Explore the vast range of titles available.

Gastric cancer (GC) is characterized by pronounced molecular and clinical heterogeneity, creating major challenges for therapeutic decision-making. Limitations in current molecular classification hinder the development of personalized therapies, underscoring the need for improved diagnostic and prognostic frameworks. we conducted an integrated multi-omics analysis of bulk, single-cell, and spatial transcriptomic data to systematically characterize three key programmed cell death pathways—pyroptosis, apoptosis, and necroptosis (collectively abbreviated as PAN). A scoring-based clustering framework integrating multiple machine learning algorithms was developed to define high-resolution molecular subtypes and construct a deep learning signature. A hybrid CNN+BiLSTM model with cross-fusion attention was applied for transcriptomic feature extraction and subtype classification, achieving superior performance compared with existing approaches. Validation in the TCGA cohort confirmed the robustness and biological relevance of our model. Among the identified subtypes, Subtype 2 showed the most favorable prognosis. We further established a nine-gene prognostic signature with strong predictive value. High-risk patients exhibited poor survival, enhanced immune infiltration, and potential sensitivity to AKT inhibitors, with several drugs, including gefitinib and paclitaxel, identified as promising candidates. Experimental validation was conducted using the Human Protein Atlas (HPA) and RT-qPCR in clinical samples. CFLAR and TNFSF13B were upregulated and PDK4 downregulated in GC, while UACA showed no significant change. Additional prognostic genes (DFFB, PSMB6, GLP1R, HDAC9, BACH2) displayed expression patterns largely consistent across HPA, TCGA, and RT-qPCR, with minor discrepancies likely due to sample size. This study integrates multi-omics and deep learning with experimental validation, providing insights into programmed cell death regulation and offering robust biomarkers and therapeutic targets for GC.

Fast switching speeds and high switching frequencies bring serious crosstalk problems in SiC MOSFET applications. In this paper, we design a SiC MOSFET gate driver with high crosstalk suppression capability by using a multi-level drive and active Miller clamp technology. In the design, an auxiliary branch is introduced to control the source potential of the SiC MOSFET to achieve multilevel driving. The branch has a simple structure, simple control logic, no external negative voltage supply, and adjustable negative output voltage. The proposed SiC MOSFET gate driver was designed using the Central Semiconductor Manufacturing Corporation (CSMC) 0.8 μm BCD high voltage process. The designed SiC MOSFET gate driver has an area of 2967 μm × 3180 μm. The simulation verification model is based on Wolfspeed’s SiC MOSFET product C3M0075120D. Post-layout simulation results show that a SiC MOSFET gate driver with a crosstalk suppression capability of over 150 V/ns is obtained, which can reliably drive SiC MOSFET power devices.

Knowledge distillation can transfer knowledge from a powerful yet cumbersome teacher model to a less-parameterized student model, thus effectively achieving model compression. Various knowledge distillation methods have mainly focused on the task of knowledge transfer, and distillation location selection, which in turn increases the difficulty of model interpretation on the one hand, and on the other hand, there have been few works on the role of the teacher classifier in distillation. In this study, we propose a novel collaborative multi-knowledge distillation under the influence of softmax regression representation. Firstly, we propose a stage-wise logit knowledge distillation, where the teacher classifier is used as an auxiliary structure to align the features of the student and teacher models. By leveraging the teacher classifier, the student features are aligned with the teacher features in the logits space, eliminating the need for a complex feature projector that requires extensive computation to match the features between the teacher and student models. Secondly, considering the teacher classifier’s adaptability to classification features, we introduce a stage-wise feature knowledge distillation. This mechanism maps the features of the student model to a latent space with the same dimensions as the features of the teacher model, guiding the student’s features to align with the teacher’s final features using a Mean Square Error (MSE) loss. Finally, we propose a pseudo-teacher knowledge distillation loss to optimize the modeling of the deformation relationship between the student and teacher features. This loss provides additional gradient optimization information for the parameters of the feature projector. Extensive experiments on CIFAR-100 and ImageNet datasets demonstrate the superiority of the proposed model compared with the state-of-the-art methods. The code is available at https://github.com/chenKP/CMKD.git

Knowledge distillation can transfer knowledge from a powerful yet cumbersome teacher model to a less-parameterized student model, thus effectively achieving model compression. Various knowledge distillation methods have mainly focused on the task of knowledge transfer, and distillation location selection, which in turn increases the difficulty of model interpretation on the one hand, and on the other hand, there have been few works on the role of the teacher classifier in distillation. In this study, we propose a novel collaborative multi-knowledge distillation under the influence of softmax regression representation. Firstly, we propose a stage-wise logit knowledge distillation, where the teacher classifier is used as an auxiliary structure to align the features of the student and teacher models. By leveraging the teacher classifier, the student features are aligned with the teacher features in the logits space, eliminating the need for a complex feature projector that requires extensive computation to match the features between the teacher and student models. Secondly, considering the teacher classifier’s adaptability to classification features, we introduce a stage-wise feature knowledge distillation. This mechanism maps the features of the student model to a latent space with the same dimensions as the features of the teacher model, guiding the student’s features to align with the teacher’s final features using a Mean Square Error (MSE) loss. Finally, we propose a pseudo-teacher knowledge distillation loss to optimize the modeling of the deformation relationship between the student and teacher features. This loss provides additional gradient optimization information for the parameters of the feature projector. Extensive experiments on CIFAR-100 and ImageNet datasets demonstrate the superiority of the proposed model compared with the state-of-the-art methods. The code is available at https://github.com/chenKP/CMKD.git

The inspection of conductive particles after Anisotropic Conductive Film (ACF) bonding is a common and crucial step in the TFT-LCD manufacturing process since the number of high-quality conductive particles is a key indicator of ACF bonding quality. However, manual inspection under microscope is a time-consuming, tedious, and error-prone. Therefore, there is an urgent demand in industry for the automatic conductive particle inspection system. It is challenging for automatic conductive particle quality inspection due to the existence of complex background noise and diversified particle appearance, including shape, size, clustering and overlapping, etc. As a result, it lacks an effective automatic detection method to handle all the complex particle patterns. In this paper, we propose a U-shaped deep residual neural network (i.e., U-ResNet), which can learn features of particle from massively labeled data. The experimental results show that the proposed method achieves high detection accuracy and recall rate, which exceedingly outperforms the previous work. Also, our system is very efficient and can work in real time.

The importance of gas acceleration near a wellbore in radial compressible porous media flows is quantified in terms of a dimensionless parameter, and conditions are identified under which gas acceleration is mainly responsible for the change in the pore pressure distribution and mass flux. Gas acceleration and Forchheimer drag both steepen the pressure profile and have significant impact on the pressure curve near the wellbore for a given wellbore pressure. For unchoked flows, the properties of a compressible accelerating gas flow can be modeled by a Darcy–Forchheimer flow with an upward adjusted Forchheimer drag coefficient. For choked flows, the Darcy–Forchheimer equation cannot be used to mimic the accelerating flow no matter how large the Forchheimer drag coefficient is. It is demonstrated that the value of the Forchheimer drag coefficient in some previous studies was inflated due to omission of the gas acceleration in the momentum equation.

● Co-occurrence of SMX and Gd(III) enhances HGT of ARGs and MRGs. ● Gd(III) alone negatively impacts ARGs and MRGs proliferation and spread. ● Streptomyces , Pseudomonas and Thauera were abundant in the presence of SMX. ● A positive correlation between internal ARGs and MGEs. With the increasing use of antibiotics and rare earth elements (REE) during the coronavirus disease (COVID-19) pandemic, the co-occurrence of sulfamethoxazole (SMX) and gadolinium (Gd) has increased in wastewater treatment plants (WWTPs). However, the effects of SMX and Gd exposure on the transmission of antibiotic resistance genes (ARGs) and heavy metal resistance genes (MRGs) remain unknown. This study investigated the impacts of SMX and Gd on the fate of ARGs and MRGs in an activated sludge system. The diversity and relative abundance of ARGs, MRGs, and mobile genetic elements (MGEs) were detected by metagenomic sequencing. The results revealed an increased abundance of ARGs but a decreased abundance of MRGs under the joint effect of SMX and Gd. In addition, Gd alone exerted adverse effects on the proliferation and spread of ARGs and MRGs. However, SMX alone resulted in an increase in the diversity of ARGs and MRGs and promoted the growth of Pseudomonas, Thauera, and Streptomyces in the activated sludge system. Interestingly, a positive correlation was observed between most ARGs and MGEs. These findings provide comprehensive insights into the effects of co-occurring REEs and antibiotics on the fate of ARGs, MRGs, and MGEs, providing evidence to assist in controlling the spread and proliferation of ARGs and MRGs in activated sludge systems.

For the permanent magnet synchronous linear motor’s (PMSLM) force ripple, the mathematical model of detent force is established in this paper, and presents a suppression strategy based on neural network. By the designing of BP neural network force ripple observer, theoretical analysis shows can effectively restrain the force ripple. Simulation results show the correctness and validity of the suppression strategy.

De novo transcriptome assembly is an important approach in RNA-Seq data analysis and it can help us to reconstruct the tran- scriptome and investigate gene expression profiles without reference genome sequences. We carried out transcriptome assem- blies with two RNA-Seq datasets generated from human brain and cell line, respectively. We then determined an efficient way to yield an optimal overall assembly using three different strategies. We first assembled brain and cell line transcriptome using a single k-mer length. Next we tested a range of values of k-mer length and coverage cutoff in assembling. Lastly, we com- bined the assembled contigs from a range of k values to generate a final assembly. By comparing these assembly results, we found that using only one k-mer value for assembly is not enough to generate good assembly results, but combining the contigs from different k-mer values could yield longer contigs and greatly improve the overall assembly.

Modern robot navigation systems encounter difficulties in diverse and complex indoor environments. Traditional approaches rely on multiple modules with small models or rule-based systems and thus lack adaptability to new environments. To address this, we developed Astra, a comprehensive dual-model architecture, Astra-Global and Astra-Local, for mobile robot navigation. Astra-Global, a multimodal LLM, processes vision and language inputs to perform self and goal localization using a hybrid topological-semantic graph as the global map, and outperforms traditional visual place recognition methods. Astra-Local, a multitask network, handles local path planning and odometry estimation. Its 4D spatial-temporal encoder, trained through self-supervised learning, generates robust 4D features for downstream tasks. The planning head utilizes flow matching and a novel masked ESDF loss to minimize collision risks for generating local trajectories, and the odometry head integrates multi-sensor inputs via a transformer encoder to predict the relative pose of the robot. Deployed on real in-house mobile robots, Astra achieves high end-to-end mission success rate across diverse indoor environments.