Multimodal sentiment analysis (MSA) aims to assess emotional states by integrating information from text, audio, and video. However, the heterogeneous nature of these modalities presents substantial challenges for accurate sentiment prediction. Existing approaches typically align pairs of modalities using attention mechanisms or contrastive learning, which are computationally expensive. Additionally, they often rely on a single granularity of alignment, either by averaging features over all time steps or aligning features at each individual time step. These approaches overlook the fact that emotional expression can vary across individuals and contexts, requiring multiple granularities to capture emotion effectively. To address these challenges, we propose a novel framework, Retrieve, Rank, and Reconstruction with Different Granularities (R3DG). R3DG segments the audio and video modalities into multiple representations at varying granularities based on their temporal durations. It then selects the most relevant representations that align closely with the text modality. To preserve the original information, R3DG reconstructs the audio and video data using the selected representations. Finally, the fused audio, video, and text features are aligned and combined for sentiment prediction, reducing the need for multiple alignment steps. Extensive experiments on 5 benchmark MSA datasets demonstrate that R3DG outperforms existing methods and achieves substantial reductions in computational time. Code is available at https://github.com/YetZzzzzz/R3DG.

Magnetic microrobots with noncontact and real-time control capabilities have garnered marked attention for targeted drug delivery in narrow, enclosed pathways within the human body. The manufacturing method of these magnetic robots plays a crucial role in determining their functionality. In this study, a photocuring 3-dimensional (3D) printing technique with in situ pixel-scale magnetic programming was developed, enabled by a 3D large-scale uniform magnetic field generator with a high strength of approximately 50 mT. Magnetic particles were rotated and aligned on demand to print intelligent structures with a spatial resolution of 50 μm. A novel key-node splicing magnetization method was introduced to control multicurved deformations in 1D strips and 2D membrane magnetic robots, enabling various modes of locomotion, such as rolling, creeping, swimming, and patch-based drug release. To support additional functions, 3D spatial magnetization was implemented for customized spiral capsule robots, allowing precise multidirectional swimming and multitarget droplet-based drug delivery. These multimode and multifunctional magnetic actuators were validated through in vivo operations in confined environments such as the gastrointestinal tract and bladder.

Lupus nephritis is recognized as a common and severe complication of systemic lupus erythematosus, without an optimal therapeutic strategy currently available. While mesenchymal stem cells (MSCs) hold therapeutic promise, their efficacy varies substantially, likely due to their plasticity and capacity to adopt pro-inflammatory (MSC1) or anti-inflammatory (MSC2) functional states in response to different microenvironments. Here, we report for the first time that IL-27, via JAK1–STAT1 signaling, up-regulates indoleamine 2,3-dioxygenase (IDO) in MSCs, driving MSC differentiation toward an IDO-positive MSC2 phenotype with low immunogenicity. These IDO-positive MSC2 cells produce kynurenine and kynurenic acid, the metabolites of tryptophan, which bind to the intracellular aryl hydrocarbon receptor. This interaction stimulates an increase in the anti-inflammatory factor TSG-6 and induces the differentiation of regulatory T cells. Notably, IL-27-conditioned MSC2 demonstrated superior therapeutic efficacy compared to conventional MSCs in a murine lupus nephritis model. In conclusion, this study revealed that IL-27 is a critical modulator of MSC immune plasticity and presented a novel therapeutic strategy utilizing IL-27-enhanced MSC2 for autoimmune diseases.

Glucocorticoid resistance increases the frequency of acute exacerbations and the risk of death in chronic obstructive pulmonary disease (COPD) patients with a history of long-term heavy smoking. In this study we aimed to investigate the role of neutrophil ferroptosis resistance and the formation of neutrophil extracellular traps (NETs) in cigarette smoke (CS)-induced glucocorticoid resistance in COPD. We collected clinical specimens from COPD patients and healthy subjects. A mouse model of COPD induced by CS exposure was established in vivo. Neutrophils were isolated from the peripheral blood of human donors and exposed to CS extract in vitro. We found extensive NET formation was observed in COPD patients with a history of long-term heavy smoking and was closely related to glucocorticoid resistance. In vivo, we found that prolonged CS exposure promoted NET formation and that rendered dexamethasone (Dex) treatment ineffective at alleviating lung inflammation in COPD model mice. However, the NET degrading agent deoxyribonuclease I could increase sensitivity to Dex in COPD model mice. In vitro experiments demonstrated that CS extract increased neutrophil cell viability by activating the Nrf2/SLC7A11/GPX4 pathway and inducing ferroptosis resistance in neutrophils. And we found that neutrophil specific GPX4 knockout inhibited CS-induced NET formation, increased sensitivity to Dex, and alleviated CS-induced glucocorticoid resistance in vivo and in vitro. In conclusion CS promotes glucocorticoid resistance in COPD by inducing ferroptosis resistance in neutrophils, further resulting in NET formation.

Advanced drug delivery systems integrating immunomodulatory functions can improve cancer therapy. Herein, we develop a novel metal-phenolic capsule for synergistic drug delivery and immunomodulation. Specifically, a stable metal-phenolic network (MPN), formed by Fe (III), tannic acid, and catechol-modified hyaluronic acid, was assembled onto doxorubicin (DOX)-loaded mannose-glycated bovine serum albumin microbubbles to construct microcapsules with both drug loading and immunomodulatory functions simultaneously. The capsules release DOX in a pH-responsive manner and induce reactive oxygen species accumulation in tumor cells, thereby enhancing immunogenic cell death (ICD). This ICD effect, combined with the direct stimulation of dendritic cells (DCs) by the capsules and the paracrine signaling from capsule-activated M1 macrophages, synergistically promotes DC maturation. In vivo bilateral tumor models demonstrate that DmTFH substantially inhibits primary and distant tumor development and markedly delays metastasis to the lungs. Flow cytometry analysis confirms robust local and systemic immune responses, characterized by enhanced DC maturation within lymph nodes and increased infiltration of activated CD8⁺ and CD4⁺ T cells in tumors and spleens. Overall, the immunoactive MPN microcapsules constructed in this study enhance the immunotherapeutic efficacy of DOX through the synergistic action of carrier-inherent immunomodulation and chemotherapy, providing a promising approach for combinatorial cancer therapy.

Reliable detection of high-g shocks in extreme impact scenarios, such as automobile collisions, is essential for ensuring occupant safety. Conventional shock sensors based on piezoresistive or capacitive mechanisms often underperform in high-g environments due to their structural complexity, resulting in delayed or missed detection. Here, we present a self-powered high-g shock sensor that combines a triboelectric transducer with a prestressed structure to deliver large signal amplitude and minimal oscillation. The prestress mechanism enhances initial contact strength, achieving a 400% increase in signal amplitude and reduced oscillation. We further developed a self-powered, compact (<4.5 cm³) microsystem that integrates the shock sensor, a signal processing module, airbag triggering circuitry, and a high-g-resistant supercapacitor as a backup power source. The microsystem achieves ultra-fast shock detection and airbag activation with a delay of less than 0.2 ms. Furthermore, its power demand is 80% lower than that of commercial high-g sensors, while the pre-charged supercapacitor ensures operational stability. To further extend the functionality of the device, we designed a lightweight collision target classification algorithm using ensemble learning and feature importance analysis, which could accurately distinguish between automotive collisions with hard, brittle, and soft materials. This study advances triboelectric nanogenerators for high-g shock sensing, offering improved reliability, performance, and real-world adaptability.

Atherosclerotic plaques pose a substantial risk for life-threatening cardiovascular complications due to their propensity to trigger acute clinical events. Foam cell formation, resulting from dysregulated lipid homeostasis, serves as a pivotal pathological hallmark in the progression of atherosclerosis. In this study, we presented a precision therapeutic strategy targeting foam cells of multiple origins. The carrier-free nanomedicine Dlut based on luteolin was constructed, featuring CD44 active targeting and stimulus-driven traceless release. Dlut demonstrated active foam cell targeting capability and complete disintegration triggered by oxidative stress and acidic microenvironment, enabling a traceless release of luteolin. Transcriptomic profiling revealed that Dlut inhibited foam cell formation by accelerating lipid efflux. In vivo studies further demonstrated that Dlut significantly reduced plaque burden and improved plaque stability, highlighting a translational potential of atherosclerosis.

The development of terahertz science and technology has shown new application prospects in artificial intelligence. Terahertz stimulation can lead to information communication of cells. Terahertz unipolar picosecond pulse train stimulation can activate cell membrane hydrophilic pores and protein ion channels. However, the effect of the envelope of the terahertz unipolar stimulation remains unknown. This paper studies the effect of the envelope on membrane communication-related variables and the accompanying energy consumption by a cell model with considerations of hydrophilic pores and Na⁺, K⁺-ATPase. According to the results, terahertz unipolar picosecond pulse train stimulation can deliver the signal contained in its envelope into the variation rates of membrane potentials no matter whether the hydrophilic pores are activated or not and also into the variation rates of the ion flow via the pores after activation of the pores. In contrast, the ion flow via Na⁺, K⁺-ATPase seems irrelevant to the signal in the envelope. Moreover, the ion flows show a modulation effect on the variation rates of membrane potentials. The accompanying power dissipations in the cases of different envelopes are similar, as low as around the level of 10⁻¹¹ W. The results lay the foundations for application in artificial intelligence, like brain–machine communications.

Stem cells have important applications in both regenerative and reproductive medicine. The cohesin complex comprises 4 core subunits, namely, SMC1, SMC3, RAD21, and STAG, and notably, it plays pivotal roles in controlling the fate determinations of stem cells by facilitating the dynamic regulation of the 3-dimensional genome architecture. We have recently reported that RAD21 forms a complex with YAP1 and NEDD4 to promote the self-renewal of human spermatogonial stem cells and inhibit their apoptosis. In this review, we address the molecular properties of the cohesin complex and its multiple regulatory mechanisms in mediating the fate decisions of various kinds of stem cells, including hematopoietic stem cells, embryonic stem cells, spermatogonial stem cells, neural stem cells, and other types of stem cells. By maintaining the chromatin loop structure, the cohesin complex is involved in DNA repair and gene transcription, which in turn controls the pluripotency, self-renewal, and differentiation of stem cells. In addition, the cohesin complex ensures faithful DNA replication and sister chromatid cohesion, which indirectly supports genetic and epigenetic programs. Variants in the subunit components of the cohesin complex and proteins’ modifications further confer functional plasticity, and its mutations can lead to abnormal stem cell functions and are correlated with diseases including cancers. Future studies need to integrate multidisciplinary approaches including single-cell multi-omics and cryo-electronic microscopy to resolve the dynamic regulatory networks of the cohesin complex in stem cell fate regulation and further explore its potential applications in regenerative and reproductive medicine.

Artificial neural networks have long been studied to emulate the cognitive capabilities of the human brain for artificial intelligence (AI) computing. However, as computational demands intensify, conventional hardware based on transistor and complementary metal oxide semiconductor (CMOS) technology faces substantial limitations due to the separation of memory and processing, a challenge commonly known as the von Neumann bottleneck. In this review, we examine how memristors, which are novel nonvolatile memory devices that exhibit memory-dependent resistance, can be harnessed to build more efficient and scalable neural networks. We provide a comprehensive background on the evolution of neural network models and memristors, as well as introduce the principles of memristive devices, which mimic the dynamic behavior of biological synapses. Various neural network architectures, including convolutional, recurrent, and spiking models, are discussed, highlighting the advantages of integrating memristors for in-memory computing and parallel processing. Our review further examines key mechanisms such as synaptic plasticity, encompassing both long-term potentiation and depression, as well as emerging learning algorithms that leverage memristive behavior. Finally, we identify current challenges, such as achieving ultra-low power consumption, high device uniformity, and seamless system integration, and propose future directions in materials science, device engineering, system integration, and industrialization. These advances suggest that memristor-based neural networks may pave the way for next-generation AI systems that combine low power consumption with high computational performance, ultimately bridging the gap between biological and electronic information processing.