Stochastic gradient descent (SGD) is a popular technique for large-scale optimization problems in machine learning. In order to parallelize SGD, minibatch training needs to be employed to reduce the communication cost. However, an increase in minibatch size typically decreases the rate of convergence. This paper introduces a technique based on approximate optimization of a conservatively regularized objective function within each minibatch. We prove that the convergence rate does not decrease with increasing minibatch size. Experiments demonstrate that with suitable implementations of approximate optimization, the resulting algorithm can outperform standard SGD in many scenarios.

Cyclic GMP–AMP synthase (cGAS) is a cytosolic DNA sensor that produces the second messenger cG[2′–5′]pA[3′–5′]p (2′3′-cGAMP) and controls activation of innate immunity in mammalian cells1–5. Animal genomes typically encode multiple proteins with predicted homology to cGAS6–10, but the function of these uncharacterized enzymes is unknown. Here we show that cGAS-like receptors (cGLRs) are innate immune sensors that are capable of recognizing divergent molecular patterns and catalysing synthesis of distinct nucleotide second messenger signals. Crystal structures of human and insect cGLRs reveal a nucleotidyltransferase signalling core shared with cGAS and a diversified primary ligand-binding surface modified with notable insertions and deletions. We demonstrate that surface remodelling of cGLRs enables altered ligand specificity and used a forward biochemical screen to identify cGLR1 as a double-stranded RNA sensor in the model organism Drosophila melanogaster. We show that RNA recognition activates Drosophila cGLR1 to synthesize the novel product cG[3′–5′]pA[2′–5′]p (3′2′-cGAMP). A crystal structure of Drosophila stimulator of interferon genes (dSTING) in complex with 3′2′-cGAMP explains selective isomer recognition, and 3′2′-cGAMP induces an enhanced antiviral state in vivo that protects from viral infection. Similar to radiation of Toll-like receptors in pathogen immunity, our results establish cGLRs as a diverse family of metazoan pattern recognition receptors.

No abstract

In this paper, we present a new learning scenario, heterogeneous transfer learning, which improves learning performance when the data can be in different feature spaces and where no correspondence between data instances in these spaces is provided. In the past, we have classified Chinese text documents using English training data under the heterogeneous transfer learning framework. In this paper, we present image clustering as an example to illustrate how unsupervised learning can be improved by transferring knowledge from auxiliary heterogeneous data obtained from the social Web. Image clustering is useful for image sense disambiguation in query-based image search, but its quality is often low due to imagedata sparsity problem. We extend PLSA to help transfer the knowledge from social Web data, which have mixed feature representations. Experiments on image-object clustering and scene clustering tasks show that our approach in heterogeneous transfer learning based on the auxiliary data is indeed effective and promising.

No abstract

ObjectiveTo perform a systematic review and meta-analysis regarding the efficacy and safety of dipeptidyl peptidase-4 (DDP-4) inhibitors (“gliptins”) for the treatment of type 2 diabetes mellitus (T2DM) patients with moderate to severe renal impairment.MethodsAll available randomized-controlled trials (RCTs) that assessed the efficacy and safety of DDP-4 inhibitors compared with placebo, no treatment, or active drugs were identified using PubMed, EMBASE, Cochrane CENTRAL, conference abstracts, clinical trials.gov, pharmaceutical company websites, the FDA, and the EMA (up to June 2014). Two independent reviewers extracted the data, and a random-effects model was applied to estimate summary effects.ResultsThirteen reports of ten studies with a total of 1,915 participants were included in the final analysis. Compared with placebo or no treatment, DPP-4 inhibitors reduced HbA1c significantly (−0.52%, 95%CI −0.64 to −0.39) and had no increased risk of hypoglycemia (RR 1.10, 95%CI 0.92 to 1.32) or weight gain. In contrast to glipizide monotherapy, DPP-4 inhibitors showed no difference in HbA1c lowering effect (−0.08%, 95% CI −0.40 to 0.25) but had a lower incidence of hypoglycemia (RR 0.40, 95%CI 0.23 to 0.69). Furthermore, DPP-4 inhibitors were well-tolerated, without any additional mortality and adverse events. However, the quality of evidence was mostly as low, as assessed using the GRADE system for each outcome.ConclusionsDPP-4 inhibitors are effective at lowering HbA1c in T2DM patients with moderate to severe renal impairment. DPP-4 inhibitors also have a potential advantage in lowering the risk of adverse events. Regarding the low quality of the evidence according to GRADE, additional well-designed randomized trials that focus on the safety and efficacy of DPP-4 inhibitors in various CKD stages are needed urgently.

Glomerular mesangial cells (MCs) hypertrophy is one of the earliest pathological abnormalities in diabetic nephropathy (DN), which correlates with eventual glomerulosclerosis. This study aimed to investigate the therapeutic role of miRNA in diabetic glomerular MCs hypertrophy and synthesis of extracellular matrix (ECM). Microarray analysis revealed a significant up-regulation of miR-214 in the renal cortex of diabetic db/db mice, which was confirmed by real-time PCR of isolated glomeruli and primary cultured human MCs. In vitro studies showed that inhibition of miR-214 significantly reduced expression of α-SMA, SM22 and collagen IV, and partially restored phosphatase and tensin homolog (PTEN) protein level in high glucose-stimulated human MCs. Furthermore, we identified PTEN as the target of miR-214 by a luciferase assay in HEK293 cells. Moreover, overexpression of PTEN ameliorated miR-214-mediated diabetic MC hypertrophy while knockdown of PTEN mimicked the MC hypertrophy. In vivo study further confirmed that inhibition of miR-214 significantly decreased the expression of SM22, α-SMA and collagen IV, partially restored PTEN level, and attenuated albuminuria and mesangial expansion in db/db mice. In conclusion, cross talk between miR-214 and PTEN attenuated glomerular hypertrophy under diabetic conditions in vivo and in vitro. Therefore, miR-214 may represent a novel therapeutic target for DN.

Despite recent advances in the therapy of non-small cell lung cancer (NSCLC), the chemotherapy efficacy against NSCLC is still unsatisfactory. Previous studies show the herbal antimalarial drug dihydroartemisinin (DHA) displays cytotoxic to multiple human tumors. Here, we showed that DHA decreased cell viability and colony formation, induced apoptosis in A549 and PC-9 cells. Additionally, we first revealed DHA inhibited glucose uptake in NSCLC cells. Moreover, glycolytic metabolism was attenuated by DHA, including inhibition of ATP and lactate production. Consequently, we demonstrated that the phosphorylated forms of both S6 ribosomal protein and mechanistic target of rapamycin (mTOR), and GLUT1 levels were abrogated by DHA treatment in NSCLC cells. Furthermore, the upregulation of mTOR activation by high expressed Rheb increased the level of glycolytic metabolism and cell viability inhibited by DHA. These results suggested that DHA-suppressed glycolytic metabolism might be associated with mTOR activation and GLUT1 expression. Besides, we showed GLUT1 overexpression significantly attenuated DHA-triggered NSCLC cells apoptosis. Notably, DHA synergized with 2-Deoxy-D-glucose (2DG, a glycolysis inhibitor) to reduce cell viability and increase cell apoptosis in A549 and PC-9 cells. However, the combination of the two compounds displayed minimal toxicity to WI-38 cells, a normal lung fibroblast cell line. More importantly, 2DG synergistically potentiated DHA-induced activation of caspase-9, -8 and -3, as well as the levels of both cytochrome c and AIF of cytoplasm. However, 2DG failed to increase the reactive oxygen species (ROS) levels elicited by DHA. Overall, the data shown above indicated DHA plus 2DG induced apoptosis was involved in both extrinsic and intrinsic apoptosis pathways in NSCLC cells.