Nanobody–antigen molecular recognition underpins nanobody discovery and development, necessitating accurate determination of binding occurrence, interface residues, and affinity. Current predictors are architecturally designed for the massive, heterogeneous spectrum of general protein–protein interactions, diluting the limited, complementarity-determining region (CDR)-dominated nanobody–antigen interaction (NAI) data and masking the decisive CDR signal. The scarcity of experimental affinity data precludes direct regression-based estimation of binding affinity. Here, we present NanoBind, a mechanism-driven deep learning framework that embeds the CDR-dominated binding pattern within its encoder, enabling robust prediction of binding occurrence and interface residues from limited NAI data. Constrained by scarce affinity data, NanoBind generates quantitative affinity ranges for nanobody–antigen pairs without extra experiments. Systematic benchmarking demonstrates that NanoBind surpasses state-of-the-art methods in accuracy and robustness, and interpretability analyses confirm that the model’s decisions align with the CDR-dominated binding mechanism. When million-sequence immune repertoires are screened against 4 antigens, NanoBind reduces candidate nanobodies to fewer than 100 per target. For the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) spike receptor-binding domain (RBD)–nanobody F2 complex, NanoBind correctly predicts binding occurrence, matches experimentally validated interface residues, and generates an affinity range quantitatively supported by molecular dynamics simulations. A server is available at http://liulab.top/NanoBind/server.

The continuous development of bright x/gamma-ray sources has opened up new frontiers of science and advanced applications. Currently, there is still a lack of efficient approaches to produce high-energy gamma-rays with peak brilliance comparable to modern free-electron lasers. Here, we report a novel mechanism called beam fast pinching radiation burst to generate such gamma-ray sources. It is achieved by injecting a giga-electron volt electron beam into a submillimeter plasma with an upramp density profile, enabling violent beam pinching to occur rapidly. During this process, a burst of collimated gamma-rays is efficiently produced with photon energies ranging from mega-electron volts to giga-electron volts, electron-to-photon energy conversion efficiency above 20%, and peak brilliance exceeding 10²⁸ photons s⁻¹ mm⁻² mrad⁻² per 0.1% bandwidth. All of these are several orders of magnitude higher than existing gamma-ray sources. This opens a novel avenue for the development of extremely bright gamma-ray sources for both fundamental research and cutting-edge applications.

Persistent androgen receptor (AR) signaling, metabolic alterations, and changes in the immune microenvironment are contributing factors to the progression of prostate cancer (PCa); however, therapeutic strategies capable of simultaneously targeting these interconnected vulnerabilities remain limited. Here, we identify a combinatorial regimen consisting of dihydroartemisinin and eupatilin (D&E) as a suppressor of PCa progression through coordinated disruption of tumor-intrinsic survival signaling and protumor immune interactions. Functionally, D&E suppressed the proliferation of PCa cells. Mechanistically, D&E induces ferroptosis in PCa cells, as evidenced by lipid peroxidation, intracellular iron accumulation, malondialdehyde elevation, and glutathione reduction. We further identified AR as an important target of D&E and found that it promotes resistance to ferroptosis by sustaining the SLC7A11 antioxidant axis through transcriptional activation of SLC7A11. Upstream, D&E inhibited the NF-κB subunit p65, which transcriptionally maintains AR expression, thus establishing an NF-κB/AR/SLC7A11 signaling cascade underlying ferroptosis resistance in PCa. Beyond tumor-intrinsic effects, D&E also remodels the tumor microenvironment. Specifically, AR-high tumor cells were found to facilitate M2-like macrophage polarization. In turn, M2-like macrophages could secrete SPP1, which potentially augments malignant properties and AR signaling in tumor cells via the CD44 pathway, thereby constituting a protumorigenic positive feedback loop. Notably, D&E attenuated the M2-like phenotype, reduced SPP1 secretion, and reduced macrophage-mediated promotion of PCa proliferation. In immunocompetent mouse models, D&E further synergized with anti-PD-L1 to suppress PCa progression. Together, these findings indicate that D&E exert antitumor effects in PCa by promoting ferroptosis-associated cell death and modulating macrophage-associated microenvironmental signaling, with potential therapeutic relevance particularly in AR-positive PCa.

Computational pathology enables scalable analysis of pathology images for cancer diagnosis and research, but conventional deep learning models remain constrained by annotation dependence and task-specific development. Recent advances in self-supervised learning, transformer-based architectures, and large-scale pretraining have given rise to computational pathology foundation models (CPathFMs), which are designed to learn transferable and reusable representations from diverse pathology data. This review first traces the development of CPathFMs by examining the data resources, backbone architectures, representation levels, input modalities, and pretraining strategies that shape their design. We then examine their applications across major cancer pathology tasks, including tumor detection, grading, and subtyping, molecular biomarker and gene expression prediction, prognostic assessment, tissue phenotyping, as well as multimodal retrieval and report generation. We further discuss key challenges that limit translation and deployment, including underrepresentation of normal tissues and benign lesions, limited generalization across domains and institutions, data overlap and benchmark contamination, task-centric evaluation and insufficiently standardized metrics, high training and infrastructure costs, and ethical concerns in model governance. Overall, CPathFMs mark an important shift from task-specific pathology artificial intelligence toward reusable representation learning. Their clinical value remains to be established through improved model performance and generalizability, more comprehensive and standardized evaluation frameworks, and prospective evidence of utility in real-world pathology workflows.

Fe single-atom catalysts with well-defined porous architectures and optimized Fe–N_x microenvironments show great potential for enhancing the oxygen reduction reaction. Herein, we report Fe–N₅ single-atom catalysts, where each Fe atom coordinates with 4 pyridinic N and 1 axial pyrrolic N, for efficient oxygen reduction. A dual-confinement strategy, combining the wood framework with Fe³⁺ coordination, directs self-assembly of cellulose nanocrystals (CNCs) into a porous wood-derived architecture. Subsequent pyrolysis yields Fe–N₅ catalysts anchored on N, S-codoped carbon with hollow, hierarchically interconnected 3-dimensional pores. Notably, coordination between CNCs and Fe³⁺ guides the formation of Fe–N_x moieties within a tailored microenvironment, enabling control over the coordination number, heteroatom doping, and the electronic structure. X-ray absorption spectroscopy and density functional theory calculations reveal that FeN₅ moieties are optimized through 3 synergistic factors: Fe coordination with 4 pyridinic N and 1 axial pyrrolic N, S doping from residual sulfate ester groups in CNCs, and adjacent micropores. Collectively, these effects lower the *OH desorption barrier, accelerating the adsorption/desorption of oxygenated intermediates. Consequently, Fe–N₅ single-atom catalysts exhibit an exceptional oxygen reduction reaction activity with a half-wave potential of 0.964 V. This dual-confinement strategy enables high-performance non-precious-metal catalysts for metal–air batteries, as evidenced by Fe–N₅-based zinc–air batteries outperforming Pt/C.

Chiral spin textures in magnetic insulators promise magneto-electric (ME) spintronics with orders-of-magnitude lower power consumption than metallic systems. However, realizing the short magnetic periods required for high-density device integration remains difficult, as conventional Dzyaloshinskii–Moriya interaction (DMI)-based mechanisms typically constrain spiral periods to tens of nanometers. While theory predicts that strong single-ion anisotropy (SIA) on frustrated lattices can stabilize complex non-coplanar textures, the potential for using this mechanism to engineer such compact textures remains largely unexplored. Here, we report that a cubic chiral insulator Co₅TeO₈ provides an experimental example of this paradigm. Comprehensive neutron scattering and magnetometry reveal helimagnetic spirals with continuously tunable pitch of 5.7 to 10 nm embedded in a complex phase diagram spanning 8 distinct phases. Capacitance anomalies throughout the phase diagram indicate ME coupling, pointing to the possibility of future E-field control of these textures. The temperature and field dependence of the helical wavevector strongly support a scenario in which site-dependent SIA provides the leading contribution to the selection of the helical period from a frustration-induced degenerate manifold. Consistent with this interpretation, ab initio calculations place SIA approximately an order of magnitude above DMI, distinct from conventional helimagnets. Co₅TeO₈ thus offers an experimental realization of sub-10-nm helimagnetism and motivates a design principle for anisotropy-engineered correlated insulators.

Extensive data resources are critical for broadening the application scope and improving the performance of large language models (LLMs). Local deployment of LLMs enables users to leverage the advanced capabilities of LLMs while maintaining secure access to local data. Therefore, local LLMs deliver more domain-specific services than general LLMs. However, concerns over data security and the absence of economic incentives restrict cross-industry collaborative utilization of data resources. To address this challenge, an interdisciplinary framework integrating artificial intelligence, economic theory, and data security technologies is required to explore the data trading mechanism for local LLMs. The key aspects of a data market framework, economic evaluation for data pricing, and trusted data trading are analyzed to support efficient utilization of cross-industry data resources. By constructing a data trading mechanism, industry-specific local LLMs can be further enhanced through cross-industry knowledge integration, thereby improving productivity across sectors.

Patients with diabetic kidney disease (DKD) at chronic kidney disease (CKD) stages 3 to 4 are at high risk for rapid renal function decline within 1 year. However, owing to the multifactorial complexity of the disease, effective prognostic tools that integrate multidimensional clinical and pathological information are currently lacking for this specific population. We conducted a retrospective cohort study involving 322 patients with biopsy-proven DKD (CKD stages 3 to 4) from the China-Japan Friendship Hospital and Hebei University Affiliated Hospital. Their clinical data and 2,576 renal biopsy pathology images were used to develop and validate a multimodal model. Machine learning was applied to integrate clinical composite indices and renal biopsy images to develop a prognostic prediction tool. Four key clinical predictors were identified: estimated glomerular filtration rate, 24-h urinary protein, systemic immune inflammation index, and estimated pulse wave velocity. Among the 6 machine learning algorithms used to develop the prediction models, the random forest algorithm achieved the best performance in the test set, with an area under the receiver operating characteristic (ROC) curve (AUC) of 0.889 and a precision–recall AUC (PR-AUC) of 0.921 for predicting the 1-year composite renal endpoint. The integration of pathological features led to a marked improvement in the performance of the model (ROC-AUC: 0.923 vs. 0.898). External validation demonstrated that incorporating pathological information into the model increased the ROC-AUC from 0.885—achieved when clinical composite indices alone were used as predictors—to 0.930. In this study, machine learning-based automated image analysis of glomerular crescent-shaped changes and renal interstitial fibrosis was integrated with established clinical composite indices to construct an accurate model for predicting short-term renal prognosis of DKD at CKD stages 3 to 4 and to provide a potential tool for improved risk stratification.

Pulmonary fibrosis (PF) is a fatal interstitial lung disease characterized by excessive extracellular matrix deposition and irreversible architectural distortion. The mechanisms driving the transition from tissue repair to fibrosis are complex and remain poorly understood. By analyzing interpatient variation across 75 idiopathic PF lungs, we identified a conserved profibrotic macrophage subset, distinct from canonical M1/M2 or SPP1⁺ states, characterized by high legumain (LGMN) expression and enrichment of gene signatures implicated in leukocyte activation and matrix remodeling. LGMN⁺ macrophages localize within fibroblastic foci and are associated with disease progression and poor prognosis. Lineage-tracing and RNA velocity analyses revealed that LGMN⁺ macrophages arise from monocytes through fibroblast-derived macrophage colony-stimulating factor signaling, which activates Maf BZIP transcription factor B-dependent differentiation programs. Pharmacological inhibition or macrophage-specific deletion of Lgmn markedly attenuated bleomycin-induced lung fibrosis, reduced extracellular matrix accumulation, and improved lung architecture. LGMN activates cathepsin S to mediate degradation of basement membrane collagen IV, thereby disrupting the alveolar–capillary barrier. In parallel, secreted LGMN acts as a paracrine signal to activate fibroblasts and promote collagen I deposition, collectively fostering a profibrotic niche. Together, these findings establish LGMN as a macrophage effector that links immune activation to matrix remodeling, thereby driving the transition from tissue injury to fibrosis.

Monitoring adoptive cell therapy in solid tumors is critical for evaluating treatment efficacy and guiding clinical medication but is also hindered by poor sensitivity, high background signals, and disruptions of therapeutic functions in existing techniques. In this study, a membrane-mimicking conjugated oligoelectrolyte with second near-infrared (NIR-II) fluorescence, conjugated oligoelectrolytes-benzobisthiadiazole (COE-BBT), is applied for the first time to label and track chimeric antigen receptor (CAR)-engineered natural killer (CAR-NK) cells and T (CAR-T) cells in vivo. COE-BBT stably embeds in lipid bilayers through combined electrostatic and hydrophobic interactions, resists membrane crossing, and supports long-lasting labeling with a lighting-up property. The optimized labeling approach achieves high sensitivity, enabling the detection of as few as ~20 labeled cells in vitro and ~50 cells in vivo under NIR-II imaging. In orthotopic and subcutaneous glioma models, NIR-II fluorescence imaging enables continuous tracking of CAR-NK and CAR-T cell proliferation, migration, tumor homing, and blood–brain barrier penetration for up to 14 d posttransfer as the fluorescence signal is enhanced during proliferation, without compromising cell viability or cytotoxic function. The COE-BBT probe also exhibits favorable biosafety, underscoring its translational potential as a robust imaging strategy to improve solid tumor adoptive cell therapy monitoring and clinical guidance of therapeutic dosing.