67639 research outputs found
Sort by
Robust full waveform inversion with deep Hessian deblurring
No full textSummary
Full Waveform Inversion (FWI) is a technique widely used in geophysics to obtain high-resolution subsurface velocity models from waveform seismic data. Due to its large computation cost, most flavors of FWI rely only on the computation of the gradient of the loss function to estimate the update direction, therefore ignoring the contribution of the Hessian. Depending on the level of computational resources one can afford, an approximate of the inverse of the Hessian can be calculated and used to speed up the convergence of FWI towards the global (or a plausible local) minimum. In this work, we propose to use an approximate Hessian computed from a linearization of the wave-equation as commonly done in Least-Squares Migration (LSM). More precisely, we rely on the link between a migrated image and a doubly migrated image (i.e., an image obtained by demigration-migration of the migrated image) to estimate the inverse of the Hessian. However, instead of using non-stationary compact filters to link the two images and approximate the Hessian, we propose to use a deep neural network to directly learn the mapping between the FWI gradient (output) and its Hessian (blurred) counterpart (input). By doing so, the network learns to act as an approximate inverse Hessian: as such, when the trained network is applied to the FWI gradient, an enhanced update direction is obtained, which is shown to be beneficial for the convergence of FWI. The weights of the trained (deblurring) network are then transferred to the next FWI iteration to expedite convergence. We demonstrate the effectiveness of the proposed approach on two synthetic datasets and a field dataset.The author thanks KAUST and the DeepWave Consortium sponsors for supporting this research, as well as Equinor and the Volve license partners for releasing the field data set
Meta learning for mutant HLA class I epitope immunogenicity prediction to accelerate cancer clinical immunotherapy
No full textAccurate prediction of binding between human leukocyte antigen (HLA) class I molecules and antigenic peptide segments is a challenging task and a key bottleneck in personalized immunotherapy for cancer. Although existing prediction tools have demonstrated significant results using established datasets, most can only predict the binding affinity of antigenic peptides to HLA and do not enable the immunogenic interpretation of new antigenic epitopes. This limitation results from the training data for the computational models relying heavily on a large amount of peptide-HLA (pHLA) eluting ligand data, in which most of the candidate epitopes lack immunogenicity. Here, we propose an adaptive immunogenicity prediction model, named MHLAPre, which is trained on the large-scale MS-derived HLA I eluted ligandome (mostly presented by epitopes) that are immunogenic. Allele-specific and pan-allelic prediction models are also provided for endogenous peptide presentation. Using a meta-learning strategy, MHLAPre rapidly assessed HLA class I peptide affinities across the whole pHLA pairs and accurately identified tumor-associated endogenous antigens. During the process of adaptive immune response of T-cells, pHLA-specific binding in the antigen presentation is only a pre-task for CD8+ T-cell recognition. The key factor in activating the immune response is the interaction between pHLA complexes and T-cell receptors (TCRs). Therefore, we performed transfer learning on the pHLA model using the pHLA-TCR dataset. In pHLA binding task, MHLAPre demonstrated significant improvement in identifying neoepitope immunogenicity compared with five state-of-the-art models, proving its effectiveness and robustness. After transfer learning of the pHLA-TCR data, MHLAPre also exhibited relatively superior performance in revealing the mechanism of immunotherapy. MHLAPre is a powerful tool to identify neoepitopes that can interact with TCR and induce immune responses. We believe that the proposed method will greatly contribute to clinical immunotherapy, such as anti-tumor immunity, tumor-specific T-cell engineering, and personalized tumor vaccine.The authors acknowledge support from the National Key R and D Program of China (2022YFF1202100), National Natural Science
Foundation of China (Nos. 62202092, 62272135, and 62372135), and the King Abdullah University of Science and Technology
(KAUST) Office of Research Administration (ORA) under Award NoREI/1/5234-01-01, REI/1/5414-01-01, REI/1/5289-01-01, REI/1/5404-
01-01, REI/1/5992-01-01, URF/1/4663-01-01, Center of Excellence for Smart Health (KCSH), under award number 5932, and Center
of Excellence on Generative AI, under award number 5940
smDeepFLUOR: Single-Molecule Deep Learning Fluorescence Classification
No full textFluorescence intensity variation has long served as a primary readout for monitoring biological events. However, single-fluorophore signals arising from distinct molecular events often exhibit similar intensity profiles, making further classification challenging using conventional methods. In this study, we introduce smDeepFLUOR, a deep learning-based framework that resolves seemingly homogeneous spatiotemporal fluorescence signals by uncovering latent features imperceptible to conventional analyses. By leveraging a three-dimensional convolutional neural network trained on image sequences captured over 7 x 7 x 10 voxel windows, smDeepFLUOR reliably distinguishes specific from nonspecific protein binding, even across different experimental days, with an accuracy of up to 97%. Remarkably, smDeepFLUOR also captures real-time DNA synthesis kinetics by identifying subtle changes in the spatial distance between the fluorophore and the 3' end of nascent DNA, a feature undetectable by traditional methods. These classifications were achieved without incorporating explicit physical rules or engineered features, implying the presence of intrinsic, previously unrecognized differences in emission patterns. This approach significantly extends the analytical capabilities of single-molecule fluorescence imaging and opens new avenues for minimally labeled and label-free protein activities.This work was supported by the National Research Foundation of Korea funded by the Ministry of Science and ICT (RS-2023-00280169 and RS-2023-00218927) and by Basic Science Research Institute Fund, whose NRF grant number is RS -2021- NR060139
Thermo-Dynamic Flux Balance Analysis, a novel approach to simulate circadian metabolism
No full textCircadian metabolism arises from complex, time-dependent interactions across organs, yet experimental characterization in humans remains limited. Current whole-body metabolic models are either too large for dynamic simulation or insufficiently detailed to capture temporal physiology. We developed a multi-tissue human metabolic reconstruction, HGEM1.19+, an extensively curated model for dynamic simulation. We introduced Thermo-Dynamic Flux Balance Analysis (tdFBA), a novel formulation that integrates thermodynamic constraints, organ-specific enzyme capacities, solubility limits, osmotic balance and pH buffering. Murine circadian transcriptomics and metabolomics were integrated with human datasets to parameterize temporal dynamics. A genetic algorithm optimized internal parameters under CHOW diet conditions and model performance was evaluated using circadian metabolomics and gene-sage patterns. The framework recapitulated key metabolic oscillations under CHOW diet and partially reconstructed perturbed states including HFD, BMAL1 knockout, and tissue-specific re-entrainment. tdFBA enables physiologically grounded, multi-organ dynamic simulations and provides a foundation for exploring human circadian metabolism and perturbation responses.<br
A global database of soil microbial phospholipid fatty acids and enzyme activities.
No full textSoil microbes drive ecosystem function and play a critical role in how ecosystems respond to global change. Research surrounding soil microbial communities has rapidly increased in recent decades, and substantial data relating to phospholipid fatty acids (PLFAs) and potential enzyme activity have been collected and analysed. However, studies have mostly been restricted to local and regional scales, and their accuracy and usefulness are limited by the extent of accessible data. Here we aim to improve data availability by collating a global database of soil PLFA and potential enzyme activity measurements from 12,258 georeferenced samples located across all continents, 5.1% of which have not previously been published. The database contains data relating to 113 PLFAs and 26 enzyme activities, and includes metadata such as sampling date, sample depth, and soil pH, total carbon, and total nitrogen. This database will help researchers in conducting both global- and local-scale studies to better understand soil microbial biomass and function.We are grateful to Mark Bradford, Qiang Gao, Diana Wall and Deli Wang for providing data and advice. LGvG, JvdH, GRS and TWC acknowledge funding from DOB Ecology (EC-2021-GEM003), the Marc R. Benioff Revocable Trust in collaboration with the World Economic Forum, and the Bernina Initiative (2022-FS-318). GRS was additionally supported by a Graduate Research Fellowship from the United States National Science Foundation, a ThinkSwiss Research Scholarship from the Embassy of Switzerland in the United States of America, and an Ambizione fellowship from the Swiss National Science Foundation (PZ00P3_216194). KGP is a Canadian Institute for Advanced Research (CIFAR) Fellow in the program Fungal Kingdom: Threats and Opportunities, and is supported by the U.S. Department of Energy (DE-SC0023661), and the U.S. National Science Foundation (DEB-1845544 and DEB-1926335). JMA and TAA acknowledge funding from Qatar Petroleum (QUEX-CAS-QP-RD-18/19). ASFA acknowledges the National Council for Scientific and Technological Development (CNPq-Brazil) for his Fellowship of Research Productivity (Grant 301755/2022-1). MB was supported by the Swedish Research Councils Vetenskapsrådet (Grant 2021–03724). PVM and RF acknowledge support by national funds through FCT (UID/EMS/00285/2020). LGB and SLutz acknowledge financial support from a Helmholtz Recruiting Initiative award (I-044-16-0) and a Baillet Latour Antarctica Fellowship (2016–2018) to LAZ that supported part of the soil sample collection in Antarctica. NE, SC and CGu acknowledge support of iDiv funded by the German Research Foundation (DFG– FZT 118, 202548816). NE acknowledges funding by the DFG (Ei 862/29-1 and Ei 862/31-1) and the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program (677232). CGa acknowledges funding by the Portuguese Science and Technology Foundation (PTDC/BIA-CBI/2340/2020). HF and KPe acknowledge funding from the Academy of Finland (315415). FBa thanks the project PID2020-11942RB-I00 funded by MCIN/AEI/10.13039/501100011033, the ILINK program from CSIC (ILINK23065), the project SOMMIT funded by the European Joint Programme SOIL (862695), and Fundación Séneca (19896/GERM/15). MGB and HI acknowledge funding from the Fonds zur Förderung der wissenschaftlichen Forschung (FWF) Austria (I989-B16). MGB also acknowledges support from the Spanish State Agency for Research (AEI) for her Ramón y Cajal grant (RYC-2016-21231). VEJJ acknowledges financial support from the French National Research Agency (ANR; MIXOPEAT; ANR-17-CE01-0007). JZL and SSh acknowledge support from the German Research Foundation through the RTG 2300 (316045089). MZH and SSu acknowledge support from JSPS KAENHI (15658101 and 15380181). VK, SSh and AMP acknowledge the Deutsche Forschungsgemeinschaft (DFG; 192626868–SFB 990). SDF acknowledges support from the U.S. National Science Foundation Long-term Ecological Research (LTER) and Long-term Research in Environmental Biology (LTREB) Programs. MAA acknowledges support from a National Science Foundation Graduate Research Fellowship (DGE 1450271). RDB, AADB and WP acknowledge funding from the UK Natural Environment Research Council (NE/N009452/1). EJS was funded by a European Research Council Starting Grant (FP/2007-2013/ERC 307888). JP was supported by a European Union Horizon 2020 Marie Sklodowska-Curie grant (892654). ATN was supported by a UK NERC grant (NE/T012226). AJM was supported by Illinois Nutrient Research and Education Council (2021-4-360731-469), USDA NIFA (2021-67019-35068), and US NSF (2125626). CR acknowledges funding from the COST Action FP1305 BioLink and from the program “Sustainability and resilience – Tackling climate and environmental changes” (Vetenskapsrådet 2016-06327) jointly supported by the Swedish Research Council, the Swedish Research Council Formas, and Sida. RM acknowledges the funding from the University of Kassel and Valli Sustainability Research and Education. FTM acknowledges support by the King Abdullah University of Science and Technology (KAUST) and the KAUST Climate and Livability Initiative, the European Research Council (647038) and Generalitat Valenciana (CIDEGENT/2018/041). KZ acknowledges support from the US National Science Foundation (2244711). IH, MMo, JO, MÖ, SP, TV, and MZ acknowledge support from the Estonian Ministry of Education and Research (Centre of Excellence AgroCropFuture) and the Estonian Research Council (PRG1065, PRG1789, PSG784). CW was funded by the National Natural Science Foundation of China (32101286) and China Postdoctoral Science Foundation (2021M693360). CCD was supported by the FCT (BIPD_01_2021_FCT-PTDC/BIA-CBI/2340/2020, UIDB/05937/2020, UIDP/05937/2020). MNBG was supported by Agencia Nacional de Promoción Científica y Tecnológica of Argentina (PICT 2014-2838). KEM and NO were funded by the UK Natural Environment Research Council (NE/I027037/1). SYuE is grateful to the budgetary funding from the state assignments FWES-2024-0023. LB acknowledges financial support from the Swiss National Science Foundation (315260_149807, SPHAGNOL). NAS thanks NWO foundation for Vidi grant (016.161.318). Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government. ZZ was supported by the National Natural Science Foundation of China (32061143027). XLi was supported by the Jilin Province Science and Technology Development Plan Project (2020020s1003JC). LJ was supported by the Second Tibetan Plateau Scientific Expedition and Research (STEP) program (2019QZKK0302), and the National Natural Science Foundation of China (31872994). XLu was supported by the National Natural Science Foundation of China (41922056). PB was supported by the Ministry of Education, Youth and Sports of the Czech Republic (CZ.02.01.01/00/22_008/0004635). AP was supported by the Danish National Research Foundation within the Center for Volatile Interactions (VOLT, DNRF168). SMBR acknowledges FAPEPI and CAPES (code 001) for her PhD scholarship. JPS was supported by the strategic plan of the Centre for Functional Ecology - Science for People and the Planet (CFE) (UIDB/04004/2020), funded by FCT/MCTES through national funds (PIDDAC). PK was supported by the Czech Science Foundation (21-20802 M). EK acknowledges the DFG Priority Program 1374 ‘Infrastructure-Biodiversity-Exploratories’ (KA 1590/8-2, KA 1590/8-3) and AA acknowledges the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior – Brasil (CAPES) – Finance Code 001 – process 88881.172163/2018-01 for her fellowship. NF acknowledges the support of the Region Nouvelle-Aquitaine (AAPR2020A-2019–8472310). MMR acknowledges support from the Spanish Ministry of Science and Innovation for the I + D + i Project PID2021-123097OA-I00 funded by MCIN/ AEI /10.13039/501100011033/ and FEDER Una manera de hacer Europa and for the Project TED2021-132332A-C22, funded by MCIN/AEI /10.13039/501100011033 and Unión Europea NextGenerationEU/ PRTR. JV was supported by the Czech Science Foundation project (21-19209 M). Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.
Open access funding provided by Swiss Federal Institute of Technology Zurich
Distributed Coordination for Heterogeneous Non-Terrestrial Networks
No full textTo achieve global coverage and ubiquitous connectivity, the non-terrestrial network (NTN) has been regarded as a key enabler in the sixth generation (6G) network, which includes uncrewed aerial vehicles (UAVs), high-altitude platforms (HAPs), and satellites. Since the unique characteristics of various NTN platforms strongly affect their implementation and lead to a highly dynamic and heterogeneous NTN scenario, achieving distributed coordination remains an important research direction. However, the explicit and systematic analysis of the individual layers' challenges and corresponding distributed coordination solutions in heterogeneous NTNs has not been proposed yet. Therefore, in this paper, we summarize the unique characteristics of each NTN platform, identify communication challenges within individual layers, and propose potential delay-tolerant or delay-sensitive coordinated solutions accordingly. We further analyse the feasibility of leveraging multi-agent deep reinforcement learning (MADRL) algorithms to achieve the proposed coordinated solutions. Finally, we present a case study of the joint scheduling and trajectory optimization problem in heterogeneous NTN, where a two-timescale multi-agent deep deterministic policy gradient (TTS-MADDPG) algorithm is developed to validate the effectiveness of distributed coordination
Automated Theorem Proving with Large Language Models in Lean: An Exploration of Specialized In-Context Learning and General-Purpose Hierarchical Architectures
No full textThis thesis investigates novel strategies for Automated Theorem Proving (ATP)
using Large Language Models (LLMs) within the Lean ecosystem, addressing
challenges of cost, data dependency, and the formal-informal reasoning gap. Two
systems are presented: the Functional Equation Automated Solver (FEAS) and
Divide and Prove.
FEAS explores a cost-effective, fine-tuning-free approach for functional equa-
tions in Lean 3 using in-context learning (ICL) and heuristic-guided prompting.
Evaluations on the new FunEq benchmark show ICL’s viability for simpler prob-
lems in this niche, though the impact of explicit heuristics is model-dependent
and performance on harder problems is limited.
Divide and Prove implements a general-purpose, hierarchical ATP architec-
ture in Lean 4, using a general LLM for high-level planning and subgoal formal-
ization, and a specialized formal prover LLM for proof execution in an iterative
loop. On the miniF2F benchmark, it solved 74.59% of problems, with its itera-
tive pipeline resolving a subset intractable to initial direct proof attempts. While
demonstrating benefits of its adaptive design, its overall performance did not de-
cisively surpass state-of-the-art standalone provers employing extensive sampling
or end-to-end optimization
MODELLING SHALLOW WATER FLOW AND POLLUTANT TRANSPORT USING A WELL-BALANCED DISCONTINUOUS GALERKIN METHOD
No full textIn the present paper, a well-balanced second-order Runge-Kutta discontinuous Galerkin (RKDG) method is presented for modelling shallow water flows and pollutant transport over unstructured triangular meshes. The model is designed to handle arbitrary topography and moving wetting and drying fronts. The mathematical model describing these flows is derived from the shallow water equations (SWEs), which are coupled with the convection equation. The fluxes across cells interfaces are evaluated using the HLL Riemann solver. Treatment of topography source term is built in the DG approximation to guaranty the well-balanced property of the SWEs. The friction source term is calculated implicitly via a splitting integration approach. To improve the model, we adapt and refine positivity preserving and wetting and drying techniques originally designed for the finite volume schemes, ensuring its effective integration within the RKDG framework. The resulting method is implemented and systematically assessed for accuracy and robustness on a carefully selected suite of test problems including wetting and drying, irregular topography, and discontinuous flows. Numerical results are provided, showcasing the effectiveness of the proposed approach in accurately predicting both water depth and pollutant concentration
Advances in DNA damage repair mechanisms in stem cells and their applications.
No full textStem cells play a critical role in tissue regeneration and the maintenance of homeostasis. Due to their high replicative potential, stem cells face an elevated risk of DNA damage during DNA replication. Consequently, efficient DNA damage repair (DDR) mechanisms are essential for preserving the genomic stability and functionality of stem cells. This review summarizes the main DNA damage repair mechanisms, examines the characteristics of these DDR pathways in different stem cell types (highlighting their specific features and key molecules), and discusses the clinical significance and applications of stem cell DDR research. Furthermore, we identify current research limitations and propose potential future research directions. Collectively, this review provides a comprehensive perspective on DDR mechanisms in stem cells, laying a foundation for future investigations and potential clinical applications.We would like to extend our sincere gratitude to all those who have contributed to this review article.
First and foremost, we are deeply indebted to our supervisor, Mrs. Xianli Wang. Her extensive knowledge, insightful guidance, and meticulous supervision have been invaluable throughout the process of writing this review. From the initial conceptualization of the topic to the final refinement of the manuscript, Mrs. Xianli Wang provided constructive feedback and helped us navigate through the vast amount of literature, ensuring the accuracy and coherence of the review.
We also express our heartfelt thanks to the numerous researchers whose original works were carefully studied and cited in this review. Their contributions have laid the foundation for our understanding of the field and enriched the content of this article.
Special thanks go to the Library of Shanghai Jiao Tong University School of Medicine, which assisted us in accessing a wide range of academic resources.
The author(s) reported there is no funding associated with the work featured in this article
Thanos: A Block-wise Pruning Algorithm for Efficient Large Language Model Compression
No full textThis paper presents Thanos, a novel weightpruning algorithm designed to reduce the memory footprint and enhance the computational efficiency of large language models (LLMs) by removing redundant weights while maintaining accuracy. Thanos introduces a block-wise pruning strategy with adaptive masks that dynamically adjust to weight importance, enabling flexible sparsity patterns and structured formats, such as n : m sparsity, optimized for hardware acceleration. Experimental evaluations demonstrate that Thanos achieves state-of-the-art performance in structured pruning and outperforms existing methods in unstructured pruning. By providing an efficient and adaptable approach to model compression, Thanos offers a practical solution for deploying large models in resource-constrained environments