Search

Neural network models have become a major topic in many areas in recent years due to their excellent problem-solving ability. However, during the training process in the real application, neural network models usually have relatively high requirements on the performance of the platform, which limits their applications, especially in battery-driven devices or embedded platforms with relatively weak performance. Therefore, the importance of optimizing and pruning neural network models is very important. This paper analyzes four categories of pruning techniques, including channel pruning, neuron pruning, weight pruning and layer pruning. This paper investigates its main ideas and particular pruning methods and compares the optimised effects of these pruning techniques in different neural network models through experiments. This paper discusses neuron pruning on the VGG-16 model, weight pruning on ResNet and AlexNet, and layer pruning on the YOLOV3 model, respectively. The parameter amount and inference speed before and after pruning are compared, and the accuracies drop before and after pruning are verified on data sets such as CIFAR-10.

At the interface between monolayer FeSe films and SrTiO3 substrates the superconducting transition temperature (Tc) is unexpectedly high, triggering a surge of excitement. The mechanism for the Tc enhancement has been the central question, as it may present a new strategy for seeking out higher Tc materials. To reveal this enigmatic mechanism, by combining advances in high quality interface growth, 16O $$\leftrightarrow$$ ↔ 18O isotope substitution, and extensive data from angle resolved photoemission spectroscopy, we provide striking evidence that the high Tc in FeSe/SrTiO3 is the cooperative effect of the intrinsic pairing mechanism in the FeSe and interactions between the FeSe electrons and SrTiO3 phonons. Furthermore, our results point to the promising prospect that similar cooperation between different Cooper pairing channels may be a general framework to understand and design high-temperature superconductors.

This paper proposes a novel matrix rank-one decomposition for quaternion Hermitian matrices, which admits a stronger property than the previous results in (sturm2003cones,huang2007complex,ai2011new). The enhanced property can be used to drive some improved results in joint numerical range, $\mathcal{S}$-Procedure and quadratically constrained quadratic programming (QCQP) in the quaternion domain, demonstrating the capability of our new decomposition technique.

It is well known that finding a global optimum is extremely challenging for nonconvex optimization. There are some recent efforts \cite{anandkumar2016efficient, cartis2018second, cartis2020sharp, chen2019high} regarding the optimization methods for computing higher-order critical points, which can exclude the so-called degenerate saddle points and reach a solution with better quality. Desipte theoretical development in \cite{anandkumar2016efficient, cartis2018second, cartis2020sharp, chen2019high}, the corresponding numerical experiments are missing. In this paper, we propose an implementable higher-order method, named adaptive high order method (AHOM), that aims to find the third-order critical points. This is achieved by solving an ``easier'' subproblem and incorporating the adaptive strategy of parameter-tuning in each iteration of the algorithm. The iteration complexity of the proposed method is established. Some preliminary numerical results are provided to show AHOM is able to escape the degenerate saddle points, where the second-order method could possibly get stuck.

Somatic cell reprogramming by exogenous factors requires cooperation with transcriptional co-activators and co-repressors to effectively remodel the epigenetic environment. How this interplay is regulated remains poorly understood. Here, we demonstrate that NCoR/SMRT co-repressors bind to pluripotency loci to create a barrier to reprogramming with the four Yamanaka factors (OCT4, SOX2, KLF4 and c-MYC), and consequently, suppressing NCoR/SMRT significantly enhances reprogramming efficiency and kinetics. The core epigenetic subunit of the NCoR/SMRT complex, histone deacetylase 3 (HDAC3), contributes to the effects of NCoR/SMRT by inducing histone deacetylation at pluripotency loci. Among the Yamanaka factors, recruitment of NCoR/SMRT–HDAC3 to genomic loci is mostly facilitated by c-MYC. Hence, we describe how c-MYC is beneficial for the early phase of reprogramming but deleterious later. Overall, we uncover a role for NCoR/SMRT co-repressors in reprogramming and propose a dual function for c-MYC in this process. Zhuang et al. demonstrate that suppression of NCoR/SMRT enhances OSKM reprogramming efficiency, and that the barrier mechanism depends on the recruitment of HDAC3 to pluripotency loci by c-MYC.

Somatic cells and pluripotent cells display remarkable differences in most aspects of cell function. Accordingly, somatic cell reprogramming by exogenous factors requires comprehensive changes in gene transcription to induce a forced pluripotent state, which is encompassed by a simultaneous transformation of the epigenome. Nevertheless, how the reprogramming factors and other endogenous regulators coordinate to suppress the somatic cell gene program and activate the pluripotency gene network, and why the conversion is multi-phased and lengthy, remain enigmatic. We summarize the current knowledge of transcriptional regulation in somatic cell reprogramming, and highlight new perspectives that may help to reshape existing paradigms.

Emerging studies document the roles of long non-coding RNAs (LncRNAs) in regulating gene expression at chromatin level but relatively less is known how they regulate DNA methylation. Here we identify an lncRNA, Dum (developmental pluripotency-associated 2 (Dppa2) Upstream binding Muscle lncRNA) in skeletal myoblast cells. The expression of Dum is dynamically regulated during myogenesis in vitro and in vivo. It is also transcriptionally induced by MyoD binding upon myoblast differentiation. Functional analyses show that it promotes myoblast differentiation and damage-induced muscle regeneration. Mechanistically, Dum was found to silence its neighboring gene, Dppa2, in cis through recruiting Dnmt1, Dnmt3a and Dnmt3b. Furthermore, intrachromosomal looping between Dum locus and Dppa2 promoter is necessary for Dum/Dppa2 interaction. Collectively, we have identified a novel lncRNA that interacts with Dnmts to regulate myogenesis.

We combine the labeling of newly transcribed RNAs with 5-ethynyluridine with the characterization of bound proteins. This approach, named capture of the newly transcribed RNA interactome using click chemistry (RICK), systematically captures proteins bound to a wide range of RNAs, including nascent RNAs and traditionally neglected nonpolyadenylated RNAs. RICK has identified mitotic regulators amongst other novel RNA-binding proteins with preferential affinity for nonpolyadenylated RNAs, revealed a link between metabolic enzymes/factors and nascent RNAs, and expanded the known RNA-bound proteome of mouse embryonic stem cells. RICK will facilitate an in-depth interrogation of the total RNA-bound proteome in different cells and systems.

Malat1 is one of the most abundant long non-coding RNAs in various cell types; its exact cellular function is still a matter of intense investigation. In this study we characterized the function of Malat1 in skeletal muscle cells and muscle regeneration. Utilizing both in vitro and in vivo assays, we demonstrate that Malat1 has a role in regulating gene expression during myogenic differentiation of myoblast cells. Specifically, we found that knockdown of Malat1 accelerates the myogenic differentiation in cultured cells. Consistently, Malat1 knockout mice display enhanced muscle regeneration after injury and deletion of Malat1 in dystrophic mdx mice also improves the muscle regeneration. Mechanistically, in the proliferating myoblasts, Malat1 recruits Suv39h1 to MyoD-binding loci, causing trimethylation of histone 3 lysine 9 (H3K9me3), which suppresses the target gene expression. Upon differentiation, the pro-myogenic miR-181a is increased and targets the nuclear Malat1 transcripts for degradation through Ago2-dependent nuclear RNA-induced silencing complex machinery; the Malat1 decrease subsequently leads to the destabilization of Suv39h1/HP1β/HDAC1-repressive complex and displacement by a Set7-containing activating complex, which allows MyoD trans-activation to occur. Together, our findings identify a regulatory axis of miR-181a-Malat1-MyoD/Suv39h1 in myogenesis and uncover a previously unknown molecular mechanism of Malat1 action in gene regulation.

The generation of induced pluripotent stem (iPS) cells by exogenous transcription factors involves a comprehensive rearrangement of cellular functions, including the microRNA profile. The resulting cell lines are similar to embryonic stem (ES) cells and have therefore raised much interest for in vitro studies and the perspective of clinical application. Yet, microRNAs are not mere listeners of the reprogramming orchestra but play an active role in the process. In consequence, overexpression or suppression of individual microRNAs has profound effects in colony formation efficiency, and in combination they can produce iPS cells without added transcription factors. Moreover, variations in microRNA expression of iPS/ES cells can predict their differentiation potential and may have consequences at other levels. Altogether, these findings highlight the relevance of pursuing further these studies.

In the version of this Article originally published, in Fig. 2c, the '+' sign and 'OSKM' were superimposed in the label '+OSKM'. In Fig. 4e, in the labels, all instances of 'Ant' should have been 'Anti-'. And, in Fig. 7a, the label '0.0' was misplaced; it should have been on the colour scale bar. These figures have now been corrected in the online versions.

Little is known how lincRNAs are involved in skeletal myogenesis. Here we describe the discovery of Linc-YY1 from the promoter of the transcription factor (TF) Yin Yang 1 (YY1) gene. We demonstrate that Linc-YY1 is dynamically regulated during myogenesis in vitro and in vivo. Gain or loss of function of Linc-YY1 in C2C12 myoblasts or muscle satellite cells alters myogenic differentiation and in injured muscles has an impact on the course of regeneration. Linc-YY1 interacts with YY1 through its middle domain, to evict YY1/Polycomb repressive complex (PRC2) from target promoters, thus activating the gene expression in trans. In addition, Linc-YY1 also regulates PRC2-independent function of YY1. Finally, we identify a human Linc-YY1 orthologue with conserved function and show that many human and mouse TF genes are associated with lincRNAs that may modulate their activity. Altogether, we show that Linc-YY1 regulates skeletal myogenesis and uncover a previously unappreciated mechanism of gene regulation by lincRNA.

Recent studies have boosted our understanding of long noncoding RNAs (lncRNAs) in numerous biological processes, but few have examined their roles in somatic cell reprogramming. Through expression profiling and functional screening, we have identified that the large intergenic noncoding RNA p21 (lincRNA-p21) impairs reprogramming. Notably, lincRNA-p21 is induced by p53 but does not promote apoptosis or cell senescence in reprogramming. Instead, lincRNA-p21 associates with the H3K9 methyltransferase SETDB1 and the maintenance DNA methyltransferase DNMT1, which is facilitated by the RNA-binding protein HNRNPK. Consequently, lincRNA-p21 prevents reprogramming by sustaining H3K9me3 and/or CpG methylation at pluripotency gene promoters. Our results provide insight into the role of lncRNAs in reprogramming and establish a novel link between p53 and heterochromatin regulation.

In an attempt to define the binding of staphylococcal enterotoxin D (SED) to the T cell receptor Vbeta (TCR Vbeta) and major histocompatibility complex (MHC) class II molecules, site-directed mutagenesis has been used to introduce alanine substitutions at Asn23, Phe45, Leu59, Asn61, Ile92 and Phe203 in SED. SED-N23A (SEA with Asn23 replaced with alanine) and SED-F45A mutants exhibit a significantly reduced ability to induce T cell proliferation. However, SED-L59A, SED-N61A, SED-I92A and SED-F203A mutants exhibit the normal mitogenic activity. The ability binding to MHC class II and the TCR Vbeta specificity of SED-N23A and SED-F45A were then detected. SED-N23A, but not SED-F45A, was able to compete effectively with FITC-conjugated SED for binding to Raji cells. This finding indicates that the mitogenic activity defection of SED-N23A is not due to poor binding to SED-MHC class II molecules. When stimulated with SED-N23A, T cells bearing TCR Vbeta5 were significantly reduced. When stimulated with SED-F45A, T cells bearing TCR Vbeta5, TCR Vbeta8 and TCR Vbeta12.1 were all significantly reduced. These results suggest Asn23 is an important residue involved SED interacting with TCR Vbeta; Phe45 is required for effective interaction with MHC class II molecules; and the ability of SED stimulating certain TCR Vbeta+ T cells is dependent on Phe45 binding to MHC class II molecules.

To explore the postburn adhesion properties of polymorphonuclear leukocyte (PMN) onto pulmonary vascular endothelial cells (PVEC) in burn patients with acute lung injury (ALI), so as to determine the role of C5a on PVEC-PMN adhesion.Microtubule sucking technique was employed to determine the PVEC-PMN adhesion. The myeloperoxidase (MPO) was also assayed to reflect the magnitude of PVEC-PMN adhesion.The magnitude of PVEC-PMN adhesion increased and the adhesion force increased along with an increase in rh-C5a concentration. Simultaneously, the MPO activity was increased, which could be inhibited by anti-C5aR McAb in a concentration 1:104.Both C5a and C5aR participated in PVEC-PMN adhesion, which might be important in the pathogenesis of ALI.