Urban river barriers reshape transport networks, spatial interaction, and accessibility disparities by concentrating cross-river mobility on limited bridges, tunnels, and transit corridors. Taking Wuhan, China, as a representative riverine city, this study examines the modal accessibility gap between public transport and car travel in cross-river travel. Using mobile phone signaling OD data and multimodal route-planning travel-time data, we develop a demand-weighted accessibility framework at the 1 km grid scale to compare how the two modes cover the same observed cross-river motorized OD demand benchmark within specified time thresholds. A Cross-river Modal Accessibility Gap (MAG) index is constructed using mode-specific travel-time threshold sets, and Random Forest and SHAP are used to examine nonlinear associations between spatial factors and MAG. Results show that car travel provides higher and more spatially even demand coverage, particularly in the off-peak period, but this advantage weakens under morning- and evening-peak routing conditions. Public transport provides lower overall coverage and greater spatial heterogeneity, while rail-supported corridors show smaller temporal deviations and localized stability. Distance to the Yangtze River, proximity to direct cross-river metro services, total walking time, transfer burden, and station coverage are among the important spatial and transport factors associated with MAG. RF predictive performance remains substantial under spatial-block validation but decreases as the spatial holdout scale increases. These findings suggest that urban rivers are not only physical barriers but also an important spatial context for spatial segmentation, differentiating accessibility across transport modes and urban areas.
Saccharomyces boulardii (S. boulardii) is the only yeast probiotic widely used in clinical practice and consumer products. It combines gastrointestinal robustness, intrinsic antibiotic tolerance and eukaryotic protein-processing capacity, making it an attractive chassis for engineered probiotics, in situ biomanufacturing and live biotherapeutic delivery. However, host-specific characterization of regulatory elements, candidate genomic integration sites, and multicopy integration strategies remains limited in the clinically used S. boulardii CNCM I-745 background. Here, using S. boulardii CNCM I-745 as the host, we developed and characterized a genetic engineering toolkit that integrates transcriptional control elements with genome-integration strategies. We first quantified the activities of 20 endogenous constitutive promoters, two inducible promoters and six terminators. We then established a markerless CRISPR-Cas9 workflow and evaluated 10 chromosomal integration loci; six of which showed Editing efficiency above 80% while supporting comparable transgene expression without detectable growth penalties. Finally, we constructed an rDNA-repeat-based multicopy integration system that increased transgene expression by 9.8- to 22.6-fold relative to single-copy integration and revealed a significant positive but non-proportional association between estimated copy number and reporter output. Together, this toolkit provides S. boulardii with a set of tunable promoters and terminators, candidate chromosomal integration sites compatible with a markerless CRISPR-Cas9 workflow. These resources expand the genetic accessibility of this probiotic yeast and provide a foundation for developing engineered S. boulardii strains with enhanced probiotic, therapeutic and biomanufacturing functions.
Actuator-rate-sensitive applications require feedback laws that discourage abrupt changes in the control input. For unknown discrete-time linear systems, this paper develops an input-increment-aware off-policy Q-learning algorithm that learns the optimal rate-aware feedback from input-state data. The previous input is treated as part of an augmented state and the input increment as the new action. This restores Markovianity but yields a linear quadratic problem with a persistent state-action cross term, so the usual cross-term-free linear-quadratic-regulator Q-learning update is not directly applicable. We establish structural validity of the augmented problem, derive a cross-coupled off-policy regression and policy-improvement rule, and prove admissibility preservation, monotone value improvement, convergence to the optimal incremental gain, and joint convergence of the plant state and the control signal. The increment weight is further shown to provide an a-posteriori rate certificate and a certified structural lower bound on the conditioning of the policy-improvement step under regression error. Numerical tests on academic plants and a quarter-car active suspension benchmark confirm the predicted smoothing and robustness effects, and benchmark the learned closed loop against a rate-constrained model predictive control oracle that has full plant knowledge and road-disturbance preview.
Metal-organic frameworks (MOFs) and their derivatives have emerged as promising photocatalysts for CO2 reduction owing to their large surface area, tunable porosity, and adjustable coordination environment. However, a comprehensive understanding of how organic ligands govern the structure-function relationship in MOF-based photocatalysis remains limited. This review provides a ligand-centered perspective on the photocatalytic CO2 reduction (PCR) process, focusing on how ligand functionalization, electronic modulation, and coordination design influence catalytic activity, stability, and selectivity. We summarize recent progress in ligand-regulated MOF systems from both experimental and theoretical viewpoints and discuss the associated structure-activity mechanisms. Finally, current challenges and future opportunities—such as cost-effective synthesis, scalability, and sustainable catalyst design—are discussed to guide the rational development of high-performance MOF photocatalysts for efficient CO2 conversion.
Seismic exploration plays a pivotal role in subsurface characterization and has been greatly promoted by artificial intelligence techniques such as deep learning, yet its reliance on high-quality labeled data poses a significant challenge for conventional supervised learning methods. In recent years, label-efficient deep learning has emerged as a powerful solution to this bottleneck by leveraging unsupervised, self-supervised, semi-supervised, and weakly supervised paradigms. This survey provides a comprehensive overview of these learning paradigms and their growing impact on key stages of the seismic data processing workflow, including preprocessing, imaging and inversion, and geological interpretation. We first outline the theoretical foundations and representative architectures of each paradigm, highlighting their distinct supervision strategies and learning objectives. We then analyze recent advancements in applying label-efficient methods to tasks such as seismic denoising, interpolation, full waveform inversion, facies classification, fault detection, and salt body delineation. By systematically comparing methodologies across application scenarios, we identify their respective advantages, limitations, and domain-specific adaptations. Finally, we discuss the main challenges hindering large-scale deployment, including the lack of standardized benchmarks, difficulty in integrating geophysical constraints, and the interpretability gap, and we also suggest promising directions for future research. This review aims to serve as a comprehensive reference for geoscientists and machine learning practitioners seeking to harness label-efficient deep learning for intelligent and scalable seismic exploration.
