Search

Motivated by biological considerations, we study sparse neural maps from an input layer to a target layer with sparse activity, and specifically the problem of storing K input-target associations (x , y) , or memories, when the target vectors y are sparse. We mathematically prove that K undergoes a phase transition and that in general, and somewhat paradoxically, sparsity in the target layers increases the storage capacity of the map. The target vectors can be chosen arbitrarily, including in random fashion, and the memories can be both encoded and decoded by networks trained using local learning rules, including the simple Hebb rule. These results are robust under a variety of statistical assumptions on the data. The proofs rely on elegant properties of random polytopes and sub-gaussian random vector variables. Open problems and connections to capacity theories and polynomial threshold maps are discussed. [ABSTRACT FROM AUTHOR]

A long standing open problem in the theory of neural networks is the development of quantitative methods to estimate and compare the capabilities of different architectures. Here we define the capacity of an architecture by the binary logarithm of the number of functions it can compute, as the synaptic weights are varied. The capacity provides an upperbound on the number of bits that can be extracted from the training data and stored in the architecture during learning. We study the capacity of layered, fully-connected, architectures of linear threshold neurons with L layers of size n 1 , n 2 , ... , n L and show that in essence the capacity is given by a cubic polynomial in the layer sizes: C (n 1 , ... , n L) = ∑ k = 1 L − 1 min (n 1 , ... , n k) n k n k + 1 , where layers that are smaller than all previous layers act as bottlenecks. In proving the main result, we also develop new techniques (multiplexing, enrichment, and stacking) as well as new bounds on the capacity of finite sets. We use the main result to identify architectures with maximal or minimal capacity under a number of natural constraints. This leads to the notion of structural regularization for deep architectures. While in general, everything else being equal, shallow networks compute more functions than deep networks, the functions computed by deep networks are more regular and "interesting". [ABSTRACT FROM AUTHOR]

In a physical neural system, learning rules must be local both in space and time. In order for learning to occur, non-local information must be communicated to the deep synapses through a communication channel, the deep learning channel. We identify several possible architectures for this learning channel (Bidirectional, Conjoined, Twin, Distinct) and six symmetry challenges: (1) symmetry of architectures; (2) symmetry of weights; (3) symmetry of neurons; (4) symmetry of derivatives; (5) symmetry of processing; and (6) symmetry of learning rules. Random backpropagation (RBP) addresses the second and third symmetry, and some of its variations, such as skipped RBP (SRBP) address the first and the fourth symmetry. Here we address the last two desirable symmetries showing through simulations that they can be achieved and that the learning channel is particularly robust to symmetry variations. Specifically, random backpropagation and its variations can be performed with the same non-linear neurons used in the main input–output forward channel, and the connections in the learning channel can be adapted using the same algorithm used in the forward channel, removing the need for any specialized hardware in the learning channel. Finally, we provide mathematical results in simple cases showing that the learning equations in the forward and backward channels converge to fixed points, for almost any initial conditions. In symmetric architectures, if the weights in both channels are small at initialization, adaptation in both channels leads to weights that are essentially symmetric during and after learning. Biological connections are discussed. [ABSTRACT FROM AUTHOR]

Summary Cancer stem cells (CSCs) have high tumorigenic capacity. Here, we show that stem-like traits of specific human cancer cells are reduced by overexpression of the histone deacetylase sirtuin 6 (SIRT6). SIRT6-sensitive cancer cells bear mutations that activate phosphatidylinositol-3-kinase (PI3K) signaling, and overexpression of SIRT6 reduces growth, progression, and grade of breast cancer in a mouse model with PI3K activation. Tumor metabolomic and transcriptomic analyses reveal that SIRT6 overexpression dampens PI3K signaling and stem-like characteristics and causes metabolic rearrangements in this cancer model. Ablation of a PI3K activating mutation in otherwise isogenic cancer cells is sufficient to convert SIRT6-sensitive into SIRT6-insensitive cells. SIRT6 overexpression suppresses PI3K signaling at the transcriptional level and antagonizes tumor sphere formation independent of its histone deacetylase activity. Our data identify SIRT6 as a putative molecular target that hinders stemness of tumors with PI3K activation. [ABSTRACT FROM AUTHOR]

In a physical neural system, where storage and processing are intimately intertwined, the rules for adjusting the synaptic weights can only depend on variables that are available locally, such as the activity of the pre- and post-synaptic neurons, resulting in local learning rules . A systematic framework for studying the space of local learning rules is obtained by first specifying the nature of the local variables, and then the functional form that ties them together into each learning rule. Such a framework enables also the systematic discovery of new learning rules and exploration of relationships between learning rules and group symmetries. We study polynomial local learning rules stratified by their degree and analyze their behavior and capabilities in both linear and non-linear units and networks. Stacking local learning rules in deep feedforward networks leads to deep local learning . While deep local learning can learn interesting representations, it cannot learn complex input–output functions, even when targets are available for the top layer. Learning complex input–output functions requires local deep learning where target information is communicated to the deep layers through a backward learning channel . The nature of the communicated information about the targets and the structure of the learning channel partition the space of learning algorithms. For any learning algorithm, the capacity of the learning channel can be defined as the number of bits provided about the error gradient per weight, divided by the number of required operations per weight. We estimate the capacity associated with several learning algorithms and show that backpropagation outperforms them by simultaneously maximizing the information rate and minimizing the computational cost. This result is also shown to be true for recurrent networks, by unfolding them in time. The theory clarifies the concept of Hebbian learning, establishes the power and limitations of local learning rules, introduces the learning channel which enables a formal analysis of the optimality of backpropagation, and explains the sparsity of the space of learning rules discovered so far. [ABSTRACT FROM AUTHOR]

The circadian clock aligns physiology with the 24-hour rotation of Earth. Light and food are the main environmental cues (zeitgebers) regulating circadian rhythms in mammals. Yet, little is known about the interaction between specific dietary components and light in coordinating circadian homeostasis. Herein, we focused on the role of essential amino acids. Mice were fed diets depleted of specific essential amino acids and their behavioral rhythms were monitored and tryptophan was selected for downstream analyses. The role of tryptophan metabolism in modulating circadian homeostasis was studied using isotope tracing as well as transcriptomic- and metabolomic- analyses. Dietary tryptophan depletion alters behavioral rhythms in mice. Furthermore, tryptophan metabolism was shown to be regulated in a time- and light- dependent manner. A multi-omics approach and combinatory diet/light interventions demonstrated that tryptophan metabolism modulates temporal regulation of metabolism and transcription programs by buffering photic cues. Specifically, tryptophan metabolites regulate central circadian functions of the suprachiasmatic nucleus and the core clock machinery in the liver. Tryptophan metabolism is a modulator of circadian homeostasis by integrating environmental cues. Our findings propose tryptophan metabolism as a potential point for pharmacologic intervention to modulate phenotypes associated with disrupted circadian rhythms. [Display omitted] • Dietary tryptophan and threonine controls circadian behavioral rhythms. • Tryptophan is metabolized in a time- and light- dependent manner. • Light cycles and tryptophan intake coordinate temporal regulation of metabolism and transcription. • Tryptophan metabolism regulate expression of the core clock machinery. [ABSTRACT FROM AUTHOR]

Abstract: Circadian rhythms control metabolism and energy homeostasis, but the role of the skeletal muscle clock has never been explored. We generated conditional and inducible mouse lines with muscle-specific ablation of the core clock gene Bmal1. Skeletal muscles from these mice showed impaired insulin-stimulated glucose uptake with reduced protein levels of GLUT4, the insulin-dependent glucose transporter, and TBC1D1, a Rab-GTPase involved in GLUT4 translocation. Pyruvate dehydrogenase (PDH) activity was also reduced due to altered expression of circadian genes Pdk4 and Pdp1, coding for PDH kinase and phosphatase, respectively. PDH inhibition leads to reduced glucose oxidation and diversion of glycolytic intermediates to alternative metabolic pathways, as revealed by metabolome analysis. The impaired glucose metabolism induced by muscle-specific Bmal1 knockout suggests that a major physiological role of the muscle clock is to prepare for the transition from the rest/fasting phase to the active/feeding phase, when glucose becomes the predominant fuel for skeletal muscle. [Copyright &y& Elsevier]

Summary: The dogma that life without insulin is incompatible has recently been challenged by results showing the viability of insulin-deficient rodents undergoing leptin monotherapy. Yet, the mechanisms underlying these actions of leptin are unknown. Here, the metabolic outcomes of intracerebroventricular (i.c.v.) administration of leptin in mice devoid of insulin and lacking or re-expressing leptin receptors (LEPRs) only in selected neuronal groups were assessed. Our results demonstrate that concomitant re-expression of LEPRs only in hypothalamic γ-aminobutyric acid (GABA) and pro-opiomelanocortin (POMC) neurons is sufficient to fully mediate the lifesaving and antidiabetic actions of leptin in insulin deficiency. Our analyses indicate that enhanced glucose uptake by brown adipose tissue and soleus muscle, as well as improved hepatic metabolism, underlies these effects of leptin. Collectively, our data elucidate a hypothalamic-dependent pathway enabling life without insulin and hence pave the way for developing better treatments for diseases of insulin deficiency. [Copyright &y& Elsevier]

Abstract: Autoencoders are unsupervised machine learning circuits, with typically one hidden layer, whose learning goal is to minimize an average distortion measure between inputs and outputs. Linear autoencoders correspond to the special case where only linear transformations between visible and hidden variables are used. While linear autoencoders can be defined over any field, only real-valued linear autoencoders have been studied so far. Here we study complex-valued linear autoencoders where the components of the training vectors and adjustable matrices are defined over the complex field with the norm. We provide simpler and more general proofs that unify the real-valued and complex-valued cases, showing that in both cases the landscape of the error function is invariant under certain groups of transformations. The landscape has no local minima, a family of global minima associated with Principal Component Analysis, and many families of saddle points associated with orthogonal projections onto sub-space spanned by sub-optimal subsets of eigenvectors of the covariance matrix. The theory yields several iterative, convergent, learning algorithms, a clear understanding of the generalization properties of the trained autoencoders, and can equally be applied to the hetero-associative case when external targets are provided. Partial results on deep architecture as well as the differential geometry of autoencoders are also presented. The general framework described here is useful to classify autoencoders and identify general properties that ought to be investigated for each class, illuminating some of the connections between autoencoders, unsupervised learning, clustering, Hebbian learning, and information theory. [Copyright &y& Elsevier]

Abstract: The amount of information contained in a piece of data can be measured by the effect this data has on its observer. Fundamentally, this effect is to transform the observer’s prior beliefs into posterior beliefs, according to Bayes theorem. Thus the amount of information can be measured in a natural way by the distance (relative entropy) between the prior and posterior distributions of the observer over the available space of hypotheses. This facet of information, termed “surprise”, is important in dynamic situations where beliefs change, in particular during learning and adaptation. Surprise can often be computed analytically, for instance in the case of distributions from the exponential family, or it can be numerically approximated. During sequential Bayesian learning, surprise decreases as the inverse of the number of training examples. Theoretical properties of surprise are discussed, in particular how it differs and complements Shannon’s definition of information. A computer vision neural network architecture is then presented capable of computing surprise over images and video stimuli. Hypothesizing that surprising data ought to attract natural or artificial attention systems, the output of this architecture is used in a psychophysical experiment to analyze human eye movements in the presence of natural video stimuli. Surprise is found to yield robust performance at predicting human gaze (ROC-like ordinal dominance score ∼0.7 compared to ∼0.8 for human inter-observer repeatability, ∼0.6 for simpler intensity contrast-based predictor, and 0.5 for chance). The resulting theory of surprise is applicable across different spatio-temporal scales, modalities, and levels of abstraction. [Copyright &y& Elsevier]

Abstract: We propose a formal Bayesian definition of surprise to capture subjective aspects of sensory information. Surprise measures how data affects an observer, in terms of differences between posterior and prior beliefs about the world. Only data observations which substantially affect the observer’s beliefs yield surprise, irrespectively of how rare or informative in Shannon’s sense these observations are. We test the framework by quantifying the extent to which humans may orient attention and gaze towards surprising events or items while watching television. To this end, we implement a simple computational model where a low-level, sensory form of surprise is computed by simple simulated early visual neurons. Bayesian surprise is a strong attractor of human attention, with 72% of all gaze shifts directed towards locations more surprising than the average, a figure rising to 84% when focusing the analysis onto regions simultaneously selected by all observers. The proposed theory of surprise is applicable across different spatio-temporal scales, modalities, and levels of abstraction. [Copyright &y& Elsevier]

Tissue sensitivity and response to exercise vary according to the time of day and alignment of circadian clocks, but the optimal exercise time to elicit a desired metabolic outcome is not fully defined. To understand how tissues independently and collectively respond to timed exercise, we applied a systems biology approach. We mapped and compared global metabolite responses of seven different mouse tissues and serum after an acute exercise bout performed at different times of the day. Comparative analyses of intra- and inter-tissue metabolite dynamics, including temporal profiling and blood sampling across liver and hindlimb muscles, uncovered an unbiased view of local and systemic metabolic responses to exercise unique to time of day. This comprehensive atlas of exercise metabolism provides clarity and physiological context regarding the production and distribution of canonical and novel time-dependent exerkine metabolites, such as 2-hydroxybutyrate (2-HB), and reveals insight into the health-promoting benefits of exercise on metabolism. [Display omitted] • Time of exercise defines system-wide activation of energy metabolism • Exercise timing rewires intra-tissue and inter-tissue metabolite correlations • Maintenance of inter-tissue metabostasis is specified by exercise time • Comparative analyses reveal time- and tissue-dependent exerkines Tissue sensitivity and response to exercise vary according to time of day. Sato et al. present an atlas of exercise metabolism, including global metabolomics profiling of multiple tissues, combined with arteriovenous sampling of hindlimb muscle and sampling across the liver to verify net uptake and release of time- and exercise-dependent signaling biochemicals. [ABSTRACT FROM AUTHOR]

Abstract: Go is an ancient board game that poses unique opportunities and challenges for artificial intelligence. Currently, there are no computer Go programs that can play at the level of a good human player. However, the emergence of large repositories of games is opening the door for new machine learning approaches to address this challenge. Here we develop a machine learning approach to Go, and related board games, focusing primarily on the problem of learning a good evaluation function in a scalable way. Scalability is essential at multiple levels, from the library of local tactical patterns, to the integration of patterns across the board, to the size of the board itself. The system we propose is capable of automatically learning the propensity of local patterns from a library of games. Propensity and other local tactical information are fed into recursive neural networks, derived from a probabilistic Bayesian network architecture. The recursive neural networks in turn integrate local information across the board in all four cardinal directions and produce local outputs that represent local territory ownership probabilities. The aggregation of these probabilities provides an effective strategic evaluation function that is an estimate of the expected area at the end, or at various other stages, of the game. Local area targets for training can be derived from datasets of games played by human players. In this approach, while requiring a learning time proportional to , skills learned on a board of size can easily be transferred to boards of other sizes. A system trained using only 9×9 amateur game data performs surprisingly well on a test set derived from 19×19 professional game data. Possible directions for further improvements are briefly discussed. [Copyright &y& Elsevier]

Abstract: Machine learning methods that can handle variable-size structured data such as sequences and graphs include Bayesian networks (BNs) and Recursive Neural Networks (RNNs). In both classes of models, the data is modeled using a set of observed and hidden variables associated with the nodes of a directed acyclic graph. In BNs, the conditional relationships between parent and child variables are probabilistic, whereas in RNNs they are deterministic and parameterized by neural networks. Here, we study the formal relationship between both classes of models and show that when the source nodes variables are observed, RNNs can be viewed as limits, both in distribution and probability, of BNs with local conditional distributions that have vanishing covariance matrices and converge to delta functions. Conditions for uniform convergence are also given together with an analysis of the behavior and exactness of Belief Propagation (BP) in ‘deterministic’ BNs. Implications for the design of mixed architectures and the corresponding inference algorithms are briefly discussed. [Copyright &y& Elsevier]

While the timing of food intake is important, it is unclear whether the effects of exercise on energy metabolism are restricted to unique time windows. As circadian regulation is key to controlling metabolism, understanding the impact of exercise performed at different times of the day is relevant for physiology and homeostasis. Using high-throughput transcriptomic and metabolomic approaches, we identify distinct responses of metabolic oscillations that characterize exercise in either the early rest phase or the early active phase in mice. Notably, glycolytic activation is specific to exercise at the active phase. At the molecular level, HIF1α, a central regulator of glycolysis during hypoxia, is selectively activated in a time-dependent manner upon exercise, resulting in carbohydrate exhaustion, usage of alternative energy sources, and adaptation of systemic energy expenditure. Our findings demonstrate that the time of day is a critical factor to amplify the beneficial impact of exercise on both metabolic pathways within skeletal muscle and systemic energy homeostasis. • Distinct response of metabolic cycles in skeletal muscle to time-of-day exercise • Early active phase exercise exerts a robust metabolic response in skeletal muscle • The metabolic response includes glycolysis, lipid oxidation, and BCAA breakdown • Time of exercise specifies the activation of HIF1α and systemic energy expenditure Sato and colleagues investigate time-dependent impact of exercise on skeletal muscle, revealing altered daily metabolic cycles after exercise specific to the time of day. Exercise at the early active phase robustly activates the HIF1α pathway, followed by glycolytic activation, usage of alternative fuels, and adaptation of systemic energy expenditure. [ABSTRACT FROM AUTHOR]

Loss of synchrony between geophysical time and insulin action predisposes to metabolic diseases. Yet the brain and peripheral pathways linking proper insulin effect to diurnal changes in light-dark and feeding-fasting inputs are poorly understood. Here, we show that the insulin sensitivity of several metabolically relevant tissues fluctuates during the 24 h period. For example, in mice, the insulin sensitivity of skeletal muscle, liver, and adipose tissue is lowest during the light period. Mechanistically, by performing loss- and gain-of-light-action and food-restriction experiments, we demonstrate that SIRT1 in steroidogenic factor 1 (SF1) neurons of the ventromedial hypothalamic nucleus (VMH) convey photic inputs to entrain the biochemical and metabolic action of insulin in skeletal muscle. These findings uncover a critical light-SF1-neuron-skeletal-muscle axis that acts to finely tune diurnal changes in insulin sensitivity and reveal a light regulatory mechanism of skeletal muscle function. • Response to insulin of metabolically relevant tissues varies in a diurnal fashion • SF1 neurons govern diurnal variation in responsiveness to insulin in skeletal muscle • Light inputs entrain diurnal variation in responsiveness to insulin in skeletal muscle • Photic gain- or loss-of-function causes insulin resistance in skeletal muscle Aras et al. provide in vivo evidence that tissue responsiveness to insulin varies in a diurnal fashion. In skeletal muscle, the authors show that photic inputs entrain diurnal changes in clock genes expression and insulin sensitivity via SIRT1 in neurons within the hypothalamic ventromedial nucleus. [ABSTRACT FROM AUTHOR]

We introduce SPLASH units, a class of learnable activation functions shown to simultaneously improve the accuracy of deep neural networks while also improving their robustness to adversarial attacks. SPLASH units have both a simple parameterization and maintain the ability to approximate a wide range of non-linear functions. SPLASH units are: (1) continuous; (2) grounded (f (0) = 0); (3) use symmetric hinges; and (4) their hinges are placed at fixed locations which are derived from the data (i.e. no learning required). Compared to nine other learned and fixed activation functions, including ReLU and its variants, SPLASH units show superior performance across three datasets (MNIST, CIFAR-10, and CIFAR-100) and four architectures (LeNet5, All-CNN, ResNet-20, and Network-in-Network). Furthermore, we show that SPLASH units significantly increase the robustness of deep neural networks to adversarial attacks. Our experiments on both black-box and white-box adversarial attacks show that commonly-used architectures, namely LeNet5, All-CNN, Network-in-Network, and ResNet-20, can be up to 31% more robust to adversarial attacks by simply using SPLASH units instead of ReLUs. Finally, we show the benefits of using SPLASH activation functions in bigger architectures designed for non-trivial datasets such as ImageNet. [ABSTRACT FROM AUTHOR]

Summary The circadian clock operates as intrinsic time-keeping machinery to preserve homeostasis in response to the changing environment. While food is a known zeitgeber for clocks in peripheral tissues, it remains unclear how lack of food influences clock function. We demonstrate that the transcriptional response to fasting operates through molecular mechanisms that are distinct from time-restricted feeding regimens. First, fasting affects core clock genes and proteins, resulting in blunted rhythmicity of BMAL1 and REV-ERBα both in liver and skeletal muscle. Second, fasting induces a switch in temporal gene expression through dedicated fasting-sensitive transcription factors such as GR, CREB, FOXO, TFEB, and PPARs. Third, the rhythmic genomic response to fasting is sustainable by prolonged fasting and reversible by refeeding. Thus, fasting imposes specialized dynamics of transcriptional coordination between the clock and nutrient-sensitive pathways, thereby achieving a switch to fasting-specific temporal gene regulation. Graphical Abstract Highlights • Transcriptional response to fasting is robustly rhythmic in liver and muscle • Lack of food fails to sustain "free-running" conditions of peripheral circadian clocks • Genes are temporally regulated by the clock and fasting-related transcription factors • Rhythmic response to fasting is reversible by refeeding Kinouchi et al. reveal that fasting affects peripheral circadian clocks in the liver and skeletal muscle. Fasting operates by influencing the circadian clock and fasting-sensitive transcription factors, thereby cooperatively achieving fasting-specific temporal gene regulation. [ABSTRACT FROM AUTHOR]

Background & Aims The benefit of colonoscopy for colorectal cancer prevention depends on the adenoma detection rate (ADR). The ADR should reflect the adenoma prevalence rate, which is estimated to be higher than 50% in the screening-age population. However, the ADR by colonoscopists varies from 7% to 53%. It is estimated that every 1% increase in ADR lowers the risk of interval colorectal cancers by 3%–6%. New strategies are needed to increase the ADR during colonoscopy. We tested the ability of computer-assisted image analysis using convolutional neural networks (CNNs; a deep learning model for image analysis) to improve polyp detection, a surrogate of ADR. Methods We designed and trained deep CNNs to detect polyps using a diverse and representative set of 8,641 hand-labeled images from screening colonoscopies collected from more than 2000 patients. We tested the models on 20 colonoscopy videos with a total duration of 5 hours. Expert colonoscopists were asked to identify all polyps in 9 de-identified colonoscopy videos, which were selected from archived video studies, with or without benefit of the CNN overlay. Their findings were compared with those of the CNN using CNN-assisted expert review as the reference. Results When tested on manually labeled images, the CNN identified polyps with an area under the receiver operating characteristic curve of 0.991 and an accuracy of 96.4%. In the analysis of colonoscopy videos in which 28 polyps were removed, 4 expert reviewers identified 8 additional polyps without CNN assistance that had not been removed and identified an additional 17 polyps with CNN assistance (45 in total). All polyps removed and identified by expert review were detected by the CNN. The CNN had a false-positive rate of 7%. Conclusion In a set of 8,641 colonoscopy images containing 4,088 unique polyps, the CNN identified polyps with a cross-validation accuracy of 96.4% and an area under the receiver operating characteristic curve of 0.991. The CNN system detected and localized polyps well within real-time constraints using an ordinary desktop machine with a contemporary graphics processing unit. This system could increase the ADR and decrease interval colorectal cancers but requires validation in large multicenter trials. Graphical abstract [ABSTRACT FROM AUTHOR]

Summary The circadian clock orchestrates rhythms in physiology and behavior, allowing organismal adaptation to daily environmental changes. While food intake profoundly influences diurnal rhythms in the liver, how nutritional challenges are differentially interpreted by distinct tissue-specific clocks remains poorly explored. Ketogenic diet (KD) is considered to have metabolic and therapeutic value, though its impact on circadian homeostasis is virtually unknown. We show that KD has profound and differential effects on liver and intestine clocks. Specifically, the amplitude of clock-controlled genes and BMAL1 chromatin recruitment are drastically altered by KD in the liver, but not in the intestine. KD induces nuclear accumulation of PPARα in both tissues but with different circadian phase. Also, gut and liver clocks respond differently to carbohydrate supplementation to KD. Importantly, KD induces serum and intestinal β-hydroxyl-butyrate levels to robustly oscillate in a circadian manner, an event coupled to tissue-specific cyclic histone deacetylase (HDAC) activity and histone acetylation. [ABSTRACT FROM AUTHOR]

Weight-sharing is one of the pillars behind Convolutional Neural Networks and their successes. However, in physical neural systems such as the brain, weight-sharing is implausible. This discrepancy raises the fundamental question of whether weight-sharing is necessary. If so, to which degree of precision? If not, what are the alternatives? The goal of this study is to investigate these questions, primarily through simulations where the weight-sharing assumption is relaxed. Taking inspiration from neural circuitry, we explore the use of Free Convolutional Networks and neurons with variable connection patterns. Using Free Convolutional Networks, we show that while weight-sharing is a pragmatic optimization approach, it is not a necessity in computer vision applications. Furthermore, Free Convolutional Networks match the performance observed in standard architectures when trained using properly translated data (akin to video). Under the assumption of translationally augmented data, Free Convolutional Networks learn translationally invariant representations that yield an approximate form of weight-sharing. [ABSTRACT FROM AUTHOR]

Prediction models of post-liver transplant mortality are crucial so that donor organs are not allocated to recipients with unreasonably high probabilities of mortality. Machine learning algorithms, particularly deep neural networks (DNNs), can often achieve higher predictive performance than conventional models. In this study, we trained a DNN to predict 90-day post-transplant mortality using preoperative variables and compared the performance to that of the Survival Outcomes Following Liver Transplantation (SOFT) and Balance of Risk (BAR) scores, using United Network of Organ Sharing data on adult patients who received a deceased donor liver transplant between 2005 and 2015 (n = 57,544). The DNN was trained using 202 features, and the best DNN's architecture consisted of 5 hidden layers with 110 neurons each. The area under the receiver operating characteristics curve (AUC) of the best DNN model was 0.703 (95% CI: 0.682-0.726) as compared to 0.655 (95% CI: 0.633-0.678) and 0.688 (95% CI: 0.667-0.711) for the BAR score and SOFT score, respectively. In conclusion, despite the complexity of DNN, it did not achieve a significantly higher discriminative performance than the SOFT score. Future risk models will likely benefit from the inclusion of other data sources, including high-resolution clinical features for which DNNs are particularly apt to outperform conventional statistical methods. • Liver transplant mortality can be predicted by using deep neural networks that use UNOS data. • Deep neural networks outperformed the BAR score in predicting mortality. • Deep neural networks did not significantly outperform the SOFT score. • Deep learning using higher resolution features may achieve superior performance. [ABSTRACT FROM AUTHOR]

Background: Rapid, preoperative identification of patients with the highest risk for medical complications is necessary to ensure that limited infrastructure and human resources are directed towards those most likely to benefit. Existing risk scores either lack specificity at the patient level or utilise the American Society of Anesthesiologists (ASA) physical status classification, which requires a clinician to review the chart.Methods: We report on the use of machine learning algorithms, specifically random forests, to create a fully automated score that predicts postoperative in-hospital mortality based solely on structured data available at the time of surgery. Electronic health record data from 53 097 surgical patients (2.01% mortality rate) who underwent general anaesthesia between April 1, 2013 and December 10, 2018 in a large US academic medical centre were used to extract 58 preoperative features.Results: Using a random forest classifier we found that automatically obtained preoperative features (area under the curve [AUC] of 0.932, 95% confidence interval [CI] 0.910-0.951) outperforms Preoperative Score to Predict Postoperative Mortality (POSPOM) scores (AUC of 0.660, 95% CI 0.598-0.722), Charlson comorbidity scores (AUC of 0.742, 95% CI 0.658-0.812), and ASA physical status (AUC of 0.866, 95% CI 0.829-0.897). Including the ASA physical status with the preoperative features achieves an AUC of 0.936 (95% CI 0.917-0.955).Conclusions: This automated score outperforms the ASA physical status score, the Charlson comorbidity score, and the POSPOM score for predicting in-hospital mortality. Additionally, we integrate this score with a previously published postoperative score to demonstrate the extent to which patient risk changes during the perioperative period. [ABSTRACT FROM AUTHOR]

Exercise has emerged as a powerful variable that can improve cognitive function and delay age-associated cognitive decline and Alzheimer's disease (AD); however, the underlying mechanisms are poorly understood. To determine if protective mechanisms may occur at the transcriptional level, we used microarrays to investigate the relationship between physical activity levels and gene expression patterns in the cognitively intact aged human hippocampus. In parallel, hippocampal gene expression patterns associated with aging and AD were assessed using publicly available microarray data profiling hippocampus from young (20–59 years), cognitively intact aging (73–95 years) and age-matched AD cases. To identify "anti-aging/AD" transcription patterns associated with physical activity, probesets significantly associated with both physical activity and aging/AD were identified and their directions of expression change in each condition were compared. Remarkably, of the 2210 probesets significant in both data sets, nearly 95% showed opposite transcription patterns with physical activity compared with aging/AD. The majority (>70%) of these anti-aging/AD genes showed increased expression with physical activity and decreased expression in aging/AD. Enrichment analysis of the anti-aging/AD genes showing increased expression in association with physical activity revealed strong overrepresentation of mitochondrial energy production and synaptic function, along with axonal function and myelin integrity. Synaptic genes were notably enriched for synaptic vesicle priming, release and recycling, glutamate and GABA signaling, and spine plasticity. Anti-aging/AD genes showing decreased expression in association with physical activity were enriched for transcription-related function (notably negative regulation of transcription). These data reveal that physical activity is associated with a more youthful profile in the hippocampus across multiple biological processes, providing a potential molecular foundation for how physical activity can delay age- and AD-related decline of hippocampal function. • Physical activity is extensively associated with gene expression in the aged human hippocampus. • Gene expression patterns associated with physical activity restore a more youthful state in the hippocampus and opposed transcription patterns associated with aging and Alzheimer's disease. • Gene expression is primarily increased, with genes targeting multiple core cellular functions. • Genes showed robust enrichment for mitochondrial energy production, synaptic signaling mechanisms, axon and white matter health, and control of transcription. [ABSTRACT FROM AUTHOR]

The circadian clock regulates a wide range of physiological and metabolic processes, and its disruption leads to metabolic disorders such as diabetes and obesity. Accumulating evidence reveals that the circadian clock regulates levels of metabolites that, in turn, may regulate the clock. Here we demonstrate that the circadian clock regulates the intracellular levels of acetyl-CoA by modulating the enzymatic activity of acetyl-CoA Synthetase 1 (AceCS1). Acetylation of AceCS1 controls the activity of the enzyme. We show that acetylation of AceCS1 is cyclic and that its rhythmicity requires a functional circadian clock and the NAD+ -dependent deacetylase SIRT1. Cyclic acetylation of AceCS1 contributes to the rhythmicity of acetyl-CoA levels both in vivo and in cultured cells. Down-regulation of AceCS1 causes a significant decrease in the cellular acetyl-CoA pool, leading to reduction in circadian changes in fatty acid elongation. Thus, a nontranscriptional, enzymatic loop is governed by the circadian clock to control acetyl-CoA levels and fatty acid synthesis. [ABSTRACT FROM AUTHOR]

Abstract: Specific docking interactions between mitogen-activated protein kinases (MAPKs), their regulators, and their downstream substrates, are crucial for efficient and accurate signal transmission. To identify novel substrates of the c-Jun N-terminal kinase (JNK) family of MAPKs, we searched the human genome for proteins that contained (1), a predicted JNK-docking site (D-site); and (2), a cluster of putative JNK target phosphosites located close to the D-site. Here we describe a novel JNK substrate that emerged from this analysis, the functionally uncharacterized protein smoothelin-like 2 (SMTNL2). SMTNL2 protein bound with high-affinity to multiple MAPKs including JNK1-3 and ERK2; furthermore, the identity of conserved amino acids in the predicted docking site (residues 180–193) was necessary for this high-affinity binding. In addition, purified full-length SMTNL2 protein was phosphorylated by JNK1-3 in vitro, and this required the integrity of the D-site. Using mass spectrometry and mutagenesis, we identified four D-site-dependent phosphoacceptor sites in close proximity to the docking site, at S217, S241, T236 and T239. A short peptide comprised of the SMTNL2 D-site inhibited JNK-mediated phosphorylation of the ATF2 transcription factor, showing that SMTNL2 can compete with other substrates for JNK binding. Moreover, when transfected into HEK293 cells, SMTNL2 was phosphorylated by endogenous JNK in a D-site dependent manner, on the same residues identified in vitro. SMTNL2 protein was expressed in many mammalian tissues, with a notably high expression in skeletal muscle. Consistent with the hypothesis that SMTNL2 has a function in skeletal muscle, SMTNL2 protein expression was strongly induced during the transition from myoblasts to myotubes in differentiating C2C12 cells. [Copyright &y& Elsevier]

Abstract: Chlamydia trachomatis is the most common bacterial sexually transmitted pathogen in the world. In order to control this infection there is an urgent need to formulate a vaccine. Identification of protective antigens is required to implement a subunit vaccine. To identify potential antigen vaccine candidates, three strains of mice, BALB/c, C3H/HeN and C57BL/6, were inoculated with live and inactivated C. trachomatis mouse pneumonitis (MoPn) by different routes of immunization. Using a protein microarray, serum samples collected after immunization were tested for the presence of antibodies against specific chlamydial antigens. A total of 225 open reading frames (ORF) of the C. trachomatis genome were cloned, expressed, and printed in the microarray. Using this protein microarray, a total of seven C. trachomatis dominant antigens were identified (TC0052, TC0189, TC0582, TC0660, TC0726, TC0816 and, TC0828) as recognized by IgG antibodies from all three strains of animals after immunization. In addition, the microarray was probed to determine if the antibody response exhibited a Th1 or Th2 bias. Animals immunized with live organisms mounted a predominant Th1 response against most of the chlamydial antigens while mice immunized with inactivated Chlamydia mounted a Th2-biased response. In conclusion, using a high throughput protein microarray we have identified a set of novel proteins that can be tested for their ability to protect against a chlamydial infection. [Copyright &y& Elsevier]

Abstract: Increased availability of large repositories of chemical compounds is creating new challenges and opportunities for the application of machine learning methods to problems in computational chemistry and chemical informatics. Because chemical compounds are often represented by the graph of their covalent bonds, machine learning methods in this domain must be capable of processing graphical structures with variable size. Here, we first briefly review the literature on graph kernels and then introduce three new kernels (Tanimoto, MinMax, Hybrid) based on the idea of molecular fingerprints and counting labeled paths of depth up to d using depth-first search from each possible vertex. The kernels are applied to three classification problems to predict mutagenicity, toxicity, and anti-cancer activity on three publicly available data sets. The kernels achieve performances at least comparable, and most often superior, to those previously reported in the literature reaching accuracies of 91.5% on the Mutag dataset, 65–67% on the PTC (Predictive Toxicology Challenge) dataset, and 72% on the NCI (National Cancer Institute) dataset. Properties and tradeoffs of these kernels, as well as other proposed kernels that leverage 1D or 3D representations of molecules, are briefly discussed. [Copyright &y& Elsevier]

The ArcAB two-component system of Escherichia coli regulates the aerobic/anaerobic expression of genes that encode respiratory proteins whose synthesis is coordinated during aerobic/anaerobic cell growth. A genomic study of E. coli was undertaken to identify other potential targets of oxygen and ArcA regulation. A group of 175 genes generated from this study and our previous study on oxygen regulation (Salmon, K., Hung, S. P., Mekjian, K., Baldi, P., Hatfield, G. W., and Gunsalus, R. P. (2003) J. Biol. Chem. 278, 29837–29855), called our gold standard gene set, have p values <0.00013 and a posterior probability of differential expression value of 0.99. These 175 genes clustered into eight expression patterns and represent genes involved in a large number of cell processes, including small molecule biosynthesis, macromolecular synthesis, and aerobic/anaerobic respiration and fermentation. In addition, 119 of these 175 genes were also identified in our previous study of the fnr allele. A MEME/weight matrix method was used to identify a new putative ArcA-binding site for all genes of the E. coli genome. 16 new sites were identified upstream of genes in our gold standard set. The strict statistical analyses that we have performed on our data allow us to predict that 1139 genes in the E. coli genome are regulated either directly or indirectly by the ArcA protein with a 99% confidence level. [ABSTRACT FROM AUTHOR]

The work presented here is a first step toward a long term goal of systems biology, the complete elucidation of the gene regulatory networks of a living organism. To this end, we have employed DNA microarray technology to identify genes involved in the regulatory networks that facilitate the transition of Escherichia coli cells from an aerobic to an anaerobic growth state. We also report the identification of a subset of these genes that are regulated by a global regulatory protein for anaerobic metabolism, FNR. Analysis of these data demonstrated that the expression of over one-third of the genes expressed during growth under aerobic conditions are altered when E. coli cells transition to an anaerobic growth state, and that the expression of 712 (49%) of these genes are either directly or indirectly modulated by FNR. The results presented here also suggest interactions between the FNR and the leucine-responsive regulatory protein (Lrp) regulatory networks. Because computational methods to analyze and interpret high dimensional DNA microarray data are still at an early stage, and because basic issues of data analysis are still being sorted out, much of the emphasis of this work is directed toward the development of methods to identify differentially expressed genes with a high level of confidence. In particular, we describe an approach for identifying gene expression patterns (clusters) obtained from multiple perturbation experiments based on a subset of genes that exhibit high probability for differential expression values. [ABSTRACT FROM AUTHOR]

Primary cilia function as cells' antennas to detect and transduce external stimuli and play crucial roles in cell signaling and communication. The vast majority of cilia genes that are causally linked with ciliopathies are also associated with neurological deficits, such as cognitive impairments. Yet, the roles of cilia dysfunctions in the pathogenesis of psychiatric disorders have not been studied. Our aim is to identify patterns of cilia gene dysregulation in the four major psychiatric disorders: schizophrenia (SCZ), autism spectrum disorder (ASD), bipolar disorder (BP), and major depressive disorder (MDD). For this purpose, we acquired differentially expressed genes (DEGs) from the largest and most recent publicly available databases. We found that 42%, 24%, 17%, and 15% of brain-expressed cilia genes were significantly differentially expressed in SCZ, ASD, BP, and MDD, respectively. Several genes exhibited cross-disorder overlap, suggesting that typical cilia signaling pathways' dysfunctions determine susceptibility to more than one psychiatric disorder or may partially underlie their pathophysiology. Our study revealed that genes encoding proteins of almost all sub-cilia structural and functional compartments were dysregulated in the four psychiatric disorders. Strikingly, the genes of 75% of cilia GPCRs and 50% of the transition zone proteins were differentially expressed in SCZ. The present study is the first to draw associations between cilia and major psychiatric disorders, and is the first step toward understanding the role that cilia components play in their pathophysiological processes, which may lead to novel therapeutic targets for these disorders. • Cilia genes' mutations that cause ciliopathies are associated with neurological defect such as cognitive dysfunctions • We demonstrate primary cilia' gene dysregulations in major psychiatric disorders: schizophrenia, autism, bipolar disorder, and depression • We provide an evidence for primary cilia' potential role as possible pathophysiological cause or therapeutic target in psychiatric disorders [ABSTRACT FROM AUTHOR]