banner
October 28, 2022

MSR Asia Science Forum

The MSR Asia Science Forum provides a series of seminar talks on natural sciences by invited scientists. ​It is a platform for scientists from diversified areas (e.g., Physics, Chemistry, Material, Biology and Environment etc.) to communicate and exchange ideas with MSR researchers and engineers.

The Science Forum plays a crucial role in bridging computer scientists and natural scientists. It helps computer scientists to understand the pressing challenges in natural sciences and help scientists in natural science to see how AI can be an integral part of their exploration.

The Science Forum is organized on a bi-weekly basis, with a lecture from a different discipline followed by a group discussion. We hope it brings an exciting time for researchers to work at the intersection of machine learning and the natural sciences!


微软亚洲科学论坛是由微软研究院举办的自然科学讲座系列,主要目的是为来自不同领域(如物理、化学、材料、生物、环境等) 的科学家、微软研究员以及工程师提供一个思想碰撞的平台。

科学论坛是连接计算机科学家和自然科学家的重要桥梁。科学论坛将每两周举办一次,邀请自然科学领域的科学家演讲并开展小组讨论。通过演讲和探讨,我们希望该论坛能够扩大微软研究院在科学研究领域的影响,促进更广泛的合作,更好地发挥微软研究院的计算科学和人工智能技术优势,推进科学发展。

关于AI4Science:
科学智能中心(AI4Science)是微软研究院(Microsoft Research)于2022年7月成立的研究团队,致力于机器学习和自然科学交叉领域的基础性研究,通过技术创新改变人类理解自然世界以及与自然世界互动的方式,展现机器学习与自然科学交叉融合的新能力。​科学智能中心是一个全球性的团队,研究小组分布在自英国、中国和荷兰等多个国家。科学智能中心的研究团队由机器学习、计算物理、计算化学、分子生物学、软件开发等领域的世界级专家组成,同心协力解决该领域中一些最紧迫的挑战。


Talks

10/28/2022: HPC+AI: large scale molecular dynamics with ab initio accuracy for material science simulation (超级计算机+人工智能助力量子化学精度的大尺寸体系分子动力学模拟), Dr. Weile Jia

  • Abstract: Scientific simulation has become the third polar in the research of material sciences. Besides traditional scientific computing, the emerging AI for science is a new trend in solving high dimensional physical problems such as ab initio molecular dynamics. The system scale and mixed precision of modern supercomputers poses great challenges to the efficiency of the novel HPC+AI methods. In this presentation, I will introduce the algorithms and optimization both traditional scientific computing and HPC+AI methods on large-scale supercomputers, especially on exploiting the performance of entire supercomputer such as Summit, Fugaku and Sunway.  The corresponding challenges for the computer architecture will be from the perspective of large-scale calculation of materials sciences.

    Bio: Weile Jia(贾伟乐) Associate professor, Institute of Computing Technology, Chinese Academy of Sciences.  My research focuses on high-performance computing, especially in efficiently carrying out density functional theory calculation with both numerical and HPC+AI methods. I am now an Associate Professor at Institute of Computing Technology, Chinese Academy of Sciences. I am very fortunate to have received the ACM Gordon Bell prize (2020). Our recent work on large-scale simulation for complex heterostructure has been selected as one of the 2022 ACM Gordon Bell Finalists.

    poster

10/25/2022: A tale of two viruses –  the past and future of phantom hunting (双魔记 – 降魔之路的过去与未来), Dr. Lei Wei

  • Abstract: Where there is life, there are viruses. Numerous pathogenic viruses have caused tremendous loss to human lives throughout history and will continuing doing so for the foreseeable future. The study of virology is key to equip us with knowledge to fight against these viruses. One most recent success is the cure of hepatitis C virus infection, which is the first chronic viral infection that can be cured. Another half success is the control of hepatitis B virus infection. I will use these two human viruses as examples to describe the history and future of virology development, I will also talk about how data and computational analyses can revolutionize the virology field and our battle against viral infections.

    Bio: Lei Wei is currently an assistant professor at Westlake University, Center for Infectious Diseases Research. His research interest is virus-host interactions, one of his major focuses is how to cure human hepatitis B virus infection. He received his bachelor’s and master’s degrees in Biological Sciences at Tsinghua University. He obtained his doctoral degree in molecular biology and Genetics in Dr. Xiaolan Zhao’s lab in Memorial Sloan-Kettering Cancer Center, USA. His doctoral research was focused on DNA replication and repair. He then developed an interest in the interface of DNA repair and HBV virus and joined Alexander Ploss’ lab at Princeton University as a postdoc to investigate the replication of HBV genome. He is now continuing his virology research at Westlake University.

    poster

10/12/2022: Modulating the local atomic structure of electrocatalyst materials for clean energy revolution (新一代清洁能源电催化材料的局部原子结构调控), Prof. Zhao Cai

  • Abstract: Water splitting hydrogen production and carbon dioxide reduction to value-added fuels are both pivotal technologies to address the global energy crisis and environmental issues. The rational design and controllable fabrication of advanced catalyst materials holds the key to lower the overpotential and increase the energy efficiency of these electrochemical reactions. Here we focus on modulating the local structure of electrocatalyst materials at the atomic level. By means of introducing vacancy defects and regulating metal-oxygen motifs, the intrinsic activity and energy conversion efficiency of hydrogen/oxygen evolution electrocatalysts are significantly improved. Meanwhile, by enabling win-win metal-oxide sites cooperation, unprecedented catalytic functionalities of potential-dependent bifunctional CO2 conversion are achieved. Such strategy suggests the importance of local structure on tuning the surface coordination environment for electrocatalysis optimization, paving a new way for the development of advanced electrocatalysts with favorable atomic arrangement.

    Bio: Prof. Zhao Cai gained his B.S. degree and Ph.D in Chemistry from Beijing University of Chemical Technology in 2012 and 2018, respectively. After working as a visiting scholar at Yale University and postdoctoral researcher at Wuhan National Laboratory for Optoelectronics, he joined China University of Geosciences in 2022. His research focuses on developing novel transition metal nanostructures for key energy conversion and storage processes, such as electrocatalysis and aqueous batteries. He has authored and co-authored more than 40 journal articles including Nat. Commun., J. Am. Chem. Soc., Angew. Chem. Int. Ed., Adv. Energy Mater. etc. with a total citation for over 3000 times and one fourth of them are ESI highly cited papers.

    text

09/20/2022: What can CHARMM-GUI do for you?(CHARMM-GUI 能为我们做什么?), Dr. Wonpil Im

  • Abstract: Since its original development in 2006, CHARMM-GUI has proven to be an ideal web-based platform to interactively build complex molecular systems and prepare their simulation inputs with well-established and reproducible simulation protocols for state-of-the-art molecular simulations using widely used simulation packages. The CHARMM-GUI development project has been widely adopted for various purposes and now contains a number of different modules designed to set up a broad range of molecular simulation systems. Our philosophy in CHARMM-GUI development is less about providing the nuts and bolts of molecular modeling, but instead focused on helping users to achieve a task, such as building a membrane system or solvating a protein, by providing a streamlined interface. This design principle helps us to think of the workflow critically when designing the interface, which leads CHARMM-GUI to be accessible to users with little experience in modeling tools and remains useful to experts, especially for batch generation of systems. The CHARMM-GUI development project is still ongoing. CHARMM-GUI will continue to help expert and non-expert researchers from a broader range of the modeling and simulation community to build the complex molecular systems of their interest and prepare the input files for any general and advanced modeling and simulation through the large and unique scope of CHARMM-GUI functionality, allowing the research community to carry out innovative and novel molecular modeling and simulation research. In this talk, I will present the past, present, and future of the CHARMM-GUI development project, and some applications for specific modules will be also discussed.

    Bio: Wonpil Im received in bachelor’s and master’s degrees from Hanyang University in Seoul. He then earned his Ph.D. in Biochemistry from Cornell University. He did his post-doctoral research at the Scripps Research Institute in La Jolla, California. In 2005, he was hired as an assistant professor in the Center for Computational Biology and Department of Molecular Biosciences at the University of Kansas, Lawrence.  In 2011, he was promoted to associate professor and then professor in 2015. In 2016, he joined the Faculty in Departments of Biological Sciences and Bioengineering at Lehigh University, and he has been named the Presidential Endowed Chair in Health – Science and Engineering. Wonpil was awarded the Alfred P. Sloan Research Fellowship (2007), ACS HP Outstanding Junior Faculty Award (2011), J. Michael Young Undergrad Advisor Award (2011), Meredith Docking Scholar (2013), and University Scholarly Achievement Award (2015). the Friedrich Wilhelm Bessel Research Award from the Humboldt Foundation (2017), Lehigh CAS Dean’s Research Award (2019),  Libsch Research Award (2021), and was named a KIAS Scholar from the Korea Institute for Advanced Study (2016).

    Research in his lab is focused on the applications of theoretical/computational methods to chemical and physical problems in biology and material sciences. In particular, he is interested in modeling and simulations of biological membranes and associated proteins, glycoconjugates, and protein-ligand (drug) interactions. In addition, his lab has been developing CHARMM-GUI for the biomolecular modeling and simulation community.

    poster

09/13/2022: Epigenomics in health and diseases: recent advances and computational challenges(健康和疾病研究中的表观基因组学:最新进展和计算挑战), Dr. Yanxiao Zhang

  • Abstract: As we all know, DNA is the basis for inheritance. Epiegenetics is the study of heritable phenotype changes that do NOT involve alterations in the DNA sequences. Epigenetics, by the broad definition, includes DNA methylation, histone modifications, noncoding RNA and so on. They act together to dictate the fate of cells, i.e., how stem cells differentiate into hundreds of different cell types in our body. When these processes are compromised, it could result in a variety of diseases, including developmental diseases, cancer, and aging, to name a few. In this talk, I will introduce some mainstream genomics technologies that could help us measure the epigenetic states of our cells, their data analysis strategies, and their application in research. I will also try to highlight several computational challenges in the field.

    Bio: Dr. Yanxiao Zhang received his bachelor’s degree in Biotechnology from Beijing University in 2010. He then obtained a PhD degree in Bioinformatics (advisor: Dr. Maureen Sartor) and master’s degree in Statistics from the University of Michigan in 2016. He completed his postdoctoral work in Dr. Bing Ren’s lab at Ludwig Institute for Cancer Research (UCSD branch), studying the dynamics of chromatin states during mammalian development (as part of ENCODE) and 3-dimenstional chromatin structure. Dr. Zhang joined Westlake University as an Assistant Professor in January 2022. His lab will use computational and functional genomics tools to study the mechanisms of aging and cancer, with a primary focus on epigenetic regulations. For more info please visit: https://zhangyxlab.github.io/

    poster

08/30/2022: Visual information decoding and restoration(视觉信号的解码与再现), Dr. Jiayi Zhang

  • Abstract: Jiayi Zhang has utilized inter-disciplinary approaches to develop novel technologies for decoding visual information and treating neural (visual in particular) disorders. The speaker will touch on the following studies: 1) Restoration of vision in blind mice using titanium oxide nanowires as artificial photoreceptors. 2) Development of a multisite optogenetic stimulation device, which has the potential to improve rehabilitation of paralyzed upper limb. 3) a visual camouflage system that changes colors of electroluminescent fibers in real-time response to the color change in the environment. 4) Near-infrared manipulation of multiple neuronal populations via trichromatic upconversion. 5) Coding of timing information in the visual cortex. Jiayi Zhang has published multiple papers in journals including Neuron, Advanced Materials and Nature Communications (corresponding author) in the past five years, which was cited for over 1000 times.

    Bio: Dr. Jiayi Zhang received her B. Sc. Degree from Hong Kong Baptist University and Ph.D. degree from Brown University. She was a Brown-Coxe postdoctoral fellow in Yale University and joined Institutes of Brain Science at Fudan University in 2012. She is currently the vice director of State Key Laboratory of Medical Neurobiology. Her recent work focused on the decoding and restoration of vision. Her work was published in journals including Neuron, Advanced Materials and Nature Communications. She got the Young Innovative Woman Award in Shanghai in 2020. She serves as the Vice chairman of the Young Scholar Panel and fellow for Chinese Association for Physiological Sciences (CAPS) as well as the Vice chairman of the Sensory and Motor Panel, Chinese Neuroscience Society (CNS).

    poster

08/11/2022: Orbital-free Density Functional Theory: Successes, Challenges, and Opportunities(无轨道密度泛函理论:成功、挑战与机遇), Dr. W. Chuck Witt

  • Abstract: The computational materials scientist regularly faces tradeoffs between accuracy and computational expense. Density functional theory (DFT) has become an enormously popular tool because it retains much of the accuracy of quantum mechanics at modest cost. DFT calculations—on topics as wide-ranging as battery design and the nature of planetary cores—contribute to thousands of scientific papers and consume a significant fraction of worldwide supercomputing time each year. Orbital-free DFT, which bypasses all wave function operations, achieves further large efficiency gains over standard DFT; however, the orbital-free variant is presently less accurate for much of the periodic table. This talk will introduce orbital-free DFT in a manner accessible to an interdisciplinary audience, highlighting successes, challenges, and opportunities for improvement.

    Bio: Dr. Chuck Witt is a Junior Research Fellow in the Materials Theory Group at the University of Cambridge, as well as a Schmidt Science Fellow. His interests include the development of methods and software rooted in quantum-mechanical first principles, as well as their application to problems in materials science and engineering. He earned a PhD in Mechanical and Aerospace Engineering from Princeton University.

    poster

08/08/2022: Deep learning based insights into key bacterial pathways(关键细菌通路的深度学习探索), Prof. Jeffrey Skolnick

  • Abstract: Accurate descriptions of protein-protein interactions are essential for understanding biological systems. Remarkably accurate atomic structures have been recently computed for individual proteins by AlphaFold2 (AF2). We demonstrate that the same neural network models from AF2 developed for single protein sequences can be adapted to predict the structures of multimeric protein complexes without retraining. In contrast to common approaches, our method, AF2Complex, does not require paired multiple sequence alignments. It achieves higher accuracy than complex protein-protein docking strategies and provides a significant improvement over AF-Multimer, a new development of AlphaFold for multimeric proteins. Moreover, we introduce metrics for predicting direct protein-protein interactions between arbitrary protein pairs and validate AF2Complex on challenging benchmark sets and the E. coli proteome. Using the cytochrome c biogenesis system I as an example, we present high-confidence models of three sought-after assemblies formed by eight members of this system. Then, we introduce a high-throughput, deep learning pipeline to identify PPIs within the E. coli cell envelope and apply it to several proteins from an OMP biogenesis pathway. Thus, we have developed a general strategy for investigating biological pathways by using structural insights gained from deep learning-based predictions.

    Bio: Jeffrey Skolnick is a Regents’ Professor and Director of the Center for the Study of Systems Biology in the School of Biological Sciences at the Georgia Institute of Technology. He is also the Mary and Maisie Gibson Chair in Computational Systems Biology and a Georgia Research Alliance Eminent Scholar in Computational Systems Biology. He received a Ph.D. in Chemistry from Yale University. Among his awards are the Southeastern Universities Research Association (SURA) Distinguished Scientist Award, the Sigma Xi Sustained Research Award, and an Alfred P. Sloan Research Fellowship. He is a Fellow of the AAAS, the Biophysical Society, and the St. Louis Academy of Science. He has authored over 400 publications, has an h-index of 92, and has served on 16 editorial boards. Dr. Skolnick’s current research interests are in computational biology and bioinformatics with a major emphasis on precision medicine and drug discovery. He has developed and applied AI approaches to predict disease mode of action proteins, the prediction of drug efficacy and side effects and studies on the interrelationship of all human diseases. He has applied these ideas to predict the origin and possible treatments of long COVID, aging, cancer, and chronic fatigue syndrome. He has also done substantial research on the possible origins of the biochemistry of life and molecular simulations of subcellular processes.

    poster

08/02/2022: Jet Tagging in the Era of Deep Learning(深度学习时代的粒子对撞研究), Dr. Huilin Qu

  • Abstract: Machine learning has revolutionized the analysis of large-scale data samples in particle physics and greatly increased the discovery potential for new fundamental laws of nature. Specifically, deep learning has transformed how jet tagging, a critical classification task at high-energy particle colliders such as the CERN Large Hadron Collider (LHC), is performed, leading to a drastic improvement in its performance in the past few years. In this talk, Dr. Qu will go through recent progress in deep learning approaches for jet tagging and their applications in Higgs boson studies and new physics searches at the LHC. Prospects and possible future directions will also be discussed.

    Bio: Dr. Huilin Qu is a Senior Research Fellow at CERN. He received his B.S. degree from Peking University in 2014, and Ph.D. from University of California, Santa Barbara in 2019. His research has focused on searches for new physics and measurements of the Higgs boson properties with the CMS experiment at the CERN LHC, particularly using novel approaches and advanced techniques such as machine learning. He played a key role in searches for the Higgs boson decaying to a pair of charm quarks, for Higgs boson pair production in the high-momentum regime, and for supersymmetric partners of the top quark. In addition, Huilin is active in machine learning research for jet physics. He proposed a series of novel deep-learning approaches for jet tagging, which substantially improved the performance and have been widely adopted at the LHC and beyond.

    poster

07/20/2022: Paving a data-centric high-way for materials science(数据驱动材料科学), Prof. Miao Liu

  • Abstract: In this talk, the speaker will be presenting how he takes all the technologies that have disrupted other industries and start to disrupt his own. Leveraging the recent advance in the high-throughput first-principles calculations, his team has recently released a world-class materials database, namely atomly.net, to bring the high-quality materials science data to the fingertips of scientists worldwide. Atomly.net contains the high quality first-principles calculation results of more than 300k inorganic materials and many related properties, including but not limited to the crystal structure, electronic structure, elasticity, dielectric tensor, etc.
    Taking the advantage of the massive trove of data, it becomes feasible to screen, predict, and discover new materials in an expedited and cost-effective manner. For example, identify the possible Kagome materials [Chin. Phys. Lett. 39 047402 (2022)], screen the feasible superconductor [Phys. Rev. B 105, 214517 (2022)], and build generic artifactual intelligent models for property prediction [arxiv.org/abs/2108.00349].

    Bio: Miao Liu is an associate professor at the Institute of Physics, Chinese Academy of Sciences, and a joint scientist at the Songshan Lake Materials Laboratory. His research revolves around data-driven materials science for energy materials, alloys, quantum materials, etc. Dr. Liu was trained as a physics major at the beginning and got his Ph.D. from the University of Utah in Materials Science and Engineering afterward. He then joined the Materials Project team at Lawrence Berkeley National Laboratory as a postdoctoral researcher and developed the behind scene database management modules. He started his career at the Institute of Physics in the year 2018 and found the Atomly materials science database from scratch. Atomly now is one of the world’s foremost databases of its kind.

    poster

07/13/2022: 3-D Structure-based drug discovery — Rational design of small molecules, proteins and viruses for therapeutic industries(三维结构药物发现), Prof. Xiao-Dong Su

  • Abstract: There are several seminal advances in the biomedical fields in the recent 10 years that signal some major transition may take place in drug discovery research. The transition includes traditional small organic and protein molecule base drugs and therapies towards cellular and vector-based gene therapies. I will review mostly from 3-D structural viewpoints some of the traditional aspects of current drug discovery to future cellular and AAV (adeno associated viruses) vector-based gene delivery and therapies.

    Bio:
    2003.01 – present, Professor of Biochemistry and Molecular Biology (Cheung Kong Scholar), School of Life Sciences, Peking University, Beijing, P. R. China.
    2001.11 – 2002.12, Associate Professor at Department of Molecular Biophysics, Center for Chemistry and Chemical Engineering, Lund University, Sweden.
    1998.04 – 2002.04, Assistant professor at Dept. of Molecular Biophysics, Center for Chemistry and Chemical Engineering, Lund University, Sweden.
    1995.07 –1998.04, Research associate with Howard Hughes Medical Institute (HHMI) at Division of Biology, California Institute of Technology, USA.
    1988 .11- 1994.12, Ph.D. candidate at the Department of Cell and Molecular Biology, Karolinska Institute, Sweden
    1985.09-1987.07, Graduate study at the Health Science Center, Peking University, China. 1980.09-1985.07, B.Sc. in solid state physics. Department of
    Physics, Peking University, China.
    His research interests include structural biology, biochemistry and molecular biology, single-molecule biophysics, next-generation genome sequencing technology, structure-based drug and antibody engineering and rationalization design. Since receiving his Ph.D. in 1994, he has published nearly 200 academic papers in international authoritative academic journals such as Cell; Nature; Science; Nature Cancer; Nature methods; Nat. Strucut. & Mole. Biol. (NSMB); National Science Review(NSR);Annu. Rev. Pharmacol. Toxicol.; Genome Research; PNAS; Cell Research; EMBOJ.; EMBO Rep.; Cell Reports.; Cell Discovery;Nucleic Acid Research (NAR) and other internationally renowned academic journals and translated 6 books and books (including chapters in the book), and translated the best-selling textbook of structural biology, Structural Biology – From Atoms to Life.

    poster

12/14/2021: High performance large scale atomistic material simulations: the challenges and opportunities(高性能、大尺度原子级材料模拟:挑战和机遇), Prof. Lin-Wang Wang

  • Abstract: The ab initio material simulation based on quantum mechanics has been developed for more than forty years. With the maturity of the methods, the developments of user-friendly codes, and the increase of high-performance computer power, the atomistic ab initio methods, especially the methods based on density functional theory, have been used in every aspect of material research. It is not a common practice to simulate systems with hundreds of atoms. However, since the computational cost scales as the third power of the number of atoms, new strategies are needed to go beyond this barrier. Furthermore, a huge gap exists between the computable systems and many of the experimental systems, both in size scale and temporal scale. How to bridge this gap is a main challenge in order to bring the ab initio computation into industry. To overcome such challenges, we need to use multiple methods, including linear scaling method, machine leaning method, and new scheme for high performance computation. I will discuss some of these efforts in LongXun KuanTeng Inc.

    Bio: Lin-Wang Wang: Chief scientist in Semiconductor Institute, CAS, and chief scientific advisor in LongXun Kuang Teng Inc. Senior Staff Scientist, Lawrence Berkeley National Laboratory, Berkeley, CA, U.S. 1999-2021. Dr. Wang has 30 years of experience in large scale electronic structure calculations. He has worked in O(N) electronic structure calculations in early 1990s. Worked with Alex Zunger, he invented the folded spectrum method which pushed the limit of non-self-consistent electronic structure calculations from 100 atoms to thousands of atoms. He developed a linear combination of bulk bands (LCBB) method for semiconductor heterostructure electronic structure calculations, which allows the calculation of million atom devices. He developed generalized moments method which calculates the density of state and optical absorption spectra of a given system without explicit calculation of its eigenstates. He also developed a popular parallel total energy plane wave pseudopotential program (PEtot). He invented a charge patching method, which enables the ab initio accuracy thousand atom calculations for nano systems. He has developed a linear scaling three-dimensional fragment method (LS3DF), which can be used to selfconsistently calculate systems with tens of thousands of atoms. Recently, he developed a new algorithm for real-time time-dependent DFT calculations which accelerates the traditional algorithms by hundreds of times.

    poster

12/14/2021: Broadly neutralizing antibodies against SARS-CoV-2 variants(针对SARS-CoV-2变种的广谱中和抗体), Prof. Linqi Zhang

12/14/2021: Structural insights into the infection and evolution of the highly pathogenic human coronaviruses(高致病性人类冠状病毒感染和进化的结构研究), Prof. Xinquan Wang

12/14/2021: Enhancing the protein conformational sampling in molecular simulation(分子模拟中的蛋白质构象采样增强), Prof. Haipeng Gong