Dynamics of learning with restricted training sets

The dynamics of supervised learning in layered neural networks were studied in the regime where the size of the training set is proportional to the number of inputs. The evolution of macroscopic observables, including the two relevant performance measures can be predicted by using the dynamical replica theory. Three approximation schemes aimed at eliminating the need to solve a functional saddle-point equation at each time step have been derived.

Noise, regularizers, and unrealizable scenarios in online learning from restricted training sets

We study the dynamics of on-line learning in multilayer neural networks where training examples are sampled with repetition and where the number of examples scales with the number of network weights. The analysis is carried out using the dynamical replica method aimed at obtaining a closed set of coupled equations for a set of macroscopic variables from which both training and generalization errors can be calculated. We focus on scenarios whereby training examples are corrupted by additive Gaussian output noise and regularizers are introduced to improve the network performance. The dependence of the dynamics on the noise level, with and without regularizers, is examined, as well as that of the asymptotic values obtained for both training and generalization errors. We also demonstrate the ability of the method to approximate the learning dynamics in structurally unrealizable scenarios. The theoretical results show good agreement with those obtained by computer simulations.

Oxidized phospholipid inhibition of toll-like receptor (TLR) signaling is restricted to TLR2 and TLR4 roles for cd14, lps-binding protein, and md2 as targets for specificity of inhibition

The generation of reactive oxygen species is a central feature of inflammation that results in the oxidation of host phospholipids. Oxidized phospholipids, such as 1-palmitoyl-2-arachidonyl-sn-glycero-3-phosphorylcholine (OxPAPC), have been shown to inhibit signaling induced by bacterial lipopeptide or lipopolysac-charide (LPS), yet the mechanisms responsible for the inhibition of Toll-like receptor (TLR) signaling by OxPAPC remain incompletely understood. Here, we examined the mechanisms by which OxPAPC inhibits TLR signaling induced by diverse ligands in macrophages, smooth muscle cells, and epithelial cells. OxPAPC inhibited tumor necrosis factor- production, IB degradation, p38 MAPK phosphorylation, and NF-B-dependent reporter activation induced by stimulants of TLR2 and TLR4 (Pam3CSK4 and LPS) but not by stimulants of other TLRs (poly(I·C), flagellin, loxoribine, single-stranded RNA, or CpG DNA) in macrophages and HEK-293 cells transfected with respective TLRs and significantly reduced inflammatory responses in mice injected subcutaneously or intraperitoneally with Pam3CSK4. Serum proteins, including CD14 and LPS-binding protein, were identified as key targets for the specificity of TLR inhibition as supplementation with excess serum or recombinant CD14 or LBP reversed TLR2 inhibition by OxPAPC, whereas serum accessory proteins or expression of membrane CD14 potentiated signaling via TLR2 and TLR4 but not other TLRs. Binding experiments and functional assays identified MD2 as a novel additional target of OxPAPC inhibition of LPS signaling. Synthetic phospholipid oxidation products 1-palmitoyl-2-(5-oxovaleryl)-sn-glycero-3-phosphocholine and 1-palmitoyl-2-glutaryl-sn-glycero-3-phosphocholine inhibited TLR2 signaling from 30 µM. Taken together, these results suggest that oxidized phospholipid-mediated inhibition of TLR signaling occurs mainly by competitive interaction with accessory proteins that interact directly with bacterial lipids to promote signaling via TLR2 or TLR4.

Specific lipidome signatures in central nervous system from methionine-restricted mice

Membrane lipid composition is an important correlate of the rate of aging of animals. Dietary methionine restriction (MetR) increases lifespan in rodents. The underlying mechanisms have not been elucidated but could include changes in tissue lipidomes. In this work, we demonstrate that 80% MetR in mice induces marked changes in the brain, spinal cord, and liver lipidomes. Further, at least 50% of the lipids changed are common in the brain and spinal cord but not in the liver, suggesting a nervous system-specific lipidomic profile of MetR. The differentially expressed lipids includes (a) specific phospholipid species, which could reflect adaptive membrane responses, (b) sphingolipids, which could lead to changes in ceramide signaling pathways, and (c) the physiologically redox-relevant ubiquinone 9, indicating adaptations in phase II antioxidant response metabolism. In addition, specific oxidation products derived from cholesterol, phosphatidylcholine, and phosphatidylethanolamine were significantly decreased in the brain, spinal cord, and liver from MetR mice. These results demonstrate the importance of adaptive responses of membrane lipids leading to increased stress resistance as a major mechanistic contributor to the lowered rate of aging in MetR mice. © 2013 American Chemical Society.

Cardiac dysfunction and cell damage across clinical stages of severity in growth-restricted fetuses

Objective - The purpose of this study was to assess cardiac function and cell damage in intrauterine growth-restricted (IUGR) fetuses across clinical Doppler stages of deterioration. Study Design - One hundred twenty appropriate-for-gestational-age and 81 IUGR fetuses were classified in stages 1/2/3 according umbilical artery present/absent/reversed end-diastolic blood flow, respectively. Cardiac function was assessed by modified-myocardial performance index, early-to-late diastolic filling ratios, cardiac output, and cord blood B-type natriuretic peptide; myocardial cell damage was assessed by heart fatty acid–binding protein, troponin-I, and high-sensitivity C-reactive protein. Results - Modified-myocardial performance index, blood B-type natriuretic peptide, and early-to-late diastolic filling ratios were increased in a stage-dependent manner in IUGR fetuses, compared with appropriate-for-gestational-age fetuses. Heart fatty acid–binding protein levels were higher in IUGR fetuses at stage 3, compared with control fetuses. Cardiac output, troponin-I, and high-sensitivity C-reactive protein did not increase in IUGR fetuses at any stage. Conclusion - IUGR fetuses showed signs of cardiac dysfunction from early stages. Cardiac dysfunction deteriorates further with the progression of fetal compromise, together with the appearance of biochemical signs of cell damage.

Effects of lipoic acid and dihydrolipoic acid on total erythrocytic thiols under conditions of restricted glucose in vitro

The effects of lipoic acid and dihydrolipoic acid were explored on total thiol maintenance in diabetic and non-diabetic human erythrocytes in vitro over 22 hr in a 37°C incubation system with no added glucose. Over 18-22.5 hr after treatment in both non-diabetic and diabetic cells, lipoic acid (1 mM) was associated with greater loss of cellular thiols than dihydrolipoic acid (1 mM), compared to respective control values. At 0.1 mM, in non-diabetic cells, although lipoic acid-treated cells' thiol levels were significantly lower than control, there was no significant difference between dihydrolipoic acid-treated cells and control cells regarding thiol levels. In addition, at 0.1 mM, dihydrolipoic acid-treated diabetic cells showed a reduction in thiol levels compared to control. At 0.01 mM, lipoic acid-treated cells had significantly lower measured thiol levels compared with diabetic cells exposed to dihydrolipoic acid, whereas in non-diabetic cells, dihydrolipoic acid-treated erythrocytic thiol levels were significantly greater than those treated with lipoic acid, although there were no other significant differences between the groups. At 22.5 hr, control values of methaemoglobin rose to 6.4 ± 1.1% in diabetic cells and 3.6 ± 2.1% in non-diabetic cells. Lipoic acid (1 mM) showed greater methaemoglobin formation in diabetic rather than non-diabetic cells (13.6 ± 1.5% versus 11.6 ± 1.5%), whereas dihydrolipoic acid-treated diabetic and non-diabetic cells were less potent in methaemoglobin generation (8.5 ± 2.4% and 8.4 ± 1.4%, respectively). These studies suggest that in certain circumstances such as hypoglycaemia, lipoic acid administration may actually be detrimental to cellular oxidant protection status. © 2006 The Authors.

Site-restricted web searches for data collection in regional dialectology

This article presents a new method for data collection in regional dialectology based on site-restricted web searches. The method measures the usage and determines the distribution of lexical variants across a region of interest using common web search engines, such as Google or Bing. The method involves estimating the proportions of the variants of a lexical alternation variable over a series of cities by counting the number of webpages that contain the variants on newspaper websites originating from these cities through site-restricted web searches. The method is evaluated by mapping the 26 variants of 10 lexical variables with known distributions in American English. In almost all cases, the maps based on site-restricted web searches align closely with traditional dialect maps based on data gathered through questionnaires, demonstrating the accuracy of this method for the observation of regional linguistic variation. However, unlike collecting dialect data using traditional methods, which is a relatively slow process, the use of site-restricted web searches allows for dialect data to be collected from across a region as large as the United States in a matter of days.

Towards the in silico identification of class II restricted T-cell epitopes:a partial least squares iterative self-consistent algorithm for affinity prediction

Motivation: The immunogenicity of peptides depends on their ability to bind to MHC molecules. MHC binding affinity prediction methods can save significant amounts of experimental work. The class II MHC binding site is open at both ends, making epitope prediction difficult because of the multiple binding ability of long peptides. Results: An iterative self-consistent partial least squares (PLS)-based additive method was applied to a set of 66 pep- tides no longer than 16 amino acids, binding to DRB1*0401. A regression equation containing the quantitative contributions of the amino acids at each of the nine positions was generated. Its predictability was tested using two external test sets which gave r pred =0.593 and r pred=0.655, respectively. Furthermore, it was benchmarked using 25 known T-cell epitopes restricted by DRB1*0401 and we compared our results with four other online predictive methods. The additive method showed the best result finding 24 of the 25 T-cell epitopes. Availability: Peptides used in the study are available from http://www.jenner.ac.uk/JenPep. The PLS method is available commercially in the SYBYL molecular modelling software package. The final model for affinity prediction of peptides binding to DRB1*0401 molecule is available at http://www.jenner.ac.uk/MHCPred. Models developed for DRB1*0101 and DRB1*0701 also are available in MHC- Pred

Learning higher-level features with convolutional restricted Boltzmann machines for sentiment analysis

In recent years, learning word vector representations has attracted much interest in Natural Language Processing. Word representations or embeddings learned using unsupervised methods help addressing the problem of traditional bag-of-word approaches which fail to capture contextual semantics. In this paper we go beyond the vector representations at the word level and propose a novel framework that learns higher-level feature representations of n-grams, phrases and sentences using a deep neural network built from stacked Convolutional Restricted Boltzmann Machines (CRBMs). These representations have been shown to map syntactically and semantically related n-grams to closeby locations in the hidden feature space. We have experimented to additionally incorporate these higher-level features into supervised classifier training for two sentiment analysis tasks: subjectivity classification and sentiment classification. Our results have demonstrated the success of our proposed framework with 4% improvement in accuracy observed for subjectivity classification and improved the results achieved for sentiment classification over models trained without our higher level features.