近日，上海交通大学自然科学研究院/物理与天文学院/张江高等研究院洪亮课题组设计了一种微环境感知图神经网络（ProtLGN），能够从蛋白质三维结构中学习有益的氨基酸突变位点，建立自然选择下的氨基酸序列分布，用于指导蛋白质氨基酸位点设计，最终实现蛋白质指定功能的提升（定向进化）。课题组通过生物化学室实验证明，这是一个通用人工智能技术，依据目标蛋白极少实验数据甚至没有实验数据的情况下成功实现了多款蛋白指定性质的定向进化（包括抗体的亲和力/稳定性，多个荧光蛋白的荧光强度、核酸内切酶的DNA切割活性）。此外，ProtLGN能够根据单位点突变体活性准确预测多位点组合突变体活性，并在单轮湿实验中即可筛选出活性显著优于低位点突变体的高位点突变体，表明ProtLGN能够有效挖掘蛋白质定向进化中的正上位效应。因此，相较于传统的蛋白质设计方法，ProtLGN能够极大地提升蛋白质设计效率并降低实验成本。该研究成果以“Protein Engineering with Lightweight Graph Denoising Neural Networks”发表于美国化学学会出版社下Journal of Chemical Information and Modeling。人工智能的快速发展改变着生命科学领域的研究方法和思维范式。其中，蛋白质设计作为生物医药领域的关键技术之一，受到了学界和工业界的广泛关注。传统的蛋白质设计方法通常依赖于试错（蛋白质定向进化）和经验积累（蛋白质理性设计），这种方式虽然在一定程度上能够取得成功，但也存在着效率低下、成本高昂、时间耗费长等难以解决的问题。随着深度学习的快速发展和广泛应用，为蛋白质设计领域注入了新的活力和希望。深度学习模型可以通过大量的生物信息学数据进行训练，并从中学习蛋白质序列、结构与功能之间的关联，为蛋白质设计提供准确、高效的指导。近几年，基于深度学习的预测与筛选方法被提出并在实际应用中得到了验证与应用。然而，当前的方法大多是基于多序列比对（MSA）和/或蛋白质语言模型（PLM）对蛋白质序列进行特征提取。前者提取出的蛋白质共进化信息的质量高度依赖同源信息的数量，但在实际应用中，并非所有蛋白质序列都能进行同源比对，也并非所有比对都足够深以训练足够大的模型以学习氨基酸之间的复杂相互作用。后者来源于自然语言处理，因此模型主体通常需要搭建Transformer，递归神经网络，或其他自回归模型，这类方法通常需要海量的训练数据和大型且复杂的模型设计，从而需要非常高的训练成本。即使是使用当前主流的语言预训练模型的思路，考虑到每个蛋白质都有独特的性质和进化方向，不涉及任何的重新训练路径而直接把一个预训练好的模型推广到任意的任务中，对于大模型的泛化性和表达能力也是一个巨大的挑战。因此，本文设计了一款能够提取氨基酸周围微观环境信息的等变图神经网络的预训练框架ProtLGN，结合蛋白质的结构信息对一条蛋白质上的每个氨基酸进行同步编码，学习蛋白质三维结构中有益的氨基酸突变位点和突变类型，用于指导具有不同功能的蛋白质单位点突变和多位点突变设计。 查看详细>>

近日，上海交通大学电子信息与电气工程学院计算机科学与工程系人工智能教育部重点实验室盛斌教授团队，上海交通大学医学院附属第六人民医院的贾伟平教授和李华婷教授团队，以及清华大学黄天荫教授团队，汇聚来自美国、英国、马来西亚和新加坡等国的多学科专家，共同在权威学术刊物《Science Bulletin》上发表了题为“Large language models for diabetes care:Potentials and prospects”（面向糖尿病诊疗的大语言模型：潜力与前景）的展望性文章。该文章全面分析了大语言模型在糖尿病管理领域的应用，并对其未来发展前景进行了深入展望。糖尿病作为21世纪全球性的重大公共卫生议题，其影响日益扩大。据2021年的统计，全球已有5.37亿人罹患此病，且预计到2045年，这一数字将攀升至7.83亿。面对如此庞大的患者群体，糖尿病的管理面临诸多挑战，包括患者教育、实时病情监测等多个层面。大语言模型（Large Language Models,LLM），作为一种先进的生成式人工智能技术，能够接收图像和文本输入，并输出相应的文本。随着大语言模型的进一步研发与应用，我们有理由期待其为糖尿病患者和医务工作者提供数据驱动、个性化和实时的诊疗支持，从而改善糖尿病患者的健康状况和生活质量。该文详尽地探讨了大语言模型在糖尿病管理中的潜在应用及其对患者健康结局的积极影响。在糖尿病管理中，大语言模型通过健康教育、营养指导、血糖监测数据管理、用药依从性监测、并发症监测以及实时心理咨询等方式，为患者提供了持续且个性化的支持。同时，这些模型也为医务人员提供了糖尿病管理培训，有效简化了临床任务。随后，文章全面总结了当前大语言模型在糖尿病管理中面临的挑战，这些挑战涵盖了专业性、时效性、安全性、个性化需求、伦理道德、法律法规以及隐私保护等多个方面。最后，文章展望了未来的研究方向，呼吁以糖尿病专业知识为基础对现有大语言模型进行微调，探索与网络实时连接的大语言模型，建立公开的测评数据集，制定相应的伦理和法律框架，开展随机对照试验，以及将大语言模型与现有医疗体系和数字医疗技术相结合。这些努力有望进一步推动大语言模型在糖尿病管理领域的应用与发展。 查看详细>>

The radiation detectors used today for applications like inspecting cargo ships for smuggled nuclear materials are expensive and cannot operate in harsh environments,among other disadvantages.Now,in work funded largely by the U.S.Department of Homeland Security with early support from the U.S.Department of Energy,MIT engineers have demonstrated afundamentally new way to detect radiation that could allow much cheaper detectors and aplethora of new applications.They are working with Radiation Monitoring Devices,a company in Watertown,Massachusetts,to transfer the research as quickly as possible into detector products.In a2022 paper in Nature Materials,many of the same engineers reported for the first time how ultraviolet light can significantly improve the performance of fuel cells and other devices based on the movement of charged atoms,rather than those atoms’constituent electrons.In the current work,published recently in Advanced Materials,the team shows that the same concept can be extended to anew application:the detection of gamma rays emitted by the radioactive decay of nuclear materials.“Our approach involves materials and mechanisms very different than those in presently used detectors,with potentially enormous benefits in terms of reduced cost,ability to operate under harsh conditions,and simplified processing,”says Harry L.Tuller,the R.P.Simmons Professor of Ceramics and Electronic Materials in MIT’s Department of Materials Science and Engineering(DMSE). 查看详细>>

A collaborated research team led by The Hong Kong University of Science and Technology(HKUST)and Tsinghua University has theoretically proposed anew mechanism of electrical 180°switching of Néel vector and experimentally realized it in antiferromagnetic materials with spin-splitting band structure featuring the C-paired spin-valley locking,also named as altermagnet.The team also demonstrated the material‘s capability to manipulate Néel vector,paving the way for the manufacturing of ultrafast memory devices.Antiferromagnetic spintronics has sparked widespread interest due to its enormous potential for creating ultra-dense and ultrafast antiferromagnetic memory that is suitable for modern high-performance information technologies.The electrical 180°switching of the Néel vector is along-term objective for producing electrically controllable antiferromagnetic memory using opposite Néel vectors as binary"0"and"1".However,the state-of-art antiferromagnetic switching mechanisms have long been limited for 90°or 120°switching of Néel vector,which unavoidably requires multiple writing channels that contradict ultradense integration.The study of electrical 180°switching of Néel vector makes spin-splitting antiferromagnet anew potential candidate for ultrafast memory.Specifically,in collinear antiferromagnet,the Néel vector nhave two stable states n_+and n_-with symmetric energy barriers.To leave an asymmetry of energy barriers,the team led by Prof.LIU Junwei,Associate Professor at the Department of Physics at HKUST proposed to exert an external magnetic field to interact with the tiny DMI-induced moment.Then,the damping-like spin-orbit torque[2]can be used to drive Néel vector nto cross the barrier from n_+to n_-but cannot cross the opposite one(Figure 1a).As shown in Figure 1b,the atomic spin model simulation shows that ncan be deterministically switched to state n_+or n_-in 0.1 ns.Integrating the non-zero Berry curvatures on spin-splitting bands of tight-binding model,the anomalous Hall conductivities show high sensitivity to these two states n_+and n_-,shown in Figure 1c.In experiments led by Prof.PAN Feng and Prof.SONG Cheng,from the School of Materials Science and Engineering at Tsinghua University,the good cyclic performance of fabricated antiferromagnetic Mn5Si3 thin film is shown in Figure 1d,which means the current-driven 180°switching of Néel vector is robust and sustainable.In fact,the team had presented anew theory as C-paired spin-valley locking(SVL)a few years ago in the scientific journal Nature Communications,indicating anew way to induce the magnetization in antiferromagnet and laying the foundation of switching of Néel Vector.Compared to this conventional T-paired SVL materials,the C-paired SVL materials create the spin-splitting bands by the strong exchange coupling between itinerant electrons and local magnetic moments instead of SOC.Furthermore,the spin-splitting valleys are paired with opposite spin directions by preserved crystal symmetry rather than time-reversal symmetry,as shown in Figure 2.In practice,a strain/charge current can be exerted to slightly break or affect the crystal symmetry and therefore induce anet magnetization/noncollinear spin current. 查看详细>>

近日，上海交通大学船舶海洋与建筑工程学院田新亮研究小组完成了“软尾减阻”的三维风洞实验，并发现了高达22.2%的减阻效果。该研究成果以“Drag reduction using aself-adaptive flexible coating”为题，在线发表在流体力学领域权威期刊《流体物理》（Physics of Fluids 36,041701(2024)），并被选为编辑精选文章。2020年，该小组首次提出了为钝体穿“裙子”来减阻的方法（简称“软尾减阻”），即通过柔性薄膜包裹钝体使其“伪装”成流线体从而实现减阻[1,2]。前期肥皂膜水洞实验证明了二维流场中最高约为10%的减阻效果。此后，国内外多个研究团队[3-5]及该小组[6,7]分别独立采用数值模拟方法，深入分析了二维场景下“软尾减阻”的物理机制和多种参数的影响。但是，该减阻方法能否用于实际工程，一直未获直接回答。从2020年开始，研究小组在船海工程试验中心风洞循环水槽的支持下，逐步证明了“软尾减阻”方法的工程实用性。首先在风洞实验室中构造了“软尾减阻”的三维实验模型，即“刚性平板+柔性薄膜”耦合系统。然后，建立了一套精密的测量系统（图1）：设计了悬挂系统释放约束来提升测力精度，开发了基于激光和高速摄像机的柔性薄膜高时空分辨率测量方法，研制了尾流烟线显示方法和装置。为了解决薄膜在大风中被吹瘪或吹破的问题，研究小组还提出了在平板正面开若干小孔来平衡薄膜内外压强的方法，不仅缓解了薄膜在风载作用下所受的张力，还使得薄膜在任何风速下均可达到鼓胀的效果。 查看详细>>

近日，上海交通大学环境科学与工程学院朱樱副教授在环境领域著名学术期刊Environmental Science&Technology上发表了题为“Comprehensive Assessment of Environmental Emissions,Fate,and Risks of Veterinary Antibiotics in China:An Environmental Fate Modeling Approach”的研究论文。该研究结合实证数据和模型方法，率先阐释了我国2010-2020年间源于畜禽和水产养殖的兽用抗生素排放和多介质浓度的时空演变规律和多介质源汇特征，明晰了环境耐药性演化风险，首次通过纳入兽用抗生素随粪便处理的去除率，揭示了我国畜禽及水产养殖量区域分布与集约化养殖、家庭养殖方式的时空变化对抗生素排放的驱动机制。论文第一作者为上海交通大学环境科学与工程学院2021级研究生李帅奇，通讯作者为朱樱副教授，第一完成单位和通讯单位均为上海交通大学。抗生素耐药性是全球人类健康的主要威胁之一。抗生素的环境释放和污染可能促进抗生素耐药性发展和传播，这增加了人类被含耐药基因的病原体感染的可能性，进而降低医用抗生素药效。我国是全球主要抗生素消费国之一。随着收入的增加，人们对动物蛋白质需求提高，导致兽用抗生素的使用量不断增加。自2016年以来，我国制订了多种兽用抗生素使用管理政策，并将抗生素作为高关注新污染物进行环境污染控制。因此，有必要了解我国不同环境介质中兽用抗生素的排放和污染状况，以应对我国和全球范围内的耐药性发展。然而已有关于我国抗生素排放、环境迁移和分布的研究仍存在局限性。同时，包括我国在内的全球研究主要集中在调查兽用抗生素的使用情况，但很少评估环境释放和污染。这些阻碍了对我国兽用抗生素环境污染暴露水平和归趋的认识。该研究率先结合实证数据和模型方法，对我国2010-2020年间使用的兽用抗生素种类、环境排放量、多介质浓度及分布和环境耐药性演化风险做了全面的调查、估算和评价。使用我国国家尺度多介质环境归趋模型（SESAMe v3.4模型，空间分辨率0.5°），对我国兽用抗生素的多介质浓度进行预测。与其他传统多介质环境归趋模型相比，该模型充分考虑了化学物质解离的特性，并且具备相对完整的跨介质迁移过程。本研究结果可为我国兽用抗生素的污染防治提供支持。同样的研究体系可在全球范围内应用，尤其可用于面临严重抗生素排放和污染问题的国家。 查看详细>>

Gas vesicles are hollow structures made of protein found in the cells of certain microorganisms,and researchers believe they can be programmed for use in biomedical applications.“Inside cells,gas vesicles are packed in abeautiful honeycomb pattern.How this pattern is formed has never been thoroughly understood.We are presenting the first identification of aprotein that is responsible for this pattern,and we believe it can be very useful,”said George Lu,an assistant professor of bioengineering and aCancer Prevention and Research Institute of Texas scholar at Rice University.Lu and colleagues from Washington University in St.Louis and Duke University have shared their findings in apaper published in Nature Microbiology.Co-author Yifan Dai,an assistant professor of biomedical engineering at WashU’s McKelvey School of Engineering,said they were drawn to the research with the question of why the vesicles can form in the honeycomb pattern.With help from WashU colleague Alex Holehouse along with Ashutosh Chilkoti and Lingchong You,of Duke University,the team found this pattern is the most efficient use of space,and the cluster form plays apart in how it functions.Notably,these protein clusters formed in subsaturated solution,a previously identified new form of biological structure,and that drives the organization of these vesicles.Bottom line,they found the function behind this mysterious new form. 查看详细>>

A collaboration of researchers including UBC scientists have observed gravitational waves from the collision of what is most likely aneutron star and an object likely to be alight black hole,650 million light-years from Earth.The mass of the black hole is 2.5 to 4.5 times the mass of Earth’s sun,meaning it falls in the so-called‘mass gap’:heavier than heaviest known and theorized neutron stars but lighter than the lightest black holes in our galaxy.The collision,reported in apreprint paper,was detected by one part of an international network of gravitational wave detectors,comprised of the Laser Interferometer Gravitational-Wave Observatory(LIGO),the Virgo Gravitational Wave Interferometer,and the Kamioka Gravitational Wave Detector(KAGRA).“This detection,the first of our exciting results from the fourth LIGO-Virgo-KAGRA observing run,reveals that there may be ahigher rate of similar collisions between neutron stars and low mass black holes than we previously thought,”says Dr.Jess McIver,assistant professor at UBC and deputy spokesperson of the LIGO scientific collaboration. 查看详细>>

在国家“双碳”战略需求的背景下，钠离子电池凭借钠资源的低成本优势有望在未来规模储能领域实现广泛应用。作为限制钠离子电池能量密度的关键部分，正极材料的能量密度和循环寿命一直备受广大科研工作者的关注。典型的层状氧化物正极材料在实际应用过程中一直面临着钠空位有序和大体积相变的基础科学问题，这直接了限制了这类正极材料的电化学性能和能量密度。鉴于此，西安交通大学王鹏飞教授课题组提出利用Li+、Mg2+、Ti4+三种特定功能离子的协同效应有效解决了P2型正极材料低电压区域钠空位有序重排和高电压区域大体积相变的难题，设计的正极材料工作电压为3.57 V，可逆比容量达到134 mAh g–1，与硬碳负极匹配的全电池能量密度达到296 Wh kg–1。同时，王鹏飞教授团队通过高通量合成实验绘制层状相的热力学稳定相图，研制出一种P2和O3双相结构共生的新型层状正极材料，两相界面的“互锁效应”有效减小了材料充放电过程中结构应力和晶格错位概率，缓解了大尺寸钠离子脱嵌过程中的相变体积变化，得到了一种可逆比容量为144 mA hg–1，能量密度高达514 Wh kg–1的新型钠离子电池正极材料。关研究成果以《多离子协同抑制P2型正极大体积相变提高钠离子电池性能》（Mitigating the Large-Volume Phase Transition of P2-Type Cathodes by Synergetic Effect of Multiple Ions for Improved Sodium-Ion Batteries）为题于2月21日在线发表在《先进能源材料》（Advanced Energy Materials）上。同时以《一种合理的双相调控策略实现高性能钠离子电池层状正极》（A Rational Biphasic Tailoring Strategy Enabling High-Performance Layered Cathodes for Sodium-Ion Batteries）为题于3月1日在线发表在《德国应用化学》（Angewandte Chemie）上。论文第一作者均为电气学院承志伟博士，西安交通大学王鹏飞教授、肖冰教授、滑纬博研究员以及中科院化学研究所郭玉国研究员为共同通讯作者，论文第一单位均为电工材料电气绝缘全国重点实验室新型储能与能量转换纳米材料研究中心。 查看详细>>

Submarines and ships rely on towed sonar arrays(TSAs)for underwater exploration and security operations,but dragging these sensors through water,especially at high cruising speeds,creates excess noise that can mask target signals and compromise the sonar’s detection capabilities.Now,a team of Berkeley engineers is attempting to solve this problem with alittle inspiration from Mother Nature.In astudy recently published in Extreme Mechanics Letters,researchers from Berkeley,in collaboration with MIT Lincoln Laboratory,demonstrate how atextured surface designed to mimic shark skin can reduce drag and mitigate flow-based noise,potentially opening the door to anew generation of more effective and efficient TSAs.“Past studies have shown that the unique patterns on shark skin,known as riblets,can reduce drag,”said Grace Gu,assistant professor of mechanical engineering and principal investigator of this study.“We hypothesized that such bioinspired topographies could similarly reduce noise production in an underwater context and set out to adapt these naturally occurring designs to the surfaces of sonar arrays.”According to the researchers,the idea of using riblets for noise suppression was inspired by yet another natural texture,serrations,which are known to help owls achieve silent flight and used on surfaces to mitigate noise in aeronautic systems.To test their theory,the researchers used computational modeling to simulate riblet surfaces with different shapes and patterns and then simulated the movement of water around these engineered riblet surfaces on the arrays.This virtual testing environment enabled them to assess how different riblet designs affect fluid behavior and the acoustic properties of the sonar arrays under various flow conditions—from smooth,predictable laminar flows to the more common,chaotic turbulent flows encountered in natural marine environments. 查看详细>>