utility of HBM and RAA in assessing the socio-cognitive determinants of PPE use among DWCs in a developing economy. Thus, the findings highlight the combined influence of individual beliefs and organizational enforcement on DWCs’ motivation to use PPE. Interventions should pair hazard‑communication and self‑efficacy training with strict managerial enforcement to strengthen PPE compliance.

Slower searching yields higher efficiency: A case study of taxi drivers

The movement patterns of animals while searching for food have been studied extensively in recent decades. However, although human search behavior has existed since the beginning of civilization, not much is known about human search patterns, particularly regarding strategies that yield higher efficiency. Although most humans no longer need to gather and hunt in the wild, human searching remains prevalent in modern times. A common example of human searching is performed by taxi drivers looking for passengers in a city. Here, we analyze GPS data of taxi drivers in three major cities over different time periods and find that when drivers search for passengers, the higher is their efficiency, the slower is their searching, and they tend to make more short-distance turns during the search. Our study further indicates that individuals are characterized by a specific level of efficiency, and thus, efficient drivers are consistently efficient across different days as they follow their own search strategies. Interestingly, only about 10% of drivers adopt the most efficient strategy, earning nearly 20% more than the average driver. Our findings shed light on human search behavior, a fundamental aspect of human decision-making in competitive and fast-paced environments.

5-HT1A receptor antagonism decreases motor activity and influences dopamine and serotonin metabolization pathways, primarily in cingulate cortex and striatum

Abstract We assessed the effect of the 5-HT 1A receptor (R) antagonist WAY100,635 on motor behaviors, object place learning and the regional levels of dopamine (DA), serotonin (5-HT) and their metabolites in the rat brain. After a single dose of either WAY100,635 (0.4 mg/kg) or vehicle (0.9% NaCl), recognition memory was assessed together with motor/exploratory behaviors. After sacrifice, regional DA, 5-HT and metabolite levels were determined with HPLC. Overall activity and exploratory behavior were reduced by WAY100,635. Object place recognition did not differ between treatments. WAY100,635 promoted DA metabolization (1) by both monoamine oxidase (MAO) and catechol-O-methyl transferase (COMT) in cingulate, caudateputamen, thalamus and cerebellum, (2) solely by MAO in dorsal hippocampus and (3) solely by COMT in ventral hippocampus and brainstem, but suppressed DA metabolization (by both MAO and COMT) in nucleus accumbens. It promoted 5-HT metabolization (by MAO) in cingulate, caudateputamen, dorsal hippocampus and brainstem, but suppressed it in nucleus accumbens, thalamus and cerebellum. WAY100,635 altered activity and exploratory behavior as well as the quantitative relations between the neurotransmitter/metabolite levels in the individual brain regions, by inducing region-specific shifts in the metabolization pathways.

Correction: Within-subjects ultra-short sleep-wake protocol for characterising circadian variations in retinal function

Genomic and insulin-mediated control of metabolic homeostasis by the mosquito ecdysone-induced gene E93

Ecdysone-induced protein 93 (E93) is an adult specifier that governs insect pupal-adult conversion. It affects the reproductive transition in adult Aedes aegypti mosquitoes, the significant vectors of numerous devastating human diseases. Here, we show that E93 is essential for maintaining metabolic homeostasis during the reproductive cycle of mosquitoes. E93 deficiency led to insufficient production of insulin-like peptide 3 (ILP3) from insulin-producing cells in the brain, resulting in reduced phosphorylation of protein kinase B (Akt), a key regulator in the insulin signaling pathway. This reduction facilitated the nuclear translocation of FoxO and enhanced the activity of glycogen synthase kinase 3β (GSK3β), which in turn respectively activated the transcription of genes encoding phosphoenolpyruvate carboxykinase (PEPCK) during gluconeogenesis and reduced glycogen synthesis. Further insulin rescue and the luciferase activity assays demonstrated that E93 directly inhibited PEPCK transcription. Ultimately, E93 -depleted mosquitoes exhibited systemic metabolic reprogramming, characterized by the dysregulation of carbohydrate, lipid, and amino acid metabolism. Our findings establish that E93 orchestrates metabolic homeostasis by coordinating the insulin signaling cascade and directly regulating PEPCK expression, thus providing an intrinsic connection between endocrine signaling and E93-mediated reproduction in mosquitoes.

Construction and validation of a LASSO penalized logistic regression model predicting hypernatremia after pituitary adenoma surgery

“Has this been tested? Who has it helped? Who has it hurt?”: Public perceptions about California’s extreme risk protection order law

Extreme risk protection order (ERPO) laws in the United States temporarily suspend access to firearms by individuals judged to be at significant risk of harm to self or others. Evidence points toward preventive effects, but uptake of these laws remains lower than that which is likely needed to optimize their intended benefits on rates of firearm-related injury. To inform implementation efforts and policy refinements, we examined public awareness of and support for California’s ERPO law, barriers to its use, and possible alternative approaches. Using a general population sample of adults from the California Safety and Wellbeing Survey ( N  = 3531), with both closed-ended and open-ended questions, we provide updated prevalence estimates and narrative insights about the policy, with subgroup analyses by firearm ownership status and categories of race and ethnicity. Most respondents remained unaware of the law; however, after reading a brief policy description, majorities of the population, including firearm owners and respondents across all racial-ethnic subgroups, endorsed ERPOs as appropriate and said they would be willing to serve as petitioners for family members of concern. Barriers to use included knowledge gaps, particularly among non-firearm owners and Black, Latine, and Asian respondents; not trusting the system to be fair; and the perception that the scenarios of concern involved personal or family matters. More than one-third of respondents said the police were not a preferred means through which to initiate an order and a similar proportion favored holding onto a family member’s guns themselves instead of using an ERPO. Optimizing the lifesaving impacts of ERPOs will require addressing these layered concerns through coordinated and sustained investments in broader ecosystems of community safety while still facilitating firearm recovery where appropriate.

p53 regulates the expression of histone modifiers to restrict stemness and maintain differentiated luminal identity in breast cancer

Breast cancer is the leading cause of death in women under 50. The majority of breast cancers are estrogen receptor α-positive (ER+) and are commonly treated with hormonal therapies such as tamoxifen that inhibit ER activity. The TP53 tumor suppressor gene, encoding the p53 protein, is the most frequently mutated gene in breast cancer, and TP53 mutations are associated with diminished tamoxifen response and worse prognosis for breast cancer patients. Here, we report that in breast cancer cells p53 and ER cooperate to regulate the transcription of a set of genes encoding chromatin modifiers. The net result is a global increase in H3K4me3 and decrease in H3K9me3 chromatin marks. The resultant “open” chromatin is associated with increased transcription of luminal cell identity genes and enhanced tamoxifen sensitivity. Conversely, diminished p53 control of these chromatin modulators is associated with the evolution of tamoxifen resistance and cancer stem cell properties.

Palovarotene for patients with multiple hereditary exostosis: results of MO-Ped, a terminated, randomized, placebo-controlled, double-blind phase 2 trial

Abstract A phase II trial (MO-Ped; NCT03442985) assessed palovarotene in pediatric patients with multiple hereditary exostosis (MHE). Patients aged ≤ 14 years with MHE were randomized 1:1:1 to placebo, palovarotene 2.5 mg, or palovarotene 5.0 mg daily. Due to concerns of premature physeal closure (PPC), a partial clinical hold was instituted, followed by trial termination. The primary efficacy endpoint was the annualized rate of new osteochondromas (OCs). Safety was assessed. Overall, 193 patients received ≥ 1 dose of palovarotene or placebo. Due to trial termination, no patients completed planned treatment. Prior to 06 December 2019 (patients notified of clinical hold and treatment stopped), 30 patients completed Month 12 efficacy imaging. No significant differences in the annualized rate of new OCs, or change from baseline in volume of OCs or OC cartilage, were observed between treatment groups. The adverse event profile was consistent with systemic retinoids. There was no evidence of an effect of palovarotene on linear growth and no cases of PPC. Overall, palovarotene showed no clear efficacy signal in MHE, resulting in a non-favorable benefit-risk profile. Interpretation of results was limited by the reduced treatment duration and smaller than expected cohort. The trial yielded important data on the natural history of MHE. Trial registration : NCT03442985 (first posted 22 February 2018)

Predictive factors for the efficacy of brolucizumab in refractory polypoidal choroidal vasculopathy following aflibercept resistance

Purpose To identify predictors of extension of the injection interval beyond 8 weeks at the 24-month visit after switching to brolucizumab in aflibercept-resistant polypoidal choroidal vasculopathy (PCV). Design Retrospective observational study Methods 17 eyes of 16 patients with persistent or recurrent exudation on aflibercept were switched to intravitreal brolucizumab and managed with a treat-and-extend (T&E) regimen with a minimum 8-week interval after loading. The primary outcome contrasted extension (>8 weeks) versus non-extension (≤8 weeks) at month 24. Prespecified predictors were early central choroidal thickness (CCT) change from baseline to the switch visit (A0 to A1; ≥ 40% reduction) and pachychoroid. Associations were tested with Fisher’s exact tests and Firth-penalized logistic regression with the event defined as extension. Results At 24 months, 6 of 17 eyes (35%) achieved extension. A ≥ 40% early CCT reduction occurred in 0 of 6 extension eyes versus 7 of 11 non-extension eyes (Fisher exact two-sided P ≈ 0.035). In the Firth model (event = extension), < 40% CCT reduction strongly predicted extension (odds ratio 38.5; profile-likelihood 95% CI 2.0–10,000; LR P = 0.004). Non-pachychoroid showed the same direction with wide CIs (odds ratio 14.3; 95% CI 0.99–2,174; LR P = 0.006). Model fit was significant (LR χ² = 15.19, P = 0.0005) and discrimination was good (apparent AUC ≈ 0.97). We prespecified a parsimonious two-predictor model to limit overfitting; adding age, sex, prior photodynamic therapy, or number of prior aflibercept injections did not materially change coefficients or improve AICc (ΔAICc < 2). Conclusions Eyes without marked early choroidal thinning (<40% CCT reduction at A1) were more likely to extend, whereas marked thinning (≥40%) signaled difficulty extending under T&E regimen after switching to brolucizumab. Given the small sample and few events, estimates should be interpreted cautiously and considered hypothesis-generating, and warrant prospective external validation studies.

Correction for Zhu et al., Factors underlying a latitudinal gradient in the S/G lignin monomer ratio in natural poplar variants