Spatiotemporal transformable nano-assembly for on-demand drug delivery to enhance anti-tumor immunotherapy

Induction of tumor cell senescence has become a promising strategy for anti-tumor immunotherapy,but fibrotic matrix severely blocks senescence inducers penetration and immune cells infiltration.Herein,we designed a cancer-associated fibroblasts(CAFs)triggered structure-transformable nano-assembly(HSD-P@V),which can directionally deliver valsartan(Val,CAFs regulator)and doxorubicin(DOX,senescence inducer)to the specific targets.In detail,DOX is conjugated with hyaluronic acid(HA)via diselenide bonds(Se-Se)to form HSD micelles,while CAFs-sensitive peptide is grafted onto the HSD to form a hydrophilic polymer,which is coated on Val nanocrystals(VNs)surface for improving the stability and achieving responsive release.Once arriving at tumor microenvironment and touching CAFs,HSD-P@V disintegrates into VNs and HSD micelles due to sensitive peptide detachment.VNs can degrade the extracellularmatrix,leading to the enhanced penetration of HSD.HSD targets tumor cells,releases DOX to induce senescence,and recruits effector immune cells.Furthermore,senescent cells are cleared by the recruited immune cells to finish the integrated anti-tumor therapy.In vitro and in vivo results show that the nanoassembly remarkably inhibits tumor growth as well as lungmetastasis,and extends tumorbearing mice survival.This work provides a promising paradigm of programmed delivering multi-site nanomedicine for cancer immunotherapy.

Elite,transformable haploid inducers in maize

The introduction of alleles into commercial crop breeding pipelines is both time consuming and costly.Two technologies that are disrupting traditional breeding processes are doubled haploid(DH)breeding and genome editing(GE).Recently,these techniques were combined into a GE trait delivery system called HI-Edit(Haploid Inducer-Edit).In HI-Edit,the pollen of a haploid inducer line is reprogrammed to deliver GE traits to any variety,obviating recurrent selection.For HI-Edit to operate at scale,an efficient transformable HI line is needed,but most maize varieties are recalcitrant to transformation,and haploid inducers are especially difficult to transform given their aberrant reproductive behaviors.Leveraging marker assisted selection and a three-tiered testing scheme,we report the development of new Iodent and Stiff Stalk maize germplasm that are transformable,have high haploid induction rates,and exhibit a robust,genetically-dominant anthocyanin native trait that may be used for rapid haploid identification.We show that transformation of these elite‘‘HI-Edit”lines is enhanced using the BABYBOOM and WUSCHEL morphogenetic factors.Finally,we evaluate the HI-Edit performance of one of the lines against both Stiff Stalk and non-Stiff Stalk testers.The strategy and results of this study should facilitate the development of commercially scalable HI-Edit systems in diverse crops.

Geometric Study of Two-Dimension Stellated Reentrant Auxetic Structures to Transformable Architecture

Transformable architecture is totally linked to the study and knowledge of geometry.There are some materials in nature,whose geometric invariants establish equivalent structural behavior regarding the scalar transformations,developing different spatial typologies according to dimensional variation.Auxetic materials are characterized by their negative Poisson’s ratio.They can change their geometric configuration from a line to a surface,and from a surface to a volume or spatial framework.This paper is based on establishing and comparing those stellated reentrant auxetic geometries to be able to build new spaces defined by their capacity for architectural transformation,studying analytically geometric properties of stellated reentrant auxetic structures that,from the molecular to the macroscopic level,can be part of the architecture construction.In this investigation,a comparative study by means of CAD of stellated reentrant auxetic patterns has been realized.A Computer-Aided Design study of stellated reentrant auxetic structures will be realized to use them in architecture.The geometric behavior of the different stellated reentrant auxetic patterns is analyzed from the developed study to generate a systematic comparison,evaluating properties of these forms,such as their maximum achievable area reductions in relation to the total length of bars of the structure,in order to obtain a growth factor.

A Sub-Nanostructural Transformable Nanozyme for Tumor Photocatalytic Therapy

The structural change-mediated catalytic activity regulation plays a significant role in the biological functions of natural enzymes.However,there is virtually no artificial nanozyme reported that can achieve natural enzyme-like stringent spatiotemporal structure-based catalytic activity regulation.Here,we report a subnanostructural transformable gold@ceria(STGC-PEG)nanozyme that performs tunable catalytic activities via near-infrared(NIR)light-mediated sub-nanostructural transformation.The gold core in STGC-PEG can generate energetic hot electrons upon NIR irradiation,wherein an internal sub-nanostructural transformation is initiated by the conversion between CeO;and electron-rich state of CeO;-x,and active oxygen vacancies generation via the hot-electron injection.Interestingly,the sub-nanostructural transformation of STGC-PEG enhances peroxidase-like activity and unprecedentedly activates plasmon-promoted oxidase-like activity,allowing highly efficient low-power NIR light(50 m W cm;)-activated photocatalytic therapy of tumors.Our atomic-level design and fabrication provide a platform to precisely regulate the catalytic activities of nanozymes via a light-mediated sub-nanostructural transformation,approaching natural enzyme-like activity control in complex living systems.

Optimization of body configuration and joint-driven attitude stabilization for transformable spacecraft under solar radiation pressure

The solar sail is one of the most promising space exploration systems due to its theoretically infinite specific impulse achieved through solar radiation pressure(SRP).Recently,researchers have proposed“transformable spacecraft”capable of actively reconfiguring their body configurations using actuatable joints.Transformable spacecraft,if used similarly to solar sails,are expected to significantly enhance orbit and attitude control capabilities owing to their high redundancy in control degrees of freedom.However,controlling them becomes challenging due to their large number of inputs,leading previous researchers to impose strong constraints to limit their potential control capabilities.This study focuses on novel attitude control techniques for transformable spacecraft under SRP.We developed two methods,namely,joint angle optimization to obtain arbitrary SRP force and torque,and momentum damping control driven by joint angle actuation.Our proposed methods are formulated in a general manner and can be applied to any transformable spacecraft comprising front faces that can predominantly receive the SRP on each body.The validity of our proposed method is confirmed through numerical simulations.Our study contributes to making most of the high control redundancy of transformable spacecraft without the need for expendable propellants,thus significantly enhancing the orbit and attitude control capabilities.

An enzyme-responsive and transformable PD-L1 blocking peptide-photosensitizer conjugate enables efficient photothermal

Mild photothermal therapy combined with immune checkpoint blockade has received increasing attention for the treatment of advanced or metastatic cancers due to its good therapeutic efficacy.However,it remains a challenge to facilely integrate the two therapies and make it potential for clinical translation.This work designed a peptide-photosensitizer conjugate(PPC),which consisted of a PD-L1 antagonist peptide(CVRARTR),an MMP-2 specific cleavable sequence,a self-assembling motif,and the photosensitizer Purpurin 18.The single-component PPC can self-assemble into nanospheres which is suitable for intravenous injection.The PPC nanosphere is cleaved by MMP-2 when it accumulates in tumor sites,thereby initiating the cancer-specific release of the antagonist peptide.Simultaneously,the nanospheres gradually transform into co-assembled nanofibers,which promotes the retention of the remaining parts within the tumor.In vivo studies demonstrated that PPC nanospheres under laser irradiation promote the infiltration of cytotoxic T lymphocytes and maturation of DCs,which sensitize 4T1 tumor cells to immune checkpoint blockade therapy.Therefore,PPC nanospheres inhibit tumor growth efficiently both in situ and distally and blocked the formation of lung metastases.The present study provides a simple and efficient integrated strategy for breast cancer photoimmunotherapy.

Effects of Al addition and cryogenic cyclic treatment on impact toughness of phase-transformable Ti-based bulk metallic glass composites

Developing bulk metallic glass composites(BMGCs)with high toughness is vital for their practical application.However,the influence of different microstructures on the impact toughness of BMGCs is still unclear.The effects of Al addition and cryogenic cyclic treatment(CCT)on the Charpy impact toughness,a K,at 298 and 77 K of a series of phase-transformable BMGCs are investigated in this work.It is found that deformation-induced martensitic transformation(DIMT)of theβ-Ti dendrites is the dominant toughening mechanism in the phase-transformable BMGCs at 298 K,but at 77 K,the toughness of BMGCs is primarily determined by the intrinsic toughness of the glass matrix.The addition of Al can moderately tune theβ-Ti phase stability,which then affects the amount of DIMT and impact toughness of the BMGCs at 298 K.However,at 77 K,Al addition causes a monotonic decrease in the toughness of the BMGCs due to the embrittlement of the glass matrix.It is found that CCT can effectively rejuvenate the phase-transformable BMGCs,which results in an enhanced impact toughness at 298 K.However,the toughness at 77 K monotonously decreases with increasing the number of CCT cycles,suggesting that the rejuvenation of the glass matrix affects the toughness at both 298 and 77 K of BMGCs,but in dramatically different ways.These findings reveal the influence of microstructures and CCT on the impact toughness of BMGCs and provide insights that could be useful for designing tougher BMGs and BMGCs.

The innate interfacial elastic strain field of a transformable B2 precipitate embedded in an amorphous matrix

When a transformable B2 precipitate is embedded in an amorphous matrix,it is often experimentally observed that the crystalline-amorphous interface not only serves as an initiation site for the martensitic transformation due to local stress concentrations,but also as an inhibitor to stabilize the transformation,the latter being attributed to the“confinement effect”exerted by the amorphous matrix,according to the Eshelby solution.These two seemingly incongruous factors are examined in this study using molecular dynamics simulations from an atomic interaction perspective.An innate strain gradient in the vicinity of the crystalline-amorphous interface is identified.The actual interface,the compressive/dilatative transition,and the interfacial maximum strain are investigated to differentiate from the conventional“interface”located within a distance of a few nanometers.Our innate interfacial elastic strain field model is applicable for the design of materials with a higher degree of martensitic transformation and controllable stress concentration,even in cryogenic environments.

基于Transformer和自适应特征融合的矿井低照度图像亮度提升和细节增强方法

高质量矿井影像为矿山安全生产提供保障,也有利于提高后续图像分析技术的性能。矿井影像受低照度环境的影响,易出现亮度低,照度不均,颜色失真,细节信息丢失严重等问题。针对上述问题,提出一种基于Transformer和自适应特征融合的矿井低照度图像亮度提升和细节增强方法。基于生成对抗思想搭建生成对抗式主体模型框架,使用目标图像域而非单一参考图像驱动判别器监督生成器的训练,实现对低照度图像的充分增强;基于特征表示学习理论搭建特征编码器,将图像解耦为亮度分量和反射分量,避免图像增强过程中亮度与颜色特征相互影响从而导致颜色失真问题;设计CEM-Transformer Encoder通过捕获全局上下文关系和提取局部区域特征,能够充分提升整体图像亮度并消除局部区域照度不均;在反射分量增强过程中,使用结合CEM-Cross-Transformer Encoder的跳跃连接将低级特征与深层网络处特征进行自适应融合,能够有效避免细节特征丢失,并在编码网络中添加ECA-Net,提高浅层网络的特征提取效率。制作矿井低照度图像数据集为矿井低照度图像增强任务提供数据资源。试验显示,在矿井低照度图像数据集和公共数据集中,与5种先进的低照度图像增强算法相比,该算法增强图像的质量指标PSNR、SSIM、VIF平均提高了16.564%,10.998%,16.226%和14.438%,10.888%,14.948%,证明该算法能够有效提升整体图像亮度,消除照度不均,避免颜色失真和细节丢失,实现矿井低照度图像增强。

基于RoBERTa和图增强Transformer的序列推荐方法

自推荐系统出现以来,有限的数据信息就一直制约着推荐算法的进一步发展。为降低数据稀疏性的影响,增强非评分数据的利用率,基于神经网络的文本推荐模型相继被提出,但主流的卷积或循环神经网络在文本语义理解和长距离关系捕捉方面存在明显劣势。为了更好地挖掘用户与商品之间的深层潜在特征,进一步提高推荐质量,提出一种基于Ro BERTa和图增强Transformer的序列推荐(RGT)模型。引入评论文本数据,首先利用预训练的Ro BERTa模型捕获评论文本中的字词语义特征,初步建模用户的个性化兴趣,然后根据用户与商品的历史交互信息,构建具有时序特性的商品关联图注意力机制网络模型,通过图增强Transformer的方法将图模型学习到的各个商品的特征表示以序列的形式输入Transformer编码层,最后将得到的输出向量与之前捕获的语义表征以及计算得到的商品关联图的全图表征输入全连接层,以捕获用户全局的兴趣偏好,实现用户对商品的预测评分。在3组真实亚马逊公开数据集上的实验结果表明,与Deep FM、Conv MF等经典文本推荐模型相比,RGT模型在均方根误差(RMSE)和平均绝对误差(MAE)2种指标上有显著提升,相较于最优对比模型最高分别提升4.7%和5.3%。

基于多尺度注意力特征融合的场景文本检测

针对目前文本检测中小尺度文本和长文本检测精度低的问题,提出了一种基于多尺度注意力特征融合的场景文本检测算法。该方法以Mask R-CNN为基线模型,引入Swin_Transformer作为骨干网络提取底层特征。在特征金字塔(feature pyramid networks,FPN)中,通过将多尺度注意力热图与底层特征通过横向连接相融合,使检测器的不同层级专注于特定尺度的目标,并利用相邻层注意力热图之间的关系实现了FPN结构中的纵向特征共享,避免了不同层之间梯度计算的不一致性问题。实验结果表明:在ICDAR2015数据集上,该方法的准确率、召回率和F值分别达到了88.3%、83.07%和85.61%,在CTW1500和Total-Text弯曲文本数据集上相较现有方法均有良好表现。

引入轻量级Transformer的无人机视觉跟踪

随着无人机在军事和民用领域的广泛运用,对于高精度、低功耗智能无人机跟踪系统的需求日益增加。针对目标跟踪算法在无人机跟踪场景下很难平衡跟踪精度和跟踪速度的问题,提出一种引入轻量级Transformer的孪生网络无人机目标跟踪算法SiamLT。使用Transformer对AlexNet网络进行改进,在增加最小计算量的情况下捕获全局特征信息。在目标模板与搜索区域匹配方面,联合Transformer和深度互相关运算提出一种二元相关模块,同时捕获目标模板与搜索区域之间的局部相关性和全局依赖关系。在分类回归网络中引入距离交并比,并采用多监督策略训练网络,以获取更准确的目标位置。在UAV123和UAV20L跟踪基准上的实验结果表明,SiamLT算法优于主流的目标跟踪算法,更有效地平衡了跟踪精度和跟踪速度。

考虑特征重组与改进Transformer的风电功率短期日前预测方法

短期日前风电功率预测对电力系统调度计划制定有重要意义,该文为提高风电功率预测的准确性,提出了一种基于Transformer的预测模型Powerformer。模型通过因果注意力机制挖掘序列的时序依赖;通过去平稳化模块优化因果注意力以提高数据本身的可预测性;通过设计趋势增强和周期增强模块提高模型的预测能力;通过改进解码器的多头注意力层,使模型提取周期特征和趋势特征。该文首先对风电数据进行预处理,采用完全自适应噪声集合经验模态分解(complete ensemble empirical mode decomposition with adaptive noise,CEEMDAN)将风电数据序列分解为不同频率的本征模态函数并计算其样本熵,使得风电功率序列重组为周期序列和趋势序列,然后将序列输入到Powerformer模型,实现对风电功率短期日前准确预测。结果表明,虽然训练时间长于已有预测模型,但Poweformer模型预测精度得到提升;同时,消融实验结果验证了模型各模块的必要性和有效性,具有一定的应用价值。

基于轨道板振动加速度的钢轨振动加速度反演估计与现场验证

为研究高速铁路轨道板与钢轨之间的时空关联规律,提出变分模态-转换器(Variational Mode Decomposition-Transformer, VMD-T)反演模型,该模型通过分解轨道板振动加速度来反演估计钢轨振动加速度.首先,对数据进行预处理并利用双门限法检测振动端点,分离振动信号与静默信号、干扰信号,再将提取后的振动信号整合输入到VMD-T模型.其次,利用VMD模型将轨道板振动加速度数据分解成一系列不同的子模态,并网格遍历搜索与钢轨振动加速度相关系数最大的子模态,以降低原始数据的复杂度以及非平稳性,提升Transformer模型的特征抽取能力.然后,通过Transformer模型对搜索出的轨道板振动加速度子模态与钢轨振动加速度数据进行反演估计训练.最后,将该模型应用于某城际高速铁路轨道板与钢轨实测振动加速度数据反演估计试验.现场高铁试验结果表明:与单一Transformer模型相比,VMD-T模型均方根误差(RootMean Squared Error, RMSE)、绝对平均误差(Mean Absolute Error, MAE)及决定系数(R2_score)分别提升近20%、11%及48.1%,特征学习能力更强,反演估计效果更佳,初步实现钢轨垂向振动加速度幅值非接触式监测估计.

基于Depth-wise卷积和视觉Transformer的图像分类模型

图像分类作为一种常见的视觉识别任务,有着广阔的应用场景。在处理图像分类问题时,传统的方法通常使用卷积神经网络,然而,卷积网络的感受野有限,难以建模图像的全局关系表示,导致分类精度低,难以处理复杂多样的图像数据。为了对全局关系进行建模,一些研究者将Transformer应用于图像分类任务,但为了满足Transformer的序列化和并行化要求,需要将图像分割成大小相等、互不重叠的图像块,破坏了相邻图像数据块之间的局部信息。此外,由于Transformer具有较少的先验知识,模型往往需要在大规模数据集上进行预训练,因此计算复杂度较高。为了同时建模图像相邻块之间的局部信息并充分利用图像的全局信息,提出了一种基于Depth-wise卷积的视觉Transformer(Efficient Pyramid Vision Transformer,EPVT)模型。EPVT模型可以实现以较低的计算成本提取相邻图像块之间的局部和全局信息。EPVT模型主要包含3个关键组件:局部感知模块(Local Perceptron Module,LPM)、空间信息融合模块(Spatial Information Fusion,SIF)和“+卷积前馈神经网络(Convolution Feed-forward Network,CFFN)。LPM模块用于捕获图像的局部相关性;SIF模块用于融合相邻图像块之间的局部信息,并利用不同图像块之间的远距离依赖关系,提升模型的特征表达能力,使模型学习到输出特征在不同维度下的语义信息;CFFN模块用于编码位置信息和重塑张量。在图像分类数据集ImageNet-1K上,所提模型优于现有的同等规模的视觉Transformer分类模型,取得了82.6%的分类准确度,证明了该模型在大规模数据集上具有竞争力。

基于多模态掩码Transformer网络的社会事件分类

多模态社会事件分类的关键是充分且准确地利用图像和文字2种模态的特征。然而,现有的大多数方法存在以下局限性:简单地将事件的图像特征和文本特征连接起来,不同模态之间存在不相关的上下文信息导致相互干扰。因此,仅仅考虑多模态数据模态间的关系是不够的,还要考虑模态之间不相关的上下文信息(即区域或单词)。为克服这些局限性,提出一种新颖的基于多模态掩码Transformer网络(MMTN)模型的社会事件分类方法。通过图-文编码网络来学习文本和图像的更好的表示。将获得的图像和文本表示输入多模态掩码Transformer网络来融合多模态信息,并通过计算多模态信息之间的相似性,对多模态信息的模态间的关系进行建模,掩盖模态之间的不相关上下文。在2个基准数据集上的大量实验表明:所提模型达到了最先进的性能。

基于EMDPWVD时频图像和改进ViT网络的滚动轴承智能故障诊断

滚动轴承是机械设备的关键零部件之一,其故障诊断对设备安全稳定运行至关重要。针对滚动轴承振动信号的非平稳特点,提出经验模态分解联合伪魏格纳分布(empirical mode decomposition&pseudo-Wigner-Ville distribution, EMDPWVD)时频图像联合改进Vision Transformer(ViT)网络模型的智能故障诊断新方法。首先针对实际信号研究短时傅里叶变换(short-time Fourier transform, STFT)、连续小波变换(continuous wavelet transform, CWT)和EMDPWVD三种时频分析方法,考虑STFT和CWT无法同时获得高的时间分辨率和频率分辨率,优选EMDPWVD作为智能故障诊断网络的时频图像构造方法。其次,以经典ViT作为故障诊断基础模型,将时频图像按照预定尺寸分块并线性映射为输入序列,通过自注意力机制整合图像全局信息,借助堆叠Transformer编码器完成网络传输,进而实现故障诊断。为进一步提高故障诊断准确率,将池化层作为ViT的预处理网络,获得改进的Pooling ViT(PiT)模型,实现时频图像的空间特征延展,提升模型对输入图像敏感度。结果表明,所提方法对滚动轴承不同故障类型均有高的诊断准确率,PiT较ViT的准确率提高4.40%,证明对ViT加入池化层能够实现滚动轴承故障诊断效果提升。

基于SwinT-YOLOX模型的自动扶梯行人安全检测算法

自动扶梯被广泛应用在公共场合,乘客摔倒事故如果不能被及时发现并处理,会造成严重的人身伤害,因此实现自动扶梯智能化监控管理势在必行。受自动扶梯运行环境复杂、行人多以及局部遮挡情况的影响,传统的人体姿态特征摔倒检测模型效果不佳且检测速度减慢。融合Swin Transformer和YOLOX目标检测算法的优秀策略,提出一种基于SwinT-YOLOX网络模型的自动扶梯行人摔倒检测算法。采用Swin Transformer模型作为骨干网络,颈部网络使用添加注意力机制的YOLOX模型,进一步提升特征图的多样性和表达能力。此外,利用漏斗修正线性单元视觉激活函数构建CBF模块,改进颈部网络和Head网络结构,从而获得更优的特征检测性能。实验结果表明,针对自建扶梯行人摔倒数据库和网络采集实际扶梯行人摔倒事故,与AlphaPose、OpenPose、YOLOv5等算法相比,该算法检测性能明显提高,行人摔倒平均检测精度可以达到95.92%,检测帧率为24.08帧/s,能够快速、精准地检测到乘客摔倒事故发生,监控管理平台立刻采取安全急停措施以保证乘客安全。

基于Transformer复杂运动辨识的机动星凸形扩展目标跟踪方法

针对复杂的机动扩展目标跟踪问题,利用Transformer网络设计了一种有效的星凸不规则形状机动扩展目标跟踪方法。首先,该文研究利用alpha-shape算法建立了星凸形状的变化模型,实现了静态场景下的星凸形扩展目标的形状估计。然后,通过对目标状态转移矩阵进行重新设计,结合Transformer网络对机动扩展目标运动状态转移矩阵进行实时估计,实现了对复杂机动目标运动过程的精准跟踪。进一步地,将估计得到的形状轮廓与运动状态进行融合,最终实现了对星凸形机动扩展目标的实时跟踪。最后,通过构造复杂的机动扩展目标跟踪场景,利用多重性能指标测试算法对形状和运动状态的综合估计性能,验证了算法的有效性。

Dual-Path Vision Transformer用于急性缺血性脑卒中辅助诊断

急性缺血性脑卒中是由于脑组织血液供应障碍导致的脑功能障碍,数字减影脑血管造影(DSA)是诊断脑血管疾病的金标准。基于患者的正面和侧面DSA图像,对急性缺血性脑卒中的治疗效果进行分级评估,构建基于Vision Transformer的双路径图像分类智能模型DPVF。为了提高辅助诊断速度,基于EdgeViT的轻量化设计思想进行了模型的构建;为了使模型保持轻量化的同时具有较高的精度,提出空间-通道自注意力模块,促进Transformer模型捕获更全面的特征信息,提高模型的表达能力;此外,对于DPVF的两分支的特征融合,构建交叉注意力模块对两分支输出进行交叉融合,促使模型提取更丰富的特征,从而提高模型表现。实验结果显示DPVF在测试集上的准确率达98.5%,满足实际需求。