Sci Rep. 2025 Aug 12;15(1):29528. doi: 10.1038/s41598-025-14047-3.

ABSTRACT

Biometric identification systems offer strong security by relying on unique personal traits. At the same time, they raise significant privacy concerns because compromised biometric data cannot be revoked. This paper explores the use of homomorphic encryption (HE) as a means to protect biometric data during classification and reduce the risk of exposing sensitive information. Our system comprises a feature extractor which operates locally and a classifier which processes encrypted data. We demonstrate the feasibility of our approach on a gait recognition task, employing a vision transformer as a feature extractor and training several HE-compatible classifiers. Through a comprehensive statistical analysis, we evaluate the impact of HE on accuracy and computational complexity, especially with different activation functions and their polynomial approximations. Our results demonstrate the feasibility of secure and accurate gait recognition using HE, while highlighting the trade-off between security and performance.

PMID:40796590 | DOI:10.1038/s41598-025-14047-3

Can J Ophthalmol. 2025 Aug 9:S0008-4182(25)00357-6. doi: 10.1016/j.jcjo.2025.07.009. Online ahead of print.

ABSTRACT

OBJECTIVE: To compare the efficacy of aflibercept, bevacizumab, and ranibizumab for resolution of diabetic macular edema-associated hard exudates (HEs).

DESIGN: Post hoc analysis of the Diabetic Retinopathy Clinical Research Network Protocol T trial.

PARTICIPANTS: Two hundred and forty-eight subjects with 84 eyes were treated with aflibercept, 71 eyes were treated with bevacizumab, and 93 eyes were treated with ranibizumab.

METHODS: The volume of HEs was measured by automatically quantifying hyperreflective foci on optical coherence tomography volume scans using a deep-learning algorithm. HEs were quantified within the total macula (6 × 6 mm scan area), as well as separately within the central subfield (CSF), inner ring (IR), and outer ring (OR) of the Early Treatment Diabetic Retinopathy Study grid at baseline, 4, 12, 24, and 52 weeks (w) after treatment. The extent of HEs at baseline and the change over time were compared among the groups.

RESULTS: Baseline HEs in the total macula were 0.0211 ± 0.0275, 0.0215 ± 0.0266, and 0.0223 ± 0.0272 mm3 for the aflibercept, bevacizumab, and ranibizumab groups, respectively. At 1 year, HEs significantly decreased with aflibercept and ranibizumab in the total macula and in all subregions (CSF, OR, and IR), but they only decreased with evacizumab in the CSF. Over 1 year, the reduction in HEs was greater for aflibercept and ranibizumab than for bevacizumab within the total macular region, and anibizumab was superior to bevacizumab for the IR as well.

CONCLUSIONS: HEs decreased significantly with aflibercept and ranibizumab treatment over 1 year in all regions, but not with bevacizumab, highlighting the differential efficacy among agents in resolving HEs.

PMID:40796006 | DOI:10.1016/j.jcjo.2025.07.009

Bioinformatics. 2025 Aug 9:btaf446. doi: 10.1093/bioinformatics/btaf446. Online ahead of print.

ABSTRACT

MOTIVATION: A pivotal area of research in antibody engineering is to find effective modifications that enhance antibody-antigen binding affinity. Traditional wet-lab experiments assess mutants in a costly and time-consuming manner. Emerging deep learning solutions offer an alternative by modeling antibody structures to predict binding affinity changes. However, they heavily depend on high-quality complex structures, which are frequently unavailable in practice. Therefore, we propose ProtAttBA, a deep learning model that predicts binding affinity changes based solely on the sequence information of antibody-antigen complexes.

RESULTS: ProtAttBA employs a pre-training phase to learn protein sequence patterns, following a supervised training phase using labeled antibody-antigen complex data to train a cross-attention-based regressor for predicting binding affinity changes. We evaluated ProtAttBA on three open benchmarks under different conditions. Compared to both sequence- and structure-based prediction methods, our approach achieves competitive performance, demonstrating notable robustness, especially with uncertain complex structures. Notably, our method possesses interpretability from the attention mechanism. We show that the learned attention scores can identify critical residues with impacts on binding affinity. This work introduces a rapid and cost-effective computational tool for antibody engineering, with the potential to accelerate the development of novel therapeutic antibodies.

AVAILABILITY AND IMPLEMENTATION: Source codes and data are available at https://github.com/code4luck/ProtAttBA.

SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

PMID:40795828 | DOI:10.1093/bioinformatics/btaf446

J Clin Endocrinol Metab. 2025 Aug 12:dgaf462. doi: 10.1210/clinem/dgaf462. Online ahead of print.

ABSTRACT

The pathophysiological mechanisms underlying type 2 diabetes mellitus (T2DM) remain incompletely understood, and the disease continues to impose a substantial burden on global health. In this study, we integrated the data from the largest genome-wide association study (GWAS; N = 898,130) of T2DM with human plasma protein quantitative trait locus (pQTL; N = 53,022) data to conduct the first proteome-wide association study (PWAS) of T2DM. Following Mendelian randomization and colocalization analyses, we identified nine independent putatively causal proteins. Among these, three were successfully replicated in other independent pQTL datasets, including two (HYOU1 and FLT3) that were novel and not identified in the original GWAS. Further integration with expression quantitative trait locus (eQTL) data from three diabetes- related tissues (blood, adipose tissue, and pancreas) revealed that five of the causal proteins also showed significant associations with T2DM at their cis-regulatory mRNA levels. Subsequent functional annotation supported potential pathogenic roles of the causal proteins. Notably, drug repurposing analysis identified 29 candidate drugs for T2DM treatment by targeting four causal proteins. In conclusion, our findings provide new insights into the pathogenesis of T2DM and highlight promising targets for future mechanistic and therapeutic investigations.

PMID:40796333 | DOI:10.1210/clinem/dgaf462

Biochem Pharmacol. 2025 Aug 10:117219. doi: 10.1016/j.bcp.2025.117219. Online ahead of print.

ABSTRACT

Several neurodegenerative diseases including amyotrophic lateral sclerosis (ALS) are characterized by toxic aggregates accumulation due to autophagy blockade, prompting researchers to identify new autophagy-activating drugs. Here we tested, in an in vitro ALS/PDC model, the neuroprotective effects of the antipsychotic Chlorpromazine (CPZ) and the antidepressant Clomipramine (CMI), chosen by drug repurposing approach for their ability to stimulate TPC2 lysosomal channel. Patch-clamp electrophysiology on enlarged lysosomes in NSC-34 motor neurons showed that CPZ and CMI induced large inwardly-rectifying currents, that were inhibited by TPC2 synthetic blocker trans-Ned-19. The same currents were evoked by TPC2 endogenous agonist NAADP and its mimetic agent TPC2-A1-N, and inhibited by trans-Ned-19 and siRNAs against TPC2 (siTPC2). CPZ and CMI elicited a significant [Ca2+]i increase that rapidly induced nuclear translocation of TFEB (transcription factor EB), the master regulator of autophagy. Accordingly, TPC2 stimulation by both the drugs boosted autophagy, as revealed by the activation of autophagy initiators ULK and AMPK α and modification of LC3-II/p62(SQSTM1) ratio. Furthermore, by normalizing autophagy markers, CPZ and CMI counteracted the detrimental effects induced by L-BMAA, a neurotoxin mimicking ALS/PDC. Notably, siTPC2 partially reverted CMI- and CPZ-induced neuroprotection as well as that produced by NAADP. At mitochondrial level, these drugs prevented ATP reduction and ROS overproduction in motor neurons exposed to L-BMAA for 24 h. For a longer L-BMAA exposure, CPZ and CMI counteracted LDH, cytochrome C and SMAC/DIABLO release, thus preventing cell demise. These findings suggest that TPC2 activation by drug repurposing could provide novel therapeutic options for ALS via autophagy regulation.

PMID:40796055 | DOI:10.1016/j.bcp.2025.117219

Nat Struct Mol Biol. 2025 Aug 12. doi: 10.1038/s41594-025-01614-5. Online ahead of print.

ABSTRACT

Pregnane X receptor (PXR) is a nuclear receptor considered to be a master transcription factor of xenobiotic metabolism. Here, using enhanced ultraviolet crosslinking and immunoprecipitation, we show that PXR can bind mRNAs in different cancer cell lines and normal liver tissues. PXR-bound mRNAs include genes related to metabolic reprogramming and lipid metabolism. Separately from its known nuclear transcriptional function, cytoplasmic PXR binds and stabilizes mature mRNA containing C+G-enriched sequences through its zinc-finger domain. Mechanistically, cytoplasmic PXR interacts with RNH1, an RNase inhibitor, to regulate RNA stability. In colorectal cancer cells, cytoplasmic PXR facilitates glucose uptake by stabilizing SLC2A1 mRNA. This process further promotes cell proliferation and cancer development. Our study unveils a previously unknown dimension of PXR-mediated gene regulation by characterizing PXR as an RNA-binding protein important for mRNA stability in the cytoplasm.

PMID:40797049 | DOI:10.1038/s41594-025-01614-5

Nat Commun. 2025 Aug 12;16(1):7458. doi: 10.1038/s41467-025-62598-w.

ABSTRACT

DNA transactions introduce torsional constraints that pose an inherent risk to genome integrity. While topoisomerase 1 (TOP1) activity is essential for DNA supercoil removal, the aberrant stabilization of TOP1:DNA cleavage complexes (TOP1ccs) can result in cytotoxic DNA lesions. What protects genomic hot spots of topological stress from excessive TOP1cc accumulation remains unknown. Here, we identify chromatin context as an essential means to coordinate TOP1cc resolution. Through its ability to bind poly(ADP-ribose) (PAR), the histone variant macroH2A1.1 facilitates TOP1cc repair factor recruitment and lesion turnover, thereby preventing DNA damage in response to transcription-associated topological stress. The alternatively spliced macroH2A1.2 isoform is unable to bind PAR or protect from TOP1ccs. Impaired macroH2A1.1 splicing, a frequent cancer feature, was predictive of increased sensitivity to TOP1 poisons in a pharmaco-genomic screen in breast cancer cells, and macroH2A1.1 inactivation mirrored this effect. We propose macroH2A1 alternative splicing as an epigenetic modulator of TOP1-associated genome maintenance and a potential cancer vulnerability.

PMID:40796804 | DOI:10.1038/s41467-025-62598-w

Clin Drug Investig. 2025 Aug 12. doi: 10.1007/s40261-025-01466-3. Online ahead of print.

ABSTRACT

BACKGROUND AND OBJECTIVES: Reports of suspected adverse drug reactions are of a great importance for the safety monitoring of new vaccines to identify potential safety risks promptly and to ensure necessary measures for risk mitigation. We reviewed the reports of fatal adverse drug reactions after coronavirus disease 2019 (COVID-19) vaccination with Comirnaty®, Spikevax®, and Vaxzevria® during the national vaccination campaign in Sweden.

METHODS: Swedish reports of suspected adverse drug reactions with fatal outcomes after COVID-19 vaccines were retrieved from the EudraVigilance database. Vaccination data were obtained from the National vaccination register. Reporting rates were calculated by dividing the number of adverse drug reaction reports with fatal outcomes by the number of people exposed to at least one dose of the COVID-19 vaccines or by the number of vaccine doses given. A causality assessment of adverse drug reaction reports was performed by clinically qualified reviewers.

RESULTS: More than 26 million doses of COVID-19 vaccines were administered and 456 reports of suspected adverse drug reactions with fatal outcomes were reported during 27 December, 2020-31 May, 2023. The reporting rate was 5.7 fatal outcomes per 100,000 persons vaccinated with at least one dose of any COVID-19 vaccine or 1.7 per 100,000 vaccine doses given. Most of the fatalities were related to patients' pre-existing conditions, predominantly among people aged 70 years or older. Only ten of the reported fatalities (0.1 per 100,000 persons vaccinated) were assessed as consistent with a causal association to COVID-19 vaccination.

CONCLUSIONS: Adverse drug reactions with fatal outcomes after COVID-19 vaccines in Sweden were very rare. No new safety concerns were observed in this study.

PMID:40796716 | DOI:10.1007/s40261-025-01466-3

J Microbiol Immunol Infect. 2025 Aug 8:S1684-1182(25)00151-3. doi: 10.1016/j.jmii.2025.08.003. Online ahead of print.

ABSTRACT

BACKGROUND: Mycobacterium abscessus (MABS) is a clinically significant nontuberculous mycobacterium, and its drug resistance poses substantial therapeutic challenges. Comprehensive genomic and phenotypic analyses are essential for elucidating the mechanisms underlying this resistance and enhancing understanding of its epidemiology.

METHODS: Whole-genome sequencing (WGS) using the Illumina platform was conducted on 61 clinical MABS isolates obtained from patients in Thailand. MABS subspecies classification was performed using FastANI, TYGS, and NTM-Profiler. Phenotypic drug susceptibility testing (pDST) was determined using a broth microdilution method. Resistance mutations were identified through NTM-Profiler and Snippy pipelines.

RESULTS: The analysis classified MABS isolates into three subspecies: subsp. abscessus (40/61, 65.57 %), subsp. massiliense (15/61, 24.59 %), and subsp. bolletii (6/61, 9.83 %). Phylogenetic analysis revealed genetic diversity among the majority of the MABS clinical isolates. These isolates clustered into distinct clades, separate from globally recognized clinical strains and dominant circulating clones. Inducible clarithromycin resistance was detected in 60.66 % of MABS isolates, associated with the T28 variant in erm(41). The Ile80Val mutation in erm(41) was significantly associated with inducible clarithromycin resistance (χ2 = 12.61, p < 0.001). Acquired clarithromycin resistance associated with rrl mutations (A2270C, A2270G, A2271C) and amikacin resistance linked to the rrs mutation A1375G were detected in 11.48 % and 4.92 % of isolates, respectively. The categorical agreement between WGS-based DST and pDST was 95.08 %, 88.33 %, and 96.43 % for inducible clarithromycin, clarithromycin, and amikacin, respectively.

CONCLUSION: This study provides valuable insights into the genomic diversity and antimicrobial resistance of MABS isolates in Thailand, emphasizing regional variations in dominant clones and resistance mechanisms.

PMID:40796429 | DOI:10.1016/j.jmii.2025.08.003

Nat Biotechnol. 2025 Aug 12. doi: 10.1038/s41587-025-02771-0. Online ahead of print.

ABSTRACT

Precise CRISPR-based DNA integration and editing remain challenging, largely because of insufficient control of the repair process. We find that repair at the genome-cargo interface is predictable by deep learning models and adheres to sequence-context-specific rules. On the basis of in silico predictions, we devised a strategy of base-pair tandem repeat repair arms matching microhomologies at double-strand breaks. These repeat homology arms promote frame-retentive cassette integration and reduce deletions both at the target site and within the transgene. We demonstrate precise integrations at 32 loci in HEK293T cells. Germline-transmissible transgene integration and endogenous protein tagging in Xenopus and adult mouse brains demonstrated precise integration during early embryonic cleavage and in nondividing, differentiated cells. Optimized repair arms also facilitated small edits for scarless single-nucleotide or double-nucleotide changes using oligonucleotide templates in vitro and in vivo. We provide the design tool Pythia to facilitate precise genomic integration and editing for experimental and therapeutic purposes for a wide range of target cell types and applications.

PMID:40796977 | DOI:10.1038/s41587-025-02771-0

Commun Eng. 2025 Aug 12;4(1):151. doi: 10.1038/s44172-025-00488-1.

ABSTRACT

Growing energy storage demand has solidified the dominance of lithium-ion batteries (LIBs) in modern societies but intensifies recycling pressures. Precise state-of-health (SOH) assessment is crucial to grouping retired batteries from an unknown state for secondary utilization. However, batteries in the pack exhibit distinct capacity fading behaviors due to their service scenarios and working conditions. We develop a deep-learning framework for rapid, transferable SOH estimation and battery classification. This framework integrates deep neural networks with interconnected electrochemical, mechanical, and thermal features. Our model delivers optimal accuracy with a mean absolute error (MAE) of 0.822% and a root mean square error (RMSE) of 1.048% using combined features. It demonstrates robust performance across various conditions and enables SOH prediction with data from merely one previous cycle. Moreover, the well-trained model could adapt to other electrode systems with a minimal number of additional samples. This work highlights critical features for SOH estimation and enables efficient battery classification toward sustainable recycling.

PMID:40796946 | DOI:10.1038/s44172-025-00488-1

Sci Rep. 2025 Aug 12;15(1):29497. doi: 10.1038/s41598-025-12531-4.

ABSTRACT

Fistfight detection in video data is a critical task in video surveillance systems, where identifying physical altercations in real-time can enhance safety and security in public spaces. Earlier techniques primarily emphasized capturing inter-person interactions and combining individual characteristics into group-based representations, often overlooking the critical intra-person dynamics within the human bodypose point framework. However, essential individual features can be extracted by examining human skeletal movements' progression and temporal patterns. This paper presents a novel multimodal spatio-temporal fistfight detection model (MSTFDet) that integrates RGB images and human skeletal data to identify violent behaviors accurately. The proposed framework leverages both Context-Aware Encoded Transformer (CAET) for modeling interactions between individuals and their environment and Spatial-Temporal Graph Convolutional Networks (ST-GCN) for capturing intra-person and inter-person dynamics from skeletal data. The RGB module uses a combination of spatial and temporal transformers to model contextual relationships and individual actions, while the bodypose-point module processes skeletal data to capture the fine-grained motion of individuals. We conduct evaluations on two public datasets: the Surveillance Camera Fight Dataset (SCFD) and the RWF-2000 dataset, which feature complex real-world scenarios. On the SCFD and RWF-2000 datasets, MSTFDet achieved a Multi-class classification accuracy (MCA) of 92.3% and 95.2% MCA, respectively. These results highlight the effectiveness of the proposed approach in capturing both spatial and temporal features, providing a robust solution for real-time fistfight detection in diverse and challenging environments.

PMID:40796923 | DOI:10.1038/s41598-025-12531-4

Nat Commun. 2025 Aug 12;16(1):7490. doi: 10.1038/s41467-025-62826-3.

ABSTRACT

Bone marrow adipose tissue, as a distinct adipose subtype, has been implicated in the pathophysiology of skeletal, metabolic, and hematopoietic disorders. To identify its underlying genetic factors, we utilized a deep learning algorithm capable of quantifying bone marrow fat fraction (BMFF) in the vertebrae and proximal femur using magnetic resonance imaging data of over 38,000 UK Biobank participants. Genome-wide association analyses uncovered 373 significant BMFF-associated variants (P-value < 5 × 10-9), with enrichment in bone remodeling, metabolism, and hematopoiesis pathway. Furthermore, genetic correlation highlighted a significant association between BMFF and skeletal disease. In about 300,000 individuals, polygenic risk scores derived from three proximal femur BMFF were significantly associated with increased osteoporosis risk. Notably, Mendelian randomization analyses revealed a causal link between proximal femur BMFF and osteoporosis. Here, we show critical insights into the genetic determinants of BMFF and offer perspectives on the biological mechanisms driving osteoporosis development.

PMID:40796918 | DOI:10.1038/s41467-025-62826-3

Epigenetics Chromatin. 2025 Aug 12;18(1):55. doi: 10.1186/s13072-025-00618-1.

ABSTRACT

BACKGROUND: Topologically associating domains (TADs) are believed to play a role in the regulation of gene expression by constraining or guiding interactions between the regulatory elements. While the impact of TAD perturbations is typically studied in developmental genes with highly cell-type-specific expression patterns, this study examines genes with broad expression profiles separated by a strong insulator boundary. We focused on the mouse Slc29a3/Unc5b locus, which encompasses two distinct TADs containing ubiquitously expressed and essential for viability genes. We disrupted the CTCF-boundary between these TADs and analyzed the resulting changes in gene expression.

RESULTS: Deletion of four CTCF binding sites at the TAD boundary altered local chromatin architecture, abolishing pre‑existing loops and creating novel long‑range interactions that spanned the original TAD boundary. Using UMI-assisted targeted RNA-seq we evaluated transcriptional changes of Unc5b, Slc29a3, Psap, Vsir, Cdh23, and Sgpl1 across various organs. We found that TAD boundary disruption led to variable transcriptional responses, where not only the magnitude but also the direction of gene expression changes were tissue-specific. Current hypotheses on genome architecture function, such as enhancer competition and hijacking, as well as genomic deep learning models, only partially explain these transcriptional changes, highlighting the need for further investigation into the mechanisms underlying TAD function and gene regulation.

CONCLUSIONS: Disrupting the insulator element between broadly expressed genes resulted in moderate, tissue-dependent transcriptional alterations, rather than uniformly activating or silencing the target genes. These findings show that TAD boundaries contribute to context‑specific regulation even at housekeeping loci and underscore the need for refined models to predict the effects of non‑coding structural variants.

PMID:40796890 | DOI:10.1186/s13072-025-00618-1

Sci Rep. 2025 Aug 12;15(1):29519. doi: 10.1038/s41598-025-06951-5.

ABSTRACT

In the medical field, Artificial Intelligence (AI) for diagnostic processes, particularly through deep learning techniques, has become increasingly advanced. Minor trauma, such as accidental cheek biting, sharp dental edges, or poorly fitting dentures, typically causes painful mouth ulcers and bump-like sores inside the mouth. Traditionally, diagnosing these ulcers involves a dentist or physician performing a physical examination, visually assessing the sores, and asking detailed questions about their size, location, duration, and related symptoms. Our research focuses on the advanced classification of oral ulcer stages using a convolutional neural network (CNN). To evaluate performance comprehensively, we developed and tested three custom models, comparing their effectiveness in distinguishing between different stages of oral ulcers. We also explored various optimizers and activation functions to determine the best configuration for improving model performance. Although our models show promising potential as diagnostic tools for oral ulcers, they occasionally make errors. Among the models tested, UlcerNet-2 stood out for its performance. Using the RMSprop optimizer along with Softmax and SELU activation functions, UlcerNet-2 achieved a validation accuracy of 96%. These results highlight UlcerNet-2's exceptional effectiveness in classifying oral ulcer stages, achieving a commendable balance of high accuracy, precision, and recall. This suggests UlcerNet-2 has significant potential as an advanced diagnostic tool, possibly enhancing clinical practices in detecting and staging oral ulcers. The proposed model (UlcerNet) was implemented on FogBus, the cloud framework to empirically evaluate the model performance in cloud-fog interoperable scenarios.

PMID:40796791 | DOI:10.1038/s41598-025-06951-5

Sci Rep. 2025 Aug 12;15(1):29472. doi: 10.1038/s41598-025-13862-y.

ABSTRACT

Multi-Modal Medical Image Fusion (MMMIF) has become increasingly important in clinical applications, as it enables the integration of complementary information from different imaging modalities to support more accurate diagnosis and treatment planning. The primary objective of Medical Image Fusion (MIF) is to generate a fused image that retains the most informative features from the Source Images (SI), thereby enhancing the reliability of clinical decision-making systems. However, due to inherent limitations in individual imaging modalities-such as poor spatial resolution in functional images or low contrast in anatomical scans-fused images can suffer from information degradation or distortion. To address these limitations, this study proposes a novel fusion framework that integrates the Non-Subsampled Shearlet Transform (NSST) with a Convolutional Neural Network (CNN) for effective sub-band enhancement and image reconstruction. Initially, each source image is decomposed into Low-Frequency Coefficients (LFC) and multiple High-Frequency Coefficients (HFC) using NSST. The proposed Concurrent Denoising and Enhancement Network (CDEN) is then applied to these sub-bands to suppress noise and enhance critical structural details. The enhanced LFCs are fused using an AlexNet-based activity-level fusion model, while the enhanced HFCs are combined using a Pulse Coupled Neural Network (PCNN) guided by a Novel Sum-Modified Laplacian (NSML) metric. Finally, the fused image is reconstructed via Inverse-NSST (I-NSST). Experimental results prove that the proposed method outperforms existing fusion algorithms, achieving approximately 16.5% higher performance in terms of the QAB/F (edge preservation) metric, along with strong results across both subjective visual assessments and objective quality indices.

PMID:40796770 | DOI:10.1038/s41598-025-13862-y

Sci Rep. 2025 Aug 12;15(1):29467. doi: 10.1038/s41598-025-14238-y.

ABSTRACT

Pain is a multifaceted phenomenon that significantly affects a large portion of the global population. Objective pain assessment is essential for developing effective management strategies, which in turn contribute to more efficient and responsive healthcare systems. However, accurately evaluating pain remains a complex challenge due to subtle physiological and behavioural indicators, individual-specific pain responses, and the need for continuous patient monitoring. Automatic pain assessment systems offer promising, technology-driven solutions to support and enhance various aspects of the pain evaluation process. Physiological indicators offer valuable insights into pain-related states and are generally less influenced by individual variability compared to behavioural modalities, such as facial expressions. Skin conductance, regulated by sweat gland activity, and the heart's electrical signals are both influenced by changes in the sympathetic nervous system. Biosignals, such as electrodermal activity (EDA) and electrocardiogram (ECG), can, therefore, objectively capture the body's physiological responses to painful stimuli. This paper proposes a novel multi-modal ensemble deep learning framework that combines electrodermal activity and electrocardiogram signals for automatic pain recognition. The proposed framework includes a uni-modal approach (FCN-ALSTM-Transformer) comprising a Fully Convolutional Network, Attention-based LSTM, and a Transformer block to integrate features extracted by these models. Additionally, a multi-modal approach (CrossMod-Transformer) is introduced, featuring a dedicated Transformer architecture that fuses electrodermal activity and electrocardiogram signals. Experimental evaluations were primarily conducted on the BioVid dataset, with further cross-dataset validation using the AI4PAIN 2025 dataset to assess the generalisability of the proposed method. Notably, the CrossMod-Transformer achieved an accuracy of 87.52% on Biovid and 75.83% on AI4PAIN, demonstrating strong performance across independent datasets and outperforming several state-of-the-art uni-modal and multi-modal methods. These results highlight the potential of the proposed framework to improve the reliability of automatic multi-modal pain recognition and support the development of more objective and inclusive clinical assessment tools.

PMID:40796769 | DOI:10.1038/s41598-025-14238-y

Sci Rep. 2025 Aug 12;15(1):29591. doi: 10.1038/s41598-025-12370-3.

ABSTRACT

Rolling bearings are a crucial component in rotating machinery, essential for ensuring the smooth functioning of the entire system. However, their vulnerability to damage necessitates the implementation of effective fault diagnosis. Traditional deep learning methods often struggle due to the scarcity of fault samples, leading to issues like overfitting and inadequate generalization. To address this problem, a novel Siamese Neural Network (SNN) model, integrating Deep Convolutional Neural Networks with Wide First-layer Kernel (WDCNN) and Bidirectional Long Short-Term Memory (BiLSTM) network is proposed. This model constructs a feature extraction system that combines WDCNN and BiLSTM to extract local spatial features and global temporal dependencies from vibration signals. Additionally, the SNN framework is introduced to build a feature space under small sample conditions through metric learning, enhancing the ability of model to discern sample similarities. Experiments on the CWRU and HUST datasets indicate that with only 90 training samples, the model achieves diagnostic accuracy of 83.47% and 61.48%, respectively, significantly surpassing CNN, BiLSTM, and their combined models. Furthermore, the model also shows robustness against severe noise interference, making it a viable tool for efficient fault diagnosis in rolling bearings with limited data.

PMID:40796758 | DOI:10.1038/s41598-025-12370-3

Nat Microbiol. 2025 Aug 12. doi: 10.1038/s41564-025-02061-0. Online ahead of print.

ABSTRACT

Antimicrobial resistance is one of the greatest threats facing humanity, making the need for new antibiotics more critical than ever. While most antibiotics originate from bacteria and fungi, archaea offer a largely untapped reservoir for antibiotic discovery. In this study, we leveraged deep learning to systematically explore the archaeome, uncovering promising candidates for combating antimicrobial resistance. By mining 233 archaeal proteomes, we identified 12,623 molecules with potential antimicrobial activity. These peptide compounds, termed archaeasins, have unique compositional features that differentiate them from traditional antimicrobial peptides, including a distinct amino acid profile. We synthesized 80 archaeasins, 93% of which showed antimicrobial activity in vitro against Acinetobacter baumannii, Escherichia coli, Klebsiella pneumoniae, Pseudomonas aeruginosa, Staphylococcus aureus and Enterococcus spp. Notably, in vivo validation identified archaeasin-73 as a lead candidate, significantly reducing A. baumannii loads in mouse infection models, with effectiveness comparable to that of established antibiotics such as polymyxin B. Our findings highlight the potential of archaea as a resource for developing next-generation antibiotics.

PMID:40796684 | DOI:10.1038/s41564-025-02061-0

Nat Struct Mol Biol. 2025 Aug 12. doi: 10.1038/s41594-025-01614-5. Online ahead of print.

ABSTRACT

Pregnane X receptor (PXR) is a nuclear receptor considered to be a master transcription factor of xenobiotic metabolism. Here, using enhanced ultraviolet crosslinking and immunoprecipitation, we show that PXR can bind mRNAs in different cancer cell lines and normal liver tissues. PXR-bound mRNAs include genes related to metabolic reprogramming and lipid metabolism. Separately from its known nuclear transcriptional function, cytoplasmic PXR binds and stabilizes mature mRNA containing C+G-enriched sequences through its zinc-finger domain. Mechanistically, cytoplasmic PXR interacts with RNH1, an RNase inhibitor, to regulate RNA stability. In colorectal cancer cells, cytoplasmic PXR facilitates glucose uptake by stabilizing SLC2A1 mRNA. This process further promotes cell proliferation and cancer development. Our study unveils a previously unknown dimension of PXR-mediated gene regulation by characterizing PXR as an RNA-binding protein important for mRNA stability in the cytoplasm.

PMID:40797049 | DOI:10.1038/s41594-025-01614-5