Learning to Discretize: Solving 1D Scalar Conservation Laws via Deep Reinforcement Learning

Conservation laws are considered to be fundamental laws of nature. It has broad applications in many fields, including physics, chemistry, biology, geology, and engineering. Solving the differential equations associated with conservation laws is a major branch in computational mathematics. The recent success of machine learning, especially deep learning in areas such as computer vision and natural language processing, has attracted a lot of attention from the community of computational mathematics and inspired many intriguing works in combining machine learning with traditional methods. In this paper, we are the first to view numerical PDE solvers as an MDP and to use (deep) RL to learn new solvers. As proof of concept, we focus on 1-dimensional scalar conservation laws. We deploy the machinery of deep reinforcement learning to train a policy network that can decide on how the numerical solutions should be approximated in a sequential and spatial-temporal adaptive manner. We will show that the problem of solving conservation laws can be naturally viewed as a sequential decision-making process, and the numerical schemes learned in such a way can easily enforce long-term accuracy. Furthermore, the learned policy network is carefully designed to determine a good local discrete approximation based on the current state of the solution, which essentially makes the proposed method a meta-learning approach. In other words, the proposed method is capable of learning how to discretize for a given situation mimicking human experts. Finally, we will provide details on how the policy network is trained, how well it performs compared with some state-of-the-art numerical solvers such as WENO schemes, and supervised learning based approach L3D and PINN, and how well it generalizes.     

INAUGURAL ARTICLE by a Recently Elected Academy Member:Functional redundancy of type I and type II receptors in the regulation of skeletal muscle growth by myostatin and activin A

Myostatin (MSTN) is a transforming growth factor-β (TGF-β) family member that normally acts to limit muscle growth. The function of MSTN is partially redundant with that of another TGF-β family member, activin A. MSTN and activin A are capable of signaling through a complex of type II and type I receptors. Here, we investigated the roles of two type II receptors (ACVR2 and ACVR2B) and two type I receptors (ALK4 and ALK5) in the regulation of muscle mass by these ligands by genetically targeting these receptors either alone or in combination specifically in myofibers in mice. We show that targeting signaling in myofibers is sufficient to cause significant increases in muscle mass, showing that myofibers are the direct target for signaling by these ligands in the regulation of muscle growth. Moreover, we show that there is functional redundancy between the two type II receptors as well as between the two type I receptors and that all four type II/type I receptor combinations are utilized in vivo. Targeting signaling specifically in myofibers also led to reductions in overall body fat content and improved glucose metabolism in mice fed either regular chow or a high-fat diet, demonstrating that these metabolic effects are the result of enhanced muscling. We observed no effect, however, on either bone density or muscle regeneration in mice in which signaling was targeted in myofibers. The latter finding implies that MSTN likely signals to other cells, such as satellite cells, in addition to myofibers to regulate muscle homeostasis.

Myostatin (MSTN) is a secreted signaling molecule that normally acts to limit skeletal muscle growth (for review, see ref. 1). Mice lacking MSTN exhibit dramatic increases in muscle mass throughout the body, with individual muscles growing to about twice the normal size (2). MSTN appears to play two distinct roles in regulating muscle size, one to regulate the number of muscle fibers that are formed during development and a second to regulate the growth of those fibers postnatally. The sequence of MSTN has been highly conserved through evolution, with the mature MSTN peptide being identical in species as divergent as humans and turkeys (3). The function of MSTN has also been conserved, and targeted or naturally occurring mutations in MSTN have been shown to cause increased muscling in numerous species, including cattle (3–5), sheep (6), dogs (7), rabbits (8), rats (9), swine (10), goats (11), and humans (12). Numerous pharmaceutical and biotechnology companies have developed biologic agents capable of blocking MSTN activity, and these have been tested in clinical trials for a wide range of indications, including Duchenne and facioscapulohumeral muscular dystrophy, inclusion body myositis, muscle atrophy following falls and hip fracture surgery, age-related sarcopenia, Charcot–Marie–Tooth disease, and cachexia due to chronic obstructive pulmonary disease, end-stage kidney disease, and cancer.The finding that certain inhibitors of MSTN signaling can increase muscle mass even in Mstn^−/− mice revealed that the function of MSTN as a negative regulator of muscle mass is partially redundant with at least one other TGF-β family member (13, 14), and subsequent studies have identified activin A as one of these cooperating ligands (15, 16). MSTN and activin A share many key regulatory and signaling components. For example, the activities of both MSTN and activin A can be modulated extracellularly by naturally occurring inhibitory binding proteins, including follistatin (17, 18) and the follistatin-related protein, FSTL-3 or FLRG (19, 20). Moreover, MSTN and activin A also appear to share receptor components. Based on in vitro studies, MSTN is capable of binding initially to the activin type II receptors, ACVR2 and ACVR2B (also called ActRIIA and ActRIIB) (18) followed by engagement of the type I receptors, ALK4 and ALK5 (21). In previous studies, we presented genetic evidence supporting a role for both ACVR2 and ACVR2B in mediating MSTN signaling and regulating muscle mass in vivo. Specifically, we showed that mice expressing a truncated, dominant-negative form of ACVR2B in skeletal muscle (18) or carrying deletion mutations in Acvr2 and/or Acvr2b (13) have significantly increased muscle mass. One limitation of the latter study, however, was that we could not examine the consequence of complete loss of both receptors using the deletion alleles, as double homozygous mutants die early during embryogenesis (22). Moreover, the roles that the two type I receptors, ALK4 and ALK5, play in regulating MSTN and activin A signaling in muscle in vivo have not yet been documented using genetic approaches. Here, we present the results of studies in which we used floxed alleles for each of the type II and type I receptor genes in order to target these receptors alone and in combination in muscle fibers. We show that these receptors are functionally redundant and that signaling through each of these receptors contributes to the overall control of muscle mass.     

基于寒区发病特点的龙江医家血液病中医辨治思维与实践＊

中医临床思维在诊疗中扮演着重要角色，为立法之纲纪，治病之根本。今之血液系统疾病变化趋于复杂，临床辨治思维是中医血液病诊治的关键所在。本文基于寒区发病特点、龙江地域气候与人民生活特征，总结以前辈高仲山、华廷芳先生，国家级名中医孙伟正教授为代表的龙江名医血液病诊疗特色经验，通过“因实、因虚致病”病机，以“三因制宜”为基础，系统阐述了“实者以决、结者以散、虚者宜补、劳者宜温”的血液病中医辨治思维及其实践，为寒区血液疾病的中医诊治提供了治法方向。     

改革开放以来中医药文化发展的演进与特点探析＊

中医药文化是中医药事业发展的根基和灵魂，社会主义文化建设的重要组成。本文考察改革开放40年以来中医药文化发展的演进过程，从“社会矛盾—政策方针—中医药文化发展”的中观视角，构建了中医药文化发展分析模型，发现中医药文化先后经历了资产化发展（1978年-21世纪初）、情景化发展（21世纪初-中国共产党第十八次全国代表大会（中共十八大）前后）、生态化发展（中共十八大至今）三个阶段，每个阶段中医药文化发展的策略、重点和面临的问题具有明显的不同特点。因此，本文提出以社会矛盾为导向，优化政策引导，构建中医药文化生态体系，从个人、社会和国家发展的不同层面采取多元双向发展路径，推进健康中国发展，提升文化自信，打造文化强国。     

急性乙型肝炎患者pDCs上CD86表达变化及意义

目的 探讨外周血浆细胞样树突状细胞(pDCs)在急性乙型肝炎(AHB）患者临床转归中的变化及作用。方法 纳入2015年6月至2017年5月收治的急性乙型肝炎(AHB组)患者40例， 给予退黄、降酶、保肝等药物，并加用恩替卡韦0．5 mg空腹口服，1次/d。另纳入健康体检者26例为健康对照组（ HC组）。HC组于体检当天、AHB组于治疗前及治疗后6周采集外周静脉全血，采用ELISA法检测2组血浆HBV DNA、HBsAg、HbeAg及肝功能指标水平，并应用流式细胞仪检测pDCs的频数及其功能分子CD86的表达水平。结果 治疗6周后，AHB组临床症状显著缓解，血清学指标和乙型肝炎标志物水平显著下降（P<0.01）。治疗6周后，AHB组HBV DNA转阴率为82.5%，HBsAg清除率为72.5%，HBeAg血清学转换率为75.0%，与治疗前比较差异均有统计学意义（P<0.01）, HC组pDCs频数显著高于AHB组治疗前。AHB组治疗后CD86+pDCs显著高于治疗前。HC组CD86ABC显著低于AHB组治疗前。结论 在急性乙型肝炎患者外周血pDCs的数量降低。急性乙型肝炎患者在治疗后CD86+pDCs增多。急性乙型肝炎患者pDCs上CD86表达上调。     

High infiltration of mast cells is associated with improved response to adjuvant chemotherapy in gallbladder cancer

Recent studies have reported that tumor‐infiltrating mast cells (TIM) play an important role in tumor regression, but the effect of TIM in gallbladder cancer (GBC) remains unclear. The present study aims to investigate the prognostic value of TIM in GBC patients and its responsiveness to gemcitabine‐based adjuvant chemotherapy (ACT). A total of 298 GBC patients from Zhongshan Hospital were recruited for this study. TIM infiltration was measured by immunohistochemical staining. Accumulation of TIM is significantly associated with prolonged overall survival in GBC patients. The benefit from gemcitabine‐based ACT was superior among patients with high infiltration of TIM with GBC. Multivariate analysis identified TIM infiltration as an independent prognostic factor for overall survival. A heatmap showed that TIM‐activated gene signatures were positively correlated with CD8+ T cells' gene signatures. Gene set enrichment analysis (GSEA) suggested that TIM was related to multiple T cell‐related processes and signaling pathways, including the interferon gamma signaling pathway and the leukocyte migration signaling pathway. It was confirmed that CD8+ T cell infiltration was positively correlated with high TIM infiltration in tissue microarray (TMA), suggesting that TIM infiltration was linked to the immune surveillance in GBC. TIM can be used as an independent prognostic factor and a predictor of therapeutic response of gemcitabine‐based ACT in GBC patients, which may mediate immune surveillance by recruiting and activating CD8+ T cells in GBC.     

互动式歌唱表演对轻中度阿尔茨海默病患者抑郁、精神行为和运动训练参与率的影响

目的 观察互动式歌唱表演对轻中度阿尔茨海默病（AD）患者抑郁、精神行为症状及运动训练参与率的影响。方法 选取符合入组条件≥60周岁AD患者63例，随机分为研究组（31例）和对照组（32例）。所有受试患者常规药物治疗及常规运动训练，对照组接受被动性音乐治疗，研究组接受以互动歌唱为主的主动性音乐治疗，1次/d，每次1小时，每周训练5天，持续干预6个月。于治疗前、治疗1个月后、治疗3个月后、治疗6个月后分别采用康奈尔痴呆抑郁量表（CSDD）评分、阿尔茨海默病病理行为（BEHAVE AD）评分、参与率进行评估。结果 治疗1个月、3个月后，研究组CSDD评分较治疗前均降低（P<0.05）；治疗6个月后，研究组患者CSDD评分较治疗前、治疗1个月、3个月后均显著降低（P<0.05），且与对照组比较差异有统计学意义（P<0.05）。治疗1个月、3个月后，研究组BEHAVE AD评分较治疗前均降低（P<0.05）；治疗6个月后，研究组患者BEHAVE AD评分较治疗前、治疗1个月、3个月后均显著降低（P<0.05），且与对照组比较差异有统计学意义（P<0.05）。治疗6个月后，两组运动训练参与率组间比较差异有统计学意义（P<0.05）。结论 互动式歌唱表演可能对改善轻中度AD患者的抑郁和精神行为症状有着积极的疗效，同时对提高受试者运动训练的参与率可能有着更积极的疗效。