Multi-modal classification (MMC) leverages effective fusion of information from diverse modalities to achieve superior classification performance. Existing fusion methods, however, rely on expert-designed architectures that demand substantial domain expertise and computational resources, with fixed topologies that offer limited structural flexibility across tasks and datasets. Although neural architecture search (NAS)-based fusion methods have been proposed to automatically discover high-performing architectures, these approaches are computationally expensive and predominantly rely on pairwise modality combinations, which fail to capture complex multi-variable correlations, limiting the architectures' expressiveness and flexibility. Consequently, there remains a lack of multi-modal fusion frameworks that can simultaneously achieve high efficiency and high accuracy. To break through these bottlenecks, we propose a multi-branch tree-based fusion neural architecture search framework (MBTF-NAS). For performance enhancement, MBTF-NAS employs a multi-branch tree-structured encoding strategy that enables dynamic and computationally efficient exploration of fusion topologies and substantially strengthens cross-modal interaction. A learnable model-level attention weighting mechanism further emphasizes informative modalities, improving the overall quality of multi-modal feature fusion. For efficiency improvement, MBTF-NAS leverages zero-cost proxy metrics for architecture evaluation, enabling rapid identification of high-potential candidates while dramatically reducing computational overhead. We conducted a comprehensive evaluation of MBTF-NAS on seven representative multi-modal benchmarks. The experimental results demonstrate that MBTF-NAS consistently outperforms state-of-the-art approaches, highlighting its effectiveness and generalizability.
Self-healing polymers governed by supramolecular and dynamic covalent interactions are redefining structural design and functional integration in contemporary materials. Reversible supramolecular motifs, including hydrogen bonding and electrostatic interactions, enable rapid interfacial reconstruction and damage tolerance, whereas dynamic covalent bonds, such as Diels-Alder, disulfides, provide mechanically robust, reconfigurable, and recyclable network architectures. Their synergistic integration yields materials with repeatable healing, robust mechanics, and multimodal responsiveness. These advances have accelerated progress in flexible electronics, particularly capacitive sensing, where dedicated multilayer architectures, printable conductors, and minimal conductive dopants deliver enhanced sensitivity, a broad operational window, and rapid electrical and mechanical recovery; however, challenges such as scalable processing and property reconciliation remain unresolved. This review highlights recent advances in multifunctional dynamic polymer networks by delineating supramolecular and dynamic covalent self-healing behaviors, their synergistic coupling, and their deployment in state-of-the-art self-healing sensor devices, providing insights for high-performance flexible sensing technologies.
The aims of this study were to investigate gastrocnemius medialis (GM) muscle properties in male academy soccer players (ASP) and age- and sex-matched control participants (CON); and to explore the relationships between GM characteristics and jump performance. Thirty-four participants (ASP, n = 22, age 18.8 ± 1.4 years, height 1.82 ± 0.08 m, mass 75.1 ± 5.9 kg; and CON, n = 12, 22.2 ± 2.9 years, 1.75 ± 0.05 m, 71.6 ± 7.4 kg) completed the following assessments: ultrasound measurements of GM anatomical cross-sectional area (ACSA), volume, muscle thickness (MT), fascicle pennation angle (θp) and fascicle length (Lf); isokinetic dynamometry measurements of isometric plantar flexion and dorsiflexion maximum voluntary torque; and unilateral and bilateral, vertical and horizontal, countermovement jumps (CMJ), and bilateral drop jumps on a force platform. θp (17.4° ± 2.5° vs. 14.3° ± 1.2°, P < 0.001); unilateral horizontal CMJ peak power (30.14 ± 3.53 vs. 23.18 ± 3.72 W kg- 1); and projectile range during unilateral (104 ± 16 vs. 89 ± 12 cm, P = 0.006) and bilateral (140 ± 14 vs. 129 ± 14 cm, P = 0.041) horizontal CMJ were greater in ASP vs. CON. In ASP alone, Lf correlated inversely with vertical CMJ performance but positively with horizontal CMJ performance (R2 ≥ 0.200, P ≤ 0.042). Conversely, θp correlated positively with vertical CMJ performance but inversely with horizontal CMJ performance (R2 ≥ 0.194, P ≤ 0.044). In CON only, ACSA, MT, volume and Lf all correlated inversely with vertical CMJ performance (R2 ≥ 0.366, P ≤ 0.037). The opposing θp and Lf correlations with vertical and horizontal CMJ jump performance in ASP suggest GM architecture influences CMJ performance in a direction-specific manner in this population, while the different correlation patterns between ASP and CON suggest that GM architecture contributes to CMJ performance differently in these two populations.
We report a rational design of donor-acceptor-donor (D-A-D) Zn(II)porphyrin trimers adopting two conformational architectures, ABA ("closed") and ACA ("open"), in which electron-rich terminal porphyrins flank an electron-deficient central unit linked by a flexible spacer. Minor variations in peripheral substitution at the central porphyrin trigger substantial structural reorganization, leading to pronounced differences in electronic coupling and excited-state behavior. The "closed" ABA architecture enforces close interchromophoric proximity, resulting in strong excitonic coupling and perturbation of the Soret band. In contrast, the "open" ACA conformer exhibits weaker coupling and minimal spectral distortion. As a result, ACA displays fluorescence resembling a linear superposition of its constituent monomers and a relatively long excited-state lifetime (∼3 ns). By comparison, ABA deviates significantly from monomeric behavior, consistent with enhanced intramolecular charge-transfer (ICT) interactions that shorten the fluorescence lifetime (∼800 ps) and suppress the fluorescence quantum yield. The more efficient ICT in ABA facilitates triplet-state formation, resulting in substantially higher singlet oxygen quantum yield than ACA. Accordingly, ABA exhibits superior photocatalytic activity, achieving quantitative oxidation of triphenylphosphine, whereas ACA shows markedly lower efficiency. Collectively, these results demonstrate that conformational modulation of excitonic and charge-transfer interactions provides an effective strategy for selective photooxidation.
Spiking Neural Networks (SNNs) have emerged as a promising paradigm for brain-inspired edge computing. Leveraging binary spikes and local learning rules, SNNs enable energy-efficient on-chip learning and rapid adaptation to changing environments, which is crucial for edge AI that needs to learn continuously from new data. However, many SNN processors enabling on-chip learning for edge computing confront a trade-off: small-scale task-specific designs offer low power but poor multi-task inference accuracy, while large-scale general-purpose designs achieve high multi-task accuracy at the cost of large memory and poor energy efficiency. To overcome this challenge, this paper presents ANP-R, a 22nm asynchronous SNN-based edge AI processor with coarse-grained reconfigurable architecture enabling one-shot, few-shot, batch and incremental on-chip learning. The processor integrates 64 cores containing 4096 neurons and 0.262 million synapses. Two key features are proposed: 1) An asynchronous coarse-grained reconfigurable architecture that supports various STDP-based SNN topologies. These topologies enable over 95% average accuracy across four sensory tasks; 2) an energy-efficient asynchronous training method incorporating a self-adaptive synaptic weight update mechanism reducing up to 65% redundant updates without accuracy loss, and a trained weights low-bit width coding method reducing up to 50% storage cost with 0.3% accuracy loss. Measurement results demonstrate 92.1% accuracy for hand gesture classification, 93.9% for keyword spotting, 98.6% for object recognition and 99.2% for gas identification. Compared with state-of-the-art SNN-based chips, this work achieves up to 6.02x, 8.61x and 7.1% improvement in energy efficiency, energy per step, and accuracy, respectively.
Designing thermoelectric materials that combine high conversion efficiency with mechanical robustness remains challenging-especially in metavalent-bonded chalcogenides, where weak bonds yield intrinsically low lattice thermal conductivity yet compromise mechanical integrity. Here we present an entropy-enabled defect architecture in SnTe-based alloys that steers hierarchical defect evolution-from 0D substitutional clusters to 1D dislocations and 3D coherent nanoprecipitates-enabling multiscale regulation of phonon transport and strengthening mechanisms. Broadband phonon scattering depresses lattice thermal conductivity to 0.26 W·m-1·K-1 at 873 K, while coherent (Cd,Ge)Se nanoprecipitates and dislocation networks establish effective load-transfer and pinning pathways, elevating the yield strength to 220 MPa, an improvement of ∼100 MPa (≈83%) relative to pristine SnTe (120 MPa), while retaining reasonable plasticity. In parallel, modest band-structure optimization through compositionally complex alloying within the entropy-stabilized matrix improves the power factor. Benefiting from these synergies, the optimized composition Sn0.91Cd0.03Sb0.09Te(GeSe)0.25 delivers a peak figure of merit of 1.7 and device efficiencies of 7.2% (single-leg) and 5.7% (multi-leg). This work establishes a generalizable pathway to strong, efficient thermoelectric materials, particularly applicable to metavalent bonding systems.
Sesame (Sesamum indicum L.) is a major oilseed crop with numerous oil and nutritional benefits. However, the genetic basis of complex agronomic traits remains fragmented across individual quantitative trait loci (QTL) studies. Meta-QTL (MQTL) analysis provides a robust framework for identifying stable genomic regions that control quantitatively inherited traits in diverse environments and genetic backgrounds. Here, we present the first comprehensive MQTL analysis of sesame, targeting morphological and yield components, oil content and quality, and seed and capsule traits. All published QTLs were compiled using LOD scores, phenotypic variance explained (PVE), and confidence intervals (CI), and projected onto a high-density consensus map comprising 38,972 markers using BioMercator v4.2. In total, 321 QTLs (54.9%) were effectively projected and summarized into 92 MQTLs. The average CI of MQTLs was 4.01 cM, indicating a 2.41-fold (58.57%) reduction compared with the initial QTLs (9.68 cM). Nineteen high-confidence MQTLs (CI ≤ 5 cM and ≥ 4 initial QTLs) were selected for candidate gene mining, which collectively encompassed 1,678 unique gene models. Using orthology-based prioritization, we identified 160 orthologous candidates, and functional annotation revealed 43 genes that were strongly associated with key traits within the MQTL regions. Several MQTLs co-localize with marker-trait associations reported in previous genome-wide association studies, thereby reinforcing their significance in the regulation of traits. These findings indicate that MQTL analysis substantially improves the mapping precision and provides reliable genomic targets for sesame breeding. Integrating tightly linked markers from these MQTLs into marker-assisted and genomic selection schemes offers a powerful strategy to accelerate the genetic improvement of sesame in terms of yield, plant architecture, seed, and oil quality.
For decades, eukaryotic circadian timing has been framed mainly through nuclear transcription-translation feedback loops (TTFLs). Here, we synthesize evidence supporting a broader organelle-centered model in which cellular time emerges from dynamic coupling between TTFL clocks, post-translational feedback loop (PTFL) oscillators, and entrained rhythmic modules across mitochondria, endoplasmic reticulum, lysosomes, peroxisomes, Golgi apparatus, plasma membrane, and cytoskeleton. Metabolic flux, redox cycling, proteostasis, ion handling, membrane excitability, trafficking, and mechanotransduction act as temporal currencies that either sustain selected transcription-independent rhythms or transmit phase information within a TTFL-coordinated network. In this layered architecture, the TTFL remains a central integrator that stabilizes inter-organelle phase relationships, aligns intracellular rhythms with environmental Zeitgebers, and links biochemical state to epigenetic and RNA-based regulation. We propose that circadian dysfunction reflects progressive intracellular desynchronization rather than isolated clock-gene failure, opening diagnostic and therapeutic opportunities aimed at restoring subcellular temporal coherence.
Large-volume autologous fat grafting for aesthetic augmentation is often complicated by cysts and fibrosis, suggesting that its survival mechanisms differ from those of small-volume grafts. To investigate the survival biology of large-volume fat grafts and evaluate the effects of mechanical pressure on tissue structure, viability, and long-term graft outcome. A 3-dimensional in vitro model using human adipose tissue in a scaffold was established. Constructs were cultured under 0, 6 or 12 mmHg of pressure to assess structural integrity, solute permeability, cell viability and molecular changes. The viable outer layer was then transplanted into nude mice to evaluate volume retention, necrosis and fibrosis. In vitro, a porous interstitial network supported tissue survival to a depth of approximately 8 mm. Pressure dose-dependently disrupted this network, reducing solute permeability from 76.4% to 21.3% and impairing viability. Pressure also induced a form of "latent injury," consistent with YAP-related mechanotransductive signaling, lineage-related molecular changes and metabolic stress. In vivo, although gross volume retention was similar among groups, pressure-conditioned grafts developed marked necrosis, with fibrosis increasing from 22.1% in controls to 53.3% in the 12 mmHg group at 8 weeks. Large-volume fat graft survival depends in part on the integrity of a pressure-sensitive porous network. Mechanical pressure may induce a clinically relevant "latent injury," consistent with molecular changes observed in vitro, which may contribute to later fibrosis. Minimizing mechanical stress during harvest, processing, handling and implantation may help preserve graft integrity and improve long-term outcomes.
Colloidal nanocrystals are generally regarded as rigid solid entities, rarely exhibiting the structural adaptability observed in molecular cages, such as fullerenes, which can undergo carbon framework reduction and encapsulate guest cations without structural reorganization. Here, by creating two enantiomeric pairs of high-nuclearity copper sulfide nanoclusters with a mixed-valence Cu(II)/Cu(I) configuration, we endow these nanoscale assemblies with an intrinsic capacity for electron uptake under mild reducing conditions. The resulting charge imbalance provides an effective thermodynamic driving force that realizes a positively charged metal ion migrating inward through multiple atomic layers and occupying the cluster core. This system thus represents a rare example of a nanocluster platform that simultaneously combines reduction tolerance and structural robustness, preserving its atomic framework despite the incorporation of a single atom effectively modifying the electronic structure, particularly the local chirality. In situ absorption and circular dichroism spectroscopies establish that the transformation proceeds through a continuous, single-particle process rather than a fragmentation-reconstruction pathway, while ex situ pair distribution function analysis resolves key local steps in the structural evolution, offering mechanistic insights into this unique migration behavior.
Global analyses of leaf size suggest that large leaves predominate in favourably warm and wet climates, and small leaves occur at climatic extremes. However, these patterns are dominated by data from angiosperms and may obscure drivers of leaf size variation in older, less speciose groups, such as conifers. Here we employ a novel modelling framework, multi-response phylogenetic mixed models (MRPMM), that identifies trait correlations at both phylogenetic and phylogenetically independent levels to investigate how climate influences leaf size evolution across conifers. We show overall patterns for all conifers and focus on three groups with distinctive leaf architectures: scale-leaved Cupressaceae, needle-leaved Pinus and broad-leaved Podocarpus. We found moderate to strong phylogenetic signal in conifer leaf and climate niche traits. Phylogenetic relationships explained most of the association between leaf size and climate, with temperature revealed as a stronger driver than dry season precipitation, indicating deep-time co-evolutionary associations. These patterns reflect trade-offs associated with leaf hydraulic architecture under contrasting selection pressures. In single-veined leaves, lateral water transport and thus leaf width is constrained, yet this narrow leaf form can be advantageous for survival under climatic extremes such as drought and freezing. In contrast, anatomical innovations (like accessory transfusion tissue in Podocarpus) allow for broader leaves which are favourable in competitive environments, while also potentially making leaves more vulnerable to climate extremes. Our results support previous evidence for phylogenetic niche conservatism in conifers, where species tend to track their ancestral climatic preferences rather than adapting to new environments. This conservatism, likely controlled by leaf hydraulic architecture, results in strong evolutionary constraints on current bioclimatic distributions and potential responses to changing climates in conifers. This study also highlights the importance of considering phylogenetic impacts on functional trait evolution, especially in evolutionarily conservative groups like conifers.
Facile and quantitative detection of liquid biopsy biomarkers such as microRNAs offers significant potential for precision healthcare; however, conventional biosensing methods rely on enzyme- or label-based workflows that are costly, time-consuming, and labor intensive. Microwave biosensors, particularly split-ring resonators (SRRs), offer an attractive alternative as they enable label-free, noncontact electromagnetic detection through permittivity measurements and are compatible with printed-circuit-board manufacturing. However, the sensitivity of conventional SRR platforms remains insufficient for clinically relevant biomarker detection. Here, we introduce an enzyme-free, label-free microwave biosensing architecture that integrates SRRs with microfluidic channels containing localized bioreceptor-functionalized hydrogel micropillars. Target hybridization within the hydrogel micropillars induces localized changes in complex permittivity, which are transduced into concentration-dependent shifts in the resonant frequency of the SRR capacitive gap. As a proof of concept, the platform is applied to detect the cancer-associated biomarker miR-16-5p using peptide nucleic acid (PNA) probes, which were selected for their neutral backbone, enzymatic stability, and strong hybridization affinity. The hydrogel micropillars act as three-dimensional scaffolds that enhance probe loading and maximize volumetric electromagnetic interaction, representing a departure from conventional planar biointerfaces. Compared with equivalent planar systems, this architecture achieves approximately a 20-fold improvement in detection limit, reaching subnanomolar sensitivity without any amplification or labeling while maintaining single-nucleotide specificity and strong device reproducibility. Beyond being the first demonstration of SRR-based miRNA detection, this work establishes a general strategy for three-dimensional microwave biosensing and positions hydrogel-interfaced resonators as a next-generation platform for sensitive, selective, label-free, and reusable biosensors.
Architected materials derive functionality from geometry, yet conventional unit cell-based design limits functional heterogeneity, geometric adaptability, and robustness to defects. Inspired by natural morphogenesis, we introduce RDGenCAD, a morphogenetic design framework that translates programmable growth rules into reaction-diffusion dynamics to generate self-organized, CAD-ready architectures. A database of 120,000 morphogenetic structures reveals statistically deterministic and continuous tunability of elastic properties across auxetic and conventional regimes, despite pronounced geometric irregularity. These architectures further exhibit emergent flaw insensitivity and crack deflection through stress compartmentalization, leading to synergistic gains in strength and toughness relative to regular lattices. By shifting architected material design from unit-cell tessellation to programmable morphogenetic growth, this work establishes self-organization as a generative principle for designing materials that are irregular yet predictable, heterogeneous yet "coherent," and directly manufacturable.
Background Monoclonal antibodies (mAbs) and their derivatives represent a central pillar of contemporary oncology, with expanding complexity in molecular design and clinical application. Beyond unconjugated mAbs, bispecific, trispecific (bsAbs/tsAbs) and antibody-drug conjugates (ADCs) introduce additional pharmacological dimensions that directly impact efficacy, toxicity, and resistance. BsAbs were developed as pharmacological matchmakers to simultaneously engage tumor antigens and immune receptors, physically bridging effectors and malignant cells to promote immunological synapse formation. TsAbs extend this concept by enabling coordinated engagement of three targets. ADCs exploit mAb specificity to deliver highly potent, otherwise intolerable cytotoxic payloads directly into tumor cells, expanding the therapeutic window in hematologic and solid tumors. Summary Clinical performance is tightly linked to molecular architecture: bsAbs/tsAbs, activity depends on valency, target geometry, and Fc configuration, whereas ADC efficacy reflects a tripartite pharmacology integrating target engagement, intracellular payload release, and bystander cytotoxicity. Despite their remarkable potency, antibody-based platforms remain vulnerable to resistance, which may arise through target-dependent mechanisms, including antigen downregulation and epitope masking, or target-independent processes, such as altered intracellular trafficking, lysosomal dysfunction, payload efflux, and adaptive survival signaling. Key message A comprehensive understanding of antibody architecture, target biology, pharmacokinetic/pharmacodynamic (PK/PD) behavior, toxicity profiles, and resistance mechanisms is essential to optimize treatment selection and sequencing. Collectively, mAbs, bsAbs/tsAbs, and ADCs provide the mechanistic and technological basis for next-generation platforms, including peptide-drug conjugates and antibody-radionuclide conjugates.
Intelligent responsive materials are important components for molecular machines and memory devices. However, the mechanism of guest-induced chirality inversion remains elusive because guest binding is typically too fast to resolve the process. To address this challenge, it is necessary to develop a system in which chirality inversion and guest uptake occur on comparable, slow time scales. Here, we report a triple-helical closed-cage cobalt(III) metallocryptand incorporating three bridging 1,7-heptanediamine (hpda) ligands, which creates a closed-cage architecture that significantly slows guest uptake/release. X-ray crystallography revealed the formation of a (P,R6) diastereomer with a right-handed triple-helical structure. In solution, a dynamic equilibrium between the (P,R6) and (M,R6) diastereomers was observed with slow interconversion (t1/2(app) = 20 min). Upon addition of CsCl, the P/M chirality was gradually inverted from a P-abundant state to an M-abundant state over several hours, associated with slow Cs+ uptake. The closed-cage design with bridging hpda ligands significantly slowed both Cs+ uptake and P/M interconversion, allowing the two processes to proceed on comparable, slow time scales. Kinetic analysis based on a four-species reversible model revealed that Cs+ is preferentially taken up in the less abundant M form, whereas P form does not directly bind Cs+ but instead contributes to uptake after P→M isomerization. The M form binds Cs+ more strongly than the P form, thus driving the P→M chirality inversion through this M-form-mediated pathway. In contrast, addition of Cl- shifted the equilibrium toward the more P-favored mixture by interacting with Cl- at the peripheral binding pocket, highlighting its opposite stereodynamic effect relative to Cs+. These findings demonstrate that cage closure modulates both the timing and sequence of events during the pathway of guest-induced chirality inversion, and provide a kinetic platform for probing guest effects on stereodynamic equilibria. We also anticipate that the strategies used here can be applied to the rational design of other smart molecular architectures.
This work extends a closed-loop extracorporeal vascular cleaning architecture, introduced in a prior publication, with four refinements that address practical and safety considerations not handled in the prior protocol. First, atheroma is hypothesized to consist of two mechanically distinct fractions: a weakly adhered fraction removable by gentle mechanical contact within the isolated chamber, and a structurally integrated fraction requiring chemical action. Mechanical pre-cleaning with a pointed balloon under inline pressure feedback addresses the weakly adhered fraction first; chemistry then operates against an exposed substrate. The two-regime model bounds the residual chemistry burden at the structurally integrated mass fraction, hypothesized at 40%-60% of total plaque mass and corresponding to an upper bound of 60% of the chemistry-only collagenase specification; exposed-substrate kinetics may permit further reduction, though the magnitude of any such enhancement is undefended at the concept stage and must be established empirically. Second, the pressure differential between the bypass and treatment circuits is made stage-dependent: inward-biased during chemistry and outward-biased during saline phases, with a bounded outward sweep during the final flush. Third, direct percutaneous needle access through the overlying tissue extends the architecture to deep vessels (renal, mesenteric, and retroperitoneal) currently inaccessible to catheter-based intervention. Fourth, a localized biodegradable protective coating applied to the cleaned vessel wall before blood reintroduction bridges the re-endothelialization window with local protection rather than systemic antithrombotic therapy. Each refinement uses approved or near-approved components in new combinations. Two architectural extensions are also developed: a hemodynamic synergy with systemic therapy in which cleaned segments serve as preferential redeposition sites for material mobilized from inaccessible regions, and an organ as chamber primitive generalizing topological isolation to organs with discrete arterial inflow and venous outflow. Five testable predictions are specified for validation in a porcine atherosclerosis model.
Dural arteriovenous fistula (DAVF) in the sphenoid region is a rare condition. It develops following aberrant vascular connections between arterial branches of the external and internal carotid arteries with sphenoidal venous drainage. The treatment can be endovascular embolization or open surgery to obliterate the fistula. A 62-year-old man was incidentally diagnosed with a DAVF located at the left lesser sphenoid wing during a routine checkup. Preoperative angiographic imaging revealed a high-grade fistula with arterial feeders originating from both the internal and external carotid arteries and venous drainage into the superficial middle cerebral vein (SMCV). Based on the angioarchitecture of the DAVF, surgery was performed, successfully obliterating the arterial feeders and disconnecting the fistula. The authors present an asymptomatic case of a left lesser sphenoid wing DAVF with an atypical venous drainage pattern involving the SMCV, sparing the cavernous sinus. Recognition of the vascular architecture and the high-grade nature of this vascular malformation is crucial for ensuring proper treatment and preventing serious complications. https://thejns.org/doi/10.3171/CASE2628.
COVID-19 is a highly contagious disease transmitted primarily through human contact. Therefore, understanding population mobility is essential for predicting COVID-19 case trends. In this paper, we propose a novel deep learning approach for forecasting new COVID-19 cases using a neural architecture called Neural Basis Expansion Analysis for Interpretable Time Series (N-BEATS). The N-BEATS model effectively handles long input sequences and large output horizons without information loss or increased computational complexity. We compare the performance of N-BEATS with a state-of-the-art benchmark model, LSTM-Markov, across four major countries: the United States, the United Kingdom, Russia, and Brazil. Three distinct COVID-19 datasets from Google, Apple, and Our World in Data (OWID) were used in this study. Incorporating Google and Apple mobility data as covariates enhances both the accuracy and interpretability of the N-BEATS model. Our results show that N-BEATS consistently outperforms LSTM-Markov across all datasets and countries, consistently yielding lower Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE). Furthermore, the N-BEATS model with covariates outperforms its counterpart without covariates, indicating that mobility data provide substantial value for forecasting new COVID-19 cases. Overall, this study demonstrates the effectiveness of the N-BEATS architecture in capturing pandemic dynamics and offers valuable insights for policymakers and public health officials in managing future outbreaks.
Sickle cell disease (SCD) is a major global health burden, and early, accurate diagnosis is critical for effective management. Conventional diagnostic methods are often resource-intensive and inaccessible in high-burden, low-resource settings. Artificial intelligence (AI) and machine learning (ML) technologies have emerged as promising tools to automate and enhance SCD detection. This systematic review aimed to critically evaluate the diagnostic and predictive performance of AI and ML models for SCD detection and to assess their methodological quality and readiness for clinical implementation. A systematic search of PubMed, Web of Science, Scopus, and Embase was conducted for studies published between 2021 and 2025, following Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. Original research employing AI/ML models for SCD detection, classification, severity stratification, or outcome prediction was included. Data on study characteristics, model types, and diagnostic performance metrics were extracted. The risk of bias was assessed using the Prediction Model Risk of Bias Assessment Tool (PROBAST). A narrative synthesis was performed due to substantial methodological heterogeneity precluding meta-analysis. Seventeen studies were included, demonstrating a diverse landscape of model architectures, including deep learning (DL) for blood smear image analysis, ensemble methods for classification, and prognostic models for pain and mortality prediction. Diagnostic performance was consistently high, with accuracies frequently exceeding 94% for image-based SCD detection and area under the receiver operating characteristic curve (AUC-ROC) values reaching up to 0.99 for ensemble classifiers. Prognostic models for mortality and readmission achieved C-indices and AUCs of 0.76 and 0.77, respectively. PROBAST assessment revealed that a majority of studies (14 of 17) had a low overall risk of bias, while three studies were rated as high risk due to small sample sizes and methodological reporting limitations. AI and ML models demonstrate substantial diagnostic accuracy and promising prognostic capability in SCD. However, the field remains at a proof-of-concept stage, with a predominant reliance on internal validation and a lack of standardized reporting that hinders direct model comparison. For these technologies to achieve clinical impact, a rigorous paradigm shift toward prospective, externally validated studies in high-burden populations, alongside strict adherence to emerging reporting standards, is essential.
Zinc-bromine (Zn-Br) flow batteries are promising for grid-scale energy storage due to their high safety, low cost, and scalable architecture. However, their application remains constrained by cathode-side polybromide shuttle and anode-side Zn dendrite formation and hydrogen evolution reactions (HER). Here, we propose a bidomain engineering strategy that employs acetylcholine (ACh+) as a dual-functional electrolyte additive to simultaneously address the challenges of both sides. On the cathode side, the quaternary ammonium group of ACh+ complexes with polybromides upon charging to increase their molecular size, thereby effectively inhibiting the polybromide shuttle. On the anode side, the acetyl group of ACh+ rapidly absorbs onto the Zn surface to form a water-depleted interface, inducing uniform Zn plating/stripping with suppressed HER. Consequently, the cycling life of Zn-Br flow batteries with a single ACh+ additive is extended by nearly 80-fold, from 80 cycles to over 6400 cycles, demonstrating highly durable cycling stability, together with an outstanding cumulative plating capacity of 128 Ah cm-2. This finding demonstrates that dual-function electrolyte design provides a viable pathway for grid-scale application of high-rate and long-life Zn-Br flow batteries.