Hong Kong University of Science and Technology
Hong Kong University of Science and Technology Institutional RepositoryNot a member yet
162821 research outputs found
Sort by
Unveiling decrosslinking degree on polyamide nanofiltration membranes: Structure–performance relationship and micropollutant removal
No full textPolyamide (PA) nanofiltration membranes (NFMs) possess highly crosslinked structures, leading to several technical limitations such as poor selectivity, low permeance, and severe fouling. This study devised a novel isopropanol (IPA)-induced decrosslinking strategy to systematically modulate PA crosslinking density by precisely controlling IPA concentration and temperature. IPA induces swelling effects and hydrogen bonding, selectively extracting loosely crosslinked PA chain segments and allowing for network relaxation and rearrangement to form a uniform structure with lower overall crosslinking density. The optimised NF membrane post-treated with IPA at 60 °C (NF-IPA@60) achieved 32.94 % decrosslinking with enhanced surface negative charge density and enlarged pore size. Compared to the control membrane (NF-H2O), NF-IPA@60 exhibited higher pure water permeance (27.3 LMH/bar) while maintaining 92.7 % Na2SO4 rejection. Enhanced Donnan exclusion significantly improved micropollutant removal, particularly for anionic contaminants such as perfluorooctanoic acid and perfluorooctanesulfonic acid. Pearson correlation analysis established clear structure–performance relationships for the decrosslinking strategy. Natural water testing showed that NF-IPA@60 exhibit total organic carbon removal efficiencies exceeding 88.2 %, with 3.15-fold improvement in organic/mineral selectivity and excellent antifouling properties. This study provides a novel theoretical framework and technical approach for high-performance NFM design, offering significant potential for overcoming traditional membrane technology bottlenecks.</p
Predicting immunotherapy outcomes in NSCLC using RNA and pathology from multicenter clinical trials
No full textImmune checkpoint inhibitors (ICIs) are widely used to treat advanced non-small cell lung cancer (NSCLC). However, it remains crucial to identify patients who are unlikely to benefit from immunotherapy and to explore potential combination treatment strategies. In this study, 1127 advanced NSCLC patients from multicenter randomized clinical trials (OAK, POPLAR, ORIENT-11) and an in-house cohort who received ICIs, ICIs combined with chemotherapy, or chemotherapy alone are analyzed. Using bulk RNA-seq transcriptomic data, an RNA-based model, named the Lung Cancer Immunotherapy Response Assessment (LIRA), is developed, utilizing interaction analysis and a random forest algorithm to predict immunotherapy outcomes. LIRA outperforms PD-L1 expression and tumor mutation burden in predicting responses, particularly in identifying early progression risk during ICI monotherapy (HR: 0.15, 95% CI: 0.11–0.20). Tumor profile analysis reveals that LRP8 and HDAC4 are associated with immunotherapy outcomes. Additionally, scRNA-seq analysis of NSCLC tumors indicates a higher prevalence of T cells and a reduced proportion of epithelial cells in samples with a high LIRA-score. The deep learning model pinpointed critical high-attention regions within whole-slide images that contributed decisively to the LIRA predictions. In summary, these results demonstrate that LIRA enables independent risk stratification of NSCLC patients and provides insights into potential resistance mechanisms.</p
Fast soft sensor model migration in injection molding processes: A reconstruction based deep learning framework
No full textThe current manufacturing industry is characterized by rapidly evolving product requirements, making it both essential and challenging to develop optimum soft sensors for the injection molding (IM) process for precise and timely quality prediction. While a well-trained soft sensor may perform effectively in a conventional manufacturing process, it still requires reconfiguration in a new process due to the changed process dynamics. However, collecting sufficient new labeled data is time-consuming and labor-intensive. Although unlabeled data is easier to obtain, it cannot be directly utilized in supervised learning tasks, creating a conflict with production line demands when new data is limited in quantity and quality. To address this problem, this study proposes a semi-supervised-aided deep model migration framework (SSLA-DMM) to enhance training efficiency. It migrates well-trained parameters from a base model to initialize a new soft sensor at the target production site. In addition, a semi-supervised tri-training strategy is complemented to leverage valuable information from unlabeled new data. The effectiveness and superiority of our framework are validated using practical IM process data under different raw materials. Experimental results demonstrate that the migration module provides superior initial parameters, accelerating the deep learning process and reducing the required dataset size by approximately 60 %. Integrated with semi-supervised learning via tri-training, the SSLA-DMM framework attains an RMSE of 0.0568 and R2 of 0.9985, showcasing its capability to reduce data requirement while maintaining high prediction accuracy significantly. This study presents a promising solution for the challenges posed by production process transfers in the manufacturing industry.</p
Bayesian Compressive Sensing for Site Characterization
No full textSite characterization is indispensable to good geotechnical or rock engineering practice as every site is unique, but technical, budget, time, or access constraints typically result in only a tiny fraction of the underground soil and rock in a site being visually inspected, sampled, or tested. This leads to a long- lasting challenge of sparse measurements in geo- sciences and engineering. This book introduces Bayesian compressive sensing or sampling (BCS) as a highly efficient spatial data analytic and simulation method for the efficient modelling of spatial geo- data from sparse measurements, with quantified reliability and uncertainty to further optimize site characterization. It provides the necessary theory and computational tools for setting up and solving a sparse spatial data modeling problem using BCS. This book suits graduate students, academics, researchers, and engineers interested in site characterization from sparse measurements in geotechnical and rock engineering, and also those modeling other spatially varying phenomena such as air quality data, soil or water pollution data, and meteorological data. This is supplemented with a software called Analytics of Sparse Spatial Data using Bayesian compressive sampling/sensing and illustrative examples, and enables hands- on experience of spatial data analytics and simulation using sparse measurements.</p
CNNs in drug design
No full textMachine-learning-based approaches have shown effectiveness in drug representation learning and aiding in the development of potential new drugs. Convolutional Neural Networks (CNNs), which have achieved significant success in numerous areas of artificial intelligence, also hold great potential for modeling biological data. In addition to the standard CNN, numerous specialized CNN architectures have been created to handle structured molecules and proteins. This chapter explores the different CNN models utilized in drug design. Specifically, we will examine various CNN variants based on their applications in drug design, such as drug-target interaction prediction, drug sensitivity and response prediction, drug-drug interactions side effect prediction, as well as drug-drug similarity prediction. We will also identify the challenges that need to be addressed and outline future research directions.</p
Resolved CFD-DEM for high-fidelity multiphase flow modeling in porous media of arbitrary geometry
No full textThe design of resilient coastal infrastructure requires high-fidelity modelling of complex interactions between waves, porous structures, and mobile seabed. To address this need, we develop a novel computational framework that couples Computational Fluid Dynamics (CFD) and the Discrete Element Method (DEM), explicitly integrating a resolved porous media module. This approach enables direct numerical simulation of multiphase flows and their particle-scale interactions with both stationary and mobile porous structures, such as breakwaters or armor units. The model is rigorously validated against six benchmark cases, demonstrating robust capabilities in capturing permeability, capillary effects, and fluid–solid momentum exchange. We further apply the framework to large-scale coastal scenario featuring realistic wave generation, curved and trapezoidal seawalls, and over one hundred mobile cubic armor units. The simulations provide deep insights into critical processes like wave reflection, entrapped air dynamics, and drag-induced energy dissipation. The simulation results quantitatively show that porous structures significantly enhance wave energy dissipation, leading to superior wave attenuation. This integrated framework represents a significant advancement for high-fidelity, efficient simulations of fluid–structure interactions in dynamic and porous coastal environments, with great potential for coastal engineering design and environmental fluid mechanics.</p
Contextual Search in Principal-Agent Games: The Curse of Degeneracy
No full textIn this work, we introduce and study contextual search in general principal-agent games, wherea principal repeatedly interacts with agents by offering contracts based on contextual information and historicalfeedback, without knowing the agents’ true costs or rewards. Our model generalizes classical contextual pricingby accommodating richer agent action spaces. Over T rounds with d-dimensional contexts, we establish anasymptotically tight exponential T 1−Θ(1/d) bound in terms of the pessimistic Stackelberg regret, benchmarkedagainst the best utility for the principal that is consistent with the observed feedback.We also establish a lower bound of Ω(T 1/2 − 1/2d ) on the classic Stackelberg regret for principal-agent games,demonstrating a surprising double-exponential hardness separation from the contextual pricing problem (a.k.a,the principal-agent game with two actions), which is known to admit a near-optimal O(d log log T ) regretbound [Kleinberg and Leighton, 2003, Leme and Schneider, 2018, Liu et al., 2021]. In particular, this double-exponential hardness separation occurs even in the special case with three actions and two-dimensional context.We identify that this significant increase in learning difficulty arises from a structural phenomenon that we callcontextual action degeneracy, where adversarially chosen contexts can make some actions strictly dominated (andhence unincentivizable), blocking the principal’s ability to explore or learn about them, and fundamentally limitinglearning progres
Partial duodeno-ileal diversion with magnetic compression anastomosis: a novel approach for the treatment of type 1 diabetes-associated metabolic derangements in a porcine model
No full textBackground: Metabolic surgery has shown effectiveness in the management of some patients with type 2 diabetes (T2D). However, the evidence for metabolic surgery in type 1 diabetes (T1D) remains scarce. This pilot study aimed to evaluate the feasibility, safety, and metabolic impact of partial duodeno-ileal diversion (PDID) using a novel magnetic compression small bowel anastomosis (MCA) system in a T1D porcine model. Methods: Ten adult Yorkshire pigs were subjected to intravenous infusion of streptozotocin (STZ) to induce T1D. Diabetic pigs were randomized to either PDID via MCA or control with no operation. The proximal magnet was placed in the duodenum, and the distal magnet was inserted into the distal ileum about 40 cm proximal to the ileocecal valve through a small enterotomy, and magnetic coupling ensued. Procedural feasibility (MCA device coupling efficiency, anastomotic patency) and safety (e.g. burst pressure, device-related complications, survival) were evaluated. Postoperative changes including fasting plasma glucose (FPG), body weight, Homeostatic Model Assessment of Insulin Resistance [HOMA-IR]) were assessed periodically up to 6 weeks postoperatively. Histopathologic and immunohistochemical changes in the distal ileum and pancreas were also evaluated. Results: Five adult Yorkshire pigs underwent successful PDID via MCA, and 5 unoperated counterparts served as controls. One pig in each group died soon after induction, likely due to STZ toxicity. Anastomotic magnets were expelled per rectally within 14 days of PDID without any complications. In postoperative week 3, FPG normalized to a mean of 5.13 (±0.618) mmol/L in the PDID group and remained stable, whereas in the control group, mean FPG levels were persistently elevated at a mean of 30.28 (±3.168, p < 0.001) mmol/L. The PDID group had a mean HOMA-IR value of 0.892 (±0.220), in contrast to the control group’s mean of 6.056 (±0.803, p < 0.001). Histopathological examinations showed satisfactory tissue healing at the anastomotic sites of all pigs in the PDID group, focal hyperplasia of beta cells in the pancreas of all pigs that underwent PDID, and increased immunostaining for synaptophysin, glucagon-like peptide-1 and proinsulin in the distal ileum and pancreatic tissues in the PDID group, as compared with the controls. Conclusions: PDID via MCA is safe and effective in optimization of glycemic control in this STZ-induced T1D porcine model. Further preclinical validation and evaluation of metabolic changes in response to this novel intervention in different large animal models of T1D is needed before clinical trials in humans.</p
Bio-geomorphologic effects on blue carbon accumulation in mangrove sediments
No full textBlue carbon habitats, such as mangroves, salt marshes and seagrasses, are vital carbon sinks, storing large amounts of organic carbon in sediments, which often exhibit pronounced spatial variability even within a single site. Here, we examined the bio-geomorphologic controls underlying this spatial heterogeneity through year-round field investigation in a mangrove ecosystem along China's southern coast. Field observations across an elevation gradient revealed that sediment organic carbon stocks generally decreased with lowering elevation, with vegetated zones storing significantly more organic carbon than adjacent bare mudflats. Further analysis demonstrated that wave-induced bed level variability—rather than sediment input alone—was the primary factor governing carbon storage in the sediment. Vegetated zones had much higher sediment organic carbon stocks than bare mudflats due to the reduced hydrodynamic disturbance provided by the vegetation. Despite receiving six times more sediment input, the sediment carbon stock in the top 15 cm of bare mudflats was less than 25 % of that in the vegetated zones, due to experiencing at least three times more erosion and carbon loss. These findings establish a bio-geomorphologic framework for understanding heterogeneous carbon sequestration within mangroves and extend to other vegetated blue carbon habitats. By providing empirical evidence for the regulating role of bio-geomorphologic feedbacks, this study enhances predictive understanding of organic carbon burial processes and informs long-term assessments of coastal carbon sink dynamics under changing environmental conditions.</p
Hierarchical acceptor engineering of NIR-II butterfly aggregation-induced emission luminogens for enhanced theranostic efficiency
No full textSecond near-infrared (NIR-II) multimodal phototheranostics represents an advancing frontier in oncology, demanding precise orchestration of photosensitizer architectures at molecular and aggregate levels to navigate competitive energy dissipation pathways and achieve diagnostic-therapeutic synergy. Prevailing methodologies centered predominantly on donor engineering and π-bridge optimization have engendered a paucity of investigations into acceptors, notwithstanding their pivotal role in governing phototheranostic efficacy. Herein, we presented a hierarchical acceptor engineering approach to optimize multimodal phototheranostics at both molecular and aggregate levels. Three butterfly-shaped donor-acceptor-donor photosensitizers exhibiting aggregation-induced emission (AIE) in the NIR-II region were designed by strategically introducing steric phenyl group and barrier-free trifluoromethyl (CF3) rotors as acceptor substituents stepwise. Among them, CF3Ph-TPTCF exhibited the highest molar extinction coefficient, longer NIR-II emission wavelength with moderate NIR-II relative quantum yield and typical AIE feature. Quantum chemical calculations and molecular dynamics simulations substantiated the structure-property relationship of the molecules, the excited-state energy dissipation pathways and the impact of intramolecular motions on photophysical properties. Moreover, CF3Ph-TPTCF nanoparticles demonstrated optimal generation of reactive oxygen species, superior photothermal performance and achieved complete tumor elimination in vivo. This work established an acceptor engineering paradigm and elucidated a mechanism framework providing critical insights for designing next-generation multifunctional phototheranostic platforms.</p