安徽医科大学

Structure and dynamical prediction are two core tasks in multiplex network research, yet they are often investigated separately. However, in real-world scenarios, the structural and dynamical information is typically simultaneously available but incomplete, which makes these tasks more challenging. Therefore, a natural idea to utilize such incomplete information for collaborative modeling of the two tasks. To address these challenges, we propose a data-driven collaborative learning framework (Multi-CoLF) that integrates incomplete structural and dynamical information for predicting network structure and dynamical evolutions of multiplex networks. The framework consists of a structure prediction learner and a dynamical prediction learner. Specifically, the structure prediction learner employs the Gumbel-Softmax technique to infer missing links in each layer and constructs an extended graph. Based on this extended graph, the dynamical prediction learner develops an extended graph neural network to model spreading dynamics and capture the underlying evolution rules. An alternating optimization strategy is further designed to enable collaborative optimization between the two learners, facilitating collaborative prediction of structure and dynamics. Experiments on three synthetic and five real-world networks demonstrate that our method outperforms baseline approaches on both tasks under different types of spreading dynamics, with structure prediction accuracy improving by more than 10% on certain datasets. Moreover, extensive experimental results reveal that the structure and spreading dynamics in multiplex networks are complementary: the learned network structure enhances the accuracy of dynamical prediction, while the learned spreading dynamical rules improve the accuracy of structure prediction.

Metal-ion-mediated lethal pathways offers a highly potent strategy for antibacterial treatments. However, relying on single-pathway mechanisms often fails to overwhelm bacterial antioxidant defenses, reducing antibacterial efficacy. This study designed a heterojunction nanozyme (Cu2O/Fe2O3) that induces ferroptosis-like death and copper-mediated metabolic toxicity exhibiting cuproptosis-like features, achieving highly effective antibacterial treatment. Computational chemistry analysis indicates that the Cu2O/Fe2O3 heterointerface establishes a built-in electric field through spontaneous interfacial electron transfer, thereby enhancing its peroxidase (POD)-like activity. Within the acidic bacterial microenvironment, Cu2O/Fe2O3 catalyzes the generation of large amounts of reactive oxygen species (ROS) and releases metal ions, inducing and amplifying ferroptosis-like death, which disrupts the function of ETC enzymes by attacking iron–sulfur clusters. Concurrently, copper ions disrupt the redox balance by depleting bacillithiol (BSH), indirectly increasing bacterial susceptibility and promoting ferroptosis-like death. Furthermore, the accumulation of copper ions triggers severe metabolic impairments that biochemically resemble the cuproptosis cascade, which disrupts key enzymes of the TCA cycle and impairs the synthesis of electron donors (NADH and FADH2), thereby cutting off the supply of substrates to ETC at the upstream stage. Overall, this study demonstrates that triggering bimetallic ion-mediated metabolic disruptions induces an energy crisis in bacteria, providing a reference for multimodal antibacterial strategies.

Renal fibrosis is a progressive pathological process driving chronic kidney disease (CKD) toward end-stage renal failure, yet effective anti-fibrotic therapies remain unavailable. The mTORC1 pathway critically promotes fibrosis by enhancing extracellular matrix (ECM) production and myofibroblast activation, but its upstream regulatory mechanisms in the kidney are poorly defined. Here, we identify PREX1 as a novel upstream activator of mTORC1 in renal fibrosis. We show that PREX1 drives fibrogenesis by positively regulating RPTOR, a core component of the mTORC1 complex. In human fibrotic kidneys and murine UUO models, PREX1 was markedly upregulated and closely correlated with ECM deposition. Mechanistically, PREX1 and RPTOR were found to reside in the same protein complex, and PREX1 increased RPTOR abundance; whether this association is direct was not determined. Silencing RPTOR abolished PREX1-driven S6K1 phosphorylation, placing RPTOR downstream of PREX1 in mTORC1 activation. Importantly, genetic silencing of either PREX1 or RPTOR, both in vitro and in vivo, significantly suppressed fibrotic and inflammatory responses and ameliorated renal pathology. Collectively, our findings uncover a novel PREX1-RPTOR-mTORC1 signaling axis that governs renal fibrogenesis and highlight the PREX1-RPTOR-mTORC1 axis as a promising therapeutic target for chronic kidney disease.

Accurate trace gas detection is crucial for the diagnosis of constipation and other gastrointestinal functional disorders. In this work, a gas filtering and inversion algorithm that integrates wavelet transform (WT) with a multilayer perceptron (MLP) was developed for accurate measurement of exhaled CH4. A direct absorption spectroscopy based gas sensing system was designed and constructed. The performance of the proposed WT-MLP algorithm was evaluated by comparing it with the adaptive boosting decision tree (ABDT) and the convolutional neural network (CNN) models. The relative error of WT-MLP was reduced by 67.36% and the absolute error by 56.16% compared with ABDT on the same dataset. Furthermore, a 1.80-fold improvement in signal-to-noise ratio was obtained using the WT-MLP algorithm relative to 100-time averaging. Allan variance analysis yielded a limit of detection of 63.46 ppb at an integration time of 33s. Exhaled CH4 concentrations in volunteers and patients with constipation were successfully measured, and the results validated the potential practical value of the WT-MLP algorithm for laser-spectroscopy-based breath diagnosis.

Objectives To characterise senescence-associated fibroblast alterations in periodontitis (PD), identify relevant molecular targets, and examine the relationship between CXCL12–CXCR4 signalling and T-cell responses. Methods We integrated single-cell RNA-sequencing of 26 human gingival samples with bulk and spatial transcriptomics, Xenium in situ profiling, weighted gene coexpression network analysis, five machine-learning algorithms, tissue imaging, and cell-based assays. H₂O₂-induced senescent human gingival fibroblasts and their conditioned medium were used to assess CXCL12 expression, Jurkat-cell migration, and inflammatory gene expression in the presence or absence of the CXCR4 antagonist LY2510924. Results Fibroblasts exhibited the highest senescence scores among the major gingival cell populations, with elevated scores in PD. High-senescence fibroblasts showed inflammatory, chemokine, extracellular-matrix-remodelling and metabolic alterations, together with enhanced predicted communication with immune cells. Integrated analyses prioritised CXCL12, PECAM1 and PLAT as PD- and senescence-associated candidate genes. Their combined signature yielded areas under the receiver operating characteristic curve of 0.944 and 0.902 in the discovery and validation datasets, respectively. CXCL12 was predominantly expressed by fibroblasts, enriched in high-senescence fibroblasts and spatially associated with CXCR4-positive T cells. H₂O₂-induced fibroblast senescence increased CXCL12 expression. Conditioned medium from H₂O₂-treated fibroblasts increased Jurkat-cell migration and inflammatory-gene expression, whereas LY2510924 attenuated these responses. Conclusions Multiomics and machine-learning analyses identified senescence-associated molecular alterations in PD. The findings associate fibroblast CXCL12 expression with a CXCR4-antagonist-sensitive component of Jurkat-cell migration and inflammatory transcription in vitro, supporting further investigation of this pathway as a potential host-directed target.