The explicit roles of the hardly avoidable oxygen species on carbon materials in various fields remain contentious due to the limitations of characterization techniques, which lead to a lack of fundamental understanding of carbon surface chemistry. This study delves exhaustively into the comprehension of the features of different oxygen-modified carbons through the dynamic evolution of surficial oxygen functional groups. Significant differences of thermal stability and electronic properties among various oxygen species are elucidated via in situ characterizations and theoretical calculations, providing a reliable benchmark for identifying oxygen functional groups on carbon materials. The chemical properties of the carbon materials are simultaneously investigated to show the influence of the oxygen functional groups on carbon structures, redox stability, and scalable metal adsorption. These findings not only consider the common misconception that oxygen species produced under various conditions possess identical properties but also raise awareness of understanding carbon surface chemistry in the atomic level.

Glutathione (GSH), a pivotal antioxidant, plays a critical role in Parkinson's disease (PD) pathogenesis, where oxidative stress drives neuronal damage. However, the dynamic changes of GSH in the brain remain poorly understood. Here, we develop a cyclometalated iridium complex (Ir-DO) for GSH monitoring via multiphoton phosphorescence lifetime imaging. Ir-DO exhibits exceptional optical properties, including long phosphorescence lifetime, deep tissue penetration, and robust multiphoton response. With high biocompatibility, excellent selectivity for GSH, and a low detection limit (300 nM) under physiological pH, Ir-DO enables precise tracking of GSH dynamics in cellular models and brain tissue slices. Using Ir-DO, we reveal a significant reduction of GSH in the substantia nigra of PD mice compared to healthy controls, providing direct evidence for oxidative stress involvement in PD progression. This study not only establishes Ir-DO as a powerful tool for investigating PD mechanisms but also highlights its potential for early diagnostic applications.

Machine learning models can suffer from performance degradation when applied to new tasks due to distribution shifts. Feature representation learning offers a robust solution to this issue. However, a fundamental challenge remains in devising the optimal strategy for feature selection. Existing literature is somewhat paradoxical: some advocate for learning invariant features from source domains, while others favor more diverse features. For better understanding, we propose a statistical framework that evaluates the utilities of the features (i.e., how differently the features are used in each source task) based on the variance of their correlation to y across different domains. Under our framework, we design and analyze a learning procedure consisting of learning content features (comprising both invariant and approximately shared features) from source tasks and finetuning them on the target task. Our theoretical analysis highlights the significance of learning approximately shared features-beyond strictly invariant ones-when distribution shifts occur. Our analysis also yields an improved population risk on target tasks compared to previous results. Inspired by our theory, we introduce ProjectionNet, a practical method to distinguish content features from environmental features via explicit feature space control, further consolidating our theoretical findings.

Multi-source domain adaptation aims to reduce performance degradation when applying machine learning models to unseen domains. A fundamental challenge is devising the optimal strategy for feature selection. Existing literature is somewhat paradoxical: some advocate for learning invariant features from source domains, while others favor more diverse features. To address the challenge, we propose a statistical framework that distinguishes the utilities of features based on the variance of their correlation to label y across domains. Under our framework, we design and analyze a learning procedure consisting of learning approximately shared feature representation from source tasks and fine-tuning it on the target task. Our theoretical analysis necessitates the importance of learning approximately shared features instead of only the strictly invariant features and yields an improved population risk compared to previous results on both source and target tasks, thus partly resolving the paradox mentioned above. Inspired by our theory, we proposed a more practical way to isolate the content (invariant+approximately shared) from environmental features and further consolidate our theoretical findings.

We have developed a new method for treating Alzheimer’s disease (AD) by targeting the LRP1 receptor at the blood-brain barrier (BBB) using a nanoscopic multivalent scaffold decorated with LRP1 targeting peptides. Our experiments on AD model mice have demonstrated that this treatment significantly reduces amyloid-β (Aβ) deposits and improves cognitive function. This study introduces a new approach to drug design that combines multivalent targeting with controlling membrane trafficking using the same tools for nanocarrier design, creating a novel therapeutic intervention. In doing so, we emphasize the crucial role that the BBB plays in AD pathogenesis, highlighting the vital importance of LRP1-mediated Aβ clearance. ### Competing Interest Statement The authors have declared no competing interest.

Oxygenated carbon materials exhibit outstanding electrocatalytic performance in the production of hydrogen peroxide (H2O2) through a two-electron oxygen reduction reaction. The nature of the active functional group and underlying reaction mechanism, however, remain unclear. Here, a comprehensive workflow was established to identify the active sites from the numerous possible structures. The common hydroxyl group at the notched edge demonstrates a key role in the two-electron process. The local chemical environment weakens the binding of OOH intermediate to substrate while enhancing interaction with solution, thereby promoting the H2O2 production. With increasing pH, the intramolecular hydrogen bond between OOH intermediate and hydroxyl decreases, facilitating OOH desorption. Furthermore, the rise in selectivity with increasing potential stems from the suppression of the four-electron process. The active site was further validated through experiments. Guided by theoretical understanding, optimal performance was achieved with high selectivity (>95%) and current density (2.06 mA/cm2) in experiment.

We revisit data selection in a modern context of finetuning from a fundamental perspective. Extending the classical wisdom of variance minimization in low dimensions to high-dimensional finetuning, our generalization analysis unveils the importance of additionally reducing bias induced by low-rank approximation. Inspired by the variance-bias tradeoff in high dimensions from the theory, we introduce Sketchy Moment Matching (SkMM), a scalable data selection scheme with two stages. (i) First, the bias is controlled using gradient sketching that explores the finetuning parameter space for an informative low-dimensional subspace $\mathcal{S}$; (ii) then the variance is reduced over $\mathcal{S}$ via moment matching between the original and selected datasets. Theoretically, we show that gradient sketching is fast and provably accurate: selecting $n$ samples by reducing variance over $\mathcal{S}$ preserves the fast-rate generalization $O(\dim(\mathcal{S})/n)$, independent of the parameter dimension. Empirically, we concretize the variance-bias balance via synthetic experiments and demonstrate the effectiveness of SkMM for finetuning in real vision tasks.

The poor interfacial compatibility between sodium superionic conductor (NASICON) electrolyte and metallic sodium anode will lead to severe dendrite penetration, impeding the application of NASICON electrolytes for solid state sodium batteries. Herein, a homogeneous SnS2 coating layer is sputtered on the surface of Na3.4Zn0.1Zr1.9Si2.2P0.8O12 electrolyte to in situ construct a kinetically stable Na & horbar;Sn alloy/Na2S interlayer, possessing superior Na affinity, low diffusion barrier, and electronic insulating character to suppress dendrite growth, which is confirmed by experiments and density-functional theory calculations. Benefiting from the Na & horbar;Sn alloy/Na2S interphase, the critical current density of Na3.4Zn0.1Zr1.9Si2.2P0.8O12 increases from 2.4 to 9.4 mA cm(-2). In addition, the obtained Na3V2(PO4)(3)/Na3.4Zn0.1Zr1.9Si2.2P0.8O12@SnS2/Na solid state batteries exhibit a high initial reversible discharge capacity of 115.1 mAh g(-1) at 0.1 degrees C with an initial Coulombic efficiency of 93.6%, and a capacity retention rate of 88.1% after 1000 cycles at 1 C.

The challenge of delivering therapeutics to the central nervous system due to the restrictive nature of the blood-brain barrier (BBB) is a substantial hurdle in neuropharmacology. Our research introduces a breakthrough approach using microtubule-dependent transcytosis facilitated by novel aqueous compounds. We synthesized a series of red-emitting pyran nitrile derivatives. The molecular structure of compounds, photophysical properties, and water solubility were characterized. BBB permeability of BN1 was assessed in an in vitro BBB model. The transmembrane transport mechanism was next analyzed. The derivative was injected in the wild-type mouse for evaluation of brain penetration and biodistribution in the brain. We further investigated the potential of BN1-functionalized BBB-nonpenetrated silica nanoparticles for brain targeting. This compound demonstrated an ability to form endosomes within the phospholipid layer, thus enabling efficient penetration of the BBB via microtubule-mediated transcytosis, as evidenced in vitro model. This was further confirmed by in vivo experiments that BN1 displays the excellent BBB penetration and retained in brain parenchyma. Furthermore, BBB-impermeable mesoporous silica nanoparticle codelivery system markedly enhanced the transport efficiency to the brain in vivo by BN1-functionalized. These findings indicate that our designed aqueous molecules not only are capable of traversing the BBB but also serve as a viable new strategy for central-nervous-system-targeted drug delivery.

Constructing atom-pair engineering and improving the activity of metal single-atom nanozyme (SAzyme) is significant but challenging. Herein, we design the atom-pair engineering of Zn-SA/CNCl SAzyme by simultaneously constructing Zn-N4 sites as catalytic sites and Zn-N4Cl1 sites as catalytic regulator. The Zn-N4Cl1 catalytic regulators effectively boost the peroxidase-like activities of Zn-N4 catalytic sites, resulting in a 346-fold, 1496-fold, and 133-fold increase in the maximal reaction velocity, the catalytic constant and the catalytic efficiency, compared to Zn-SA/CN SAzyme without the Zn-N4Cl1 catalytic regulator. The Zn-SA/CNCl SAzyme with excellent peroxidase-like activity effectively inhibits tumor cell growth in vitro and in vivo. The density functional theory (DFT) calculations reveal that the Zn-N4Cl1 catalytic regulators facilitate the adsorption of *H2O2 and re-exposure of Zn-N4 catalytic sites, and thus improve the reaction rate. This work provides a rational and effective strategy for improving the peroxidase-like activity of metal SAzyme by atom-pair engineering. Designing and enhancing the performance of metal single-atom nanozymes (SAzymes) through atom-pair engineering is important yet difficult. Here the authors develop the atom-pair engineering of Zn-SA/CNCl SAzyme by concurrently creating Zn-N4 sites as catalytic sites and Zn-N4Cl1 sites as catalytic regulators.

Cyclometalated iridium(III) complexes have emerged as versatile candidates for cancer theranostics, offering integrated diagnostic imaging and potent singlet oxygen (1O2) generation for photodynamic therapy (PDT). However, their application has been limited by subdued photoluminescence, primarily due to intramolecular motion-induced excited energy dissipation. In this study, we address these limitations through the design and synthesis of five novel iridium(III) complexes: IrC2, IrC4, IrC6, IrC8, and IrC12. Our approach employs meticulous side-chain extending strategy to modulate side-chain length, thereby reducing intramolecular motion and significantly enhancing both one- and three-photon emissions and 1O2 production in the aggregated state. Detailed photophysical investigations, supported by crystallographic insights, reveal that side-chain elongation substantially amplifies these properties. Among the synthesized complexes, IrC8 stands out as a superior candidate for image-guided photodynamic therapy in cellular and 3D tumor spheroid models. This investigation pioneers the simultaneous enhancement of dual-photon emissions and PDT efficacy through a novel side-chain extension strategy in iridium(III) complexes, paving the way for their translational application in clinical theranostics.

Visualization of 1O2 in cells is of great significance for revealing the laws of life activities. Although many 1O2 fluorescent probes have been reported, their mechanism of capturing 1O2 has not been innovated. Herein, we report a ruthenium(II) bipyridyl complex [Ru(bpy)2L3]PF6 with the ability to specifically identify 1O2 based on a novel mechanism of action. The non-fluorescent [Ru(bpy)2L3]PF6 can be accurately oxidized by 1O2 to generate a fluorescent substance with a large Stokes shift (184 nm). [Ru(bpy)2L3]PF6 was successfully applied for visualizing the generation process of 1O2 in living cells and monitoring the morphological changes of mitochondria during photodamage. Therefore, [Ru(bpy)2L3]PF6 has potential to be a useful molecular probe for analyzing physiological processes related to 1O2 in organisms.

Sulfide solid electrolyte membranes employed in all-solid-state lithium batteries generally show high thickness and poor chemical stability, which limit the cell-level energy density and cycle life. In this work, Li9.88GeP1.96Sb0.04S11.88Cl0.12 solid electrolyte is synthesized with Sb, Cl partial substitution of P, S, possessing excellent toluene tolerance and stability to lithium. The formed SbS43- group in Li9.88GeP1.96Sb0.04S11.88Cl0.12 exhibits low adsorption energy and reactivity for toluene molecules, confirmed by first-principles density functional theory calculation. Using toluene as the solvent, ultrathin Li9.88GeP1.96Sb0.04S11.88Cl0.12 membranes with adjustable thicknesses can be well prepared by the wet coating method, and an 8 μm thick membrane exhibits an ionic conductivity of 1.9 mS cm-1 with ultrahigh ionic conductance of 1860 mS and ultralow areal resistance of 0.68 Ω cm-2 at 25 °C. The obtained LiCoO2|Li9.88GeP1.96Sb0.04S11.88Cl0.12 membrane|Li all-solid-state lithium battery shows an initial reversible capacity of 125.6 mAh g-1 with a capacity retention of 86.3% after 250 cycles at 0.1 C under 60 °C.

The nanosized sodium sulfide species greatly enhance the performance of Na-S batteries. However, as a typical ionic compound, the diminished stability with decreasing cluster size has been rarely considered. Numerous theoretical works only simulated the Na-S binary system based on quite small sodium sulfide clusters, with no account of the stability of the sub-nanoscale cluster. Here, by using an advanced structure search algorithm, we built a binary phase diagram of the sodium sulfide clusters (NSCs). The commonly studied monomer model with low sulfur concentrations is indeed energetically unfavorable, especially for the final discharge product, the Na2S monomer. Therefore, the aggregation of monomers should take place to form multimers. According to the energy, charge, and geometry, relatively stable clusters with low sulfur concentrations are located, including (Na2S5)2, (Na2S4)2, (Na2S3)3, (Na2S2)4, and (Na2S)6 clusters. Furthermore, these multimers bind more intensely to four typical models of carbon-based substrates. The B-doped carbon material exhibits outstanding affinity to NSCs, which may facilitate overcoming the shuttle effect in applications. This work represents a significant step toward understanding the evolution mechanism of NSCs that may guide the future development of high-performance Na-S batteries.

Nanoscale Zr-based metal-organic framework (MIP-202) particles were successfully prepared via a seeds-assisted hydrothermal secondary synthesis. The uniformly-shaped MIP-202 particles exhibit excellent performance on selective CO 2 adsorption from CO 2 /CH 4 and CO 2 /N 2 mixtures. At 298 K and 1 bar, the uptake ratios of CO 2 /CH 4 and CO 2 /N 2 of MIP-202 particles are as high as 32.6 and 65.2, respectively. The IAST (ideal adsorbed solution theory)-predicted selectivities of CO 2 /N 2 (50/50, v/v) and CO 2 /CH 4 (50/50, v/v) reach to 4.5 × 10 10 and 102.2, respectively. The breakthrough experiments further demonstrate that the CO 2 /CH 4 and CO 2 /N 2 mixtures can be efficiently separated through an adsorption column packed with MIP-202 particles. In addition, the as-prepared MIP-202 particles had a low iso-enthalpy of adsorption of CO 2 of 32.95 kJ/mol, which is favorable for the regeneration of the adsorbent. Herein the as-prepared MIP-202 is a potential material for efficient separation of CO 2 from CH 4 or N 2 through an energy- and cost-saving CO 2 capture process.

Pretrained language models have shown success in various areas of natural language processing, including reading comprehension tasks.However, when applying machine learning methods to new domains, labeled data may not always be available.To address this, we use supervised pretraining on source-domain data to reduce sample complexity on domainspecific downstream tasks.We evaluate zeroshot performance on domain-specific reading comprehension tasks by combining task transfer with domain adaptation to fine-tune a pretrained model with no labelled data from the target task.Our approach outperforms Domain-Adaptive Pretraining on downstream domainspecific reading comprehension tasks in 3 out of 4 domains.

Tumour stem cells has been shown to be highly correlated with tumour occurrence, metastasis, recurrence, and drug resistance. However, current clinical methods for the identification and specific treatment of tumour stem cells are limited, complicated and ineffective. Herein, we rationally developed a novel cyclometalated iridium (III) complex, having free rotational pyridine units, with the capability in generating singlet oxygen in living cells as well as showing sensitive fluorescence lifetime response towards in situ microenvironment. In addition, we have successfully demonstrated that it can be used to distinguish tumour stem cells from tumour tissue and self-report the anti-tumour effect via fluorescence lifetime imaging microscopy (FLIM).

With the rapid development of deep learning, training Big Models (BMs) for multiple downstream tasks becomes a popular paradigm. Researchers have achieved various outcomes in the construction of BMs and the BM application in many fields. At present, there is a lack of research work that sorts out the overall progress of BMs and guides the follow-up research. In this paper, we cover not only the BM technologies themselves but also the prerequisites for BM training and applications with BMs, dividing the BM review into four parts: Resource, Models, Key Technologies and Application. We introduce 16 specific BM-related topics in those four parts, they are Data, Knowledge, Computing System, Parallel Training System, Language Model, Vision Model, Multi-modal Model, Theory&Interpretability, Commonsense Reasoning, Reliability&Security, Governance, Evaluation, Machine Translation, Text Generation, Dialogue and Protein Research. In each topic, we summarize clearly the current studies and propose some future research directions. At the end of this paper, we conclude the further development of BMs in a more general view.

In situ implanting MnO fine nanoparticles into carbon nanorod-assembled microspheres enables improved electrode stability and electrochemical performance via structural and compositional synergy.

The term `spurious correlations' has been used in NLP to informally denote any undesirable feature-label correlations. However, a correlation can be undesirable because (i) the feature is irrelevant to the label (e.g. punctuation in a review), or (ii) the feature's effect on the label depends on the context (e.g. negation words in a review), which is ubiquitous in language tasks. In case (i), we want the model to be invariant to the feature, which is neither necessary nor sufficient for prediction. But in case (ii), even an ideal model (e.g. humans) must rely on the feature, since it is necessary (but not sufficient) for prediction. Therefore, a more fine-grained treatment of spurious features is needed to specify the desired model behavior. We formalize this distinction using a causal model and probabilities of necessity and sufficiency, which delineates the causal relations between a feature and a label. We then show that this distinction helps explain results of existing debiasing methods on different spurious features, and demystifies surprising results such as the encoding of spurious features in model representations after debiasing.