Showing papers by "University of Toronto published in 2012"

ImageNet Classification with Deep Convolutional Neural Networks

Abstract: We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 different classes. On the test data, we achieved top-1 and top-5 error rates of 37.5% and 17.0% which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully-connected layers with a final 1000-way softmax. To make training faster, we used non-saturating neurons and a very efficient GPU implementation of the convolution operation. To reduce overriding in the fully-connected layers we employed a recently-developed regularization method called "dropout" that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3%, compared to 26.2% achieved by the second-best entry.

Deep Neural Networks for Acoustic Modeling in Speech Recognition: The Shared Views of Four Research Groups

Abstract: Most current speech recognition systems use hidden Markov models (HMMs) to deal with the temporal variability of speech and Gaussian mixture models (GMMs) to determine how well each state of each HMM fits a frame or a short window of frames of coefficients that represents the acoustic input. An alternative way to evaluate the fit is to use a feed-forward neural network that takes several frames of coefficients as input and produces posterior probabilities over HMM states as output. Deep neural networks (DNNs) that have many hidden layers and are trained using new methods have been shown to outperform GMMs on a variety of speech recognition benchmarks, sometimes by a large margin. This article provides an overview of this progress and represents the shared views of four research groups that have had recent successes in using DNNs for acoustic modeling in speech recognition.

Acute respiratory distress syndrome: the Berlin Definition.

Abstract: The acute respiratory distress syndrome (ARDS) was defined in 1994 by the American-European Consensus Conference (AECC); since then, issues regarding the reliability and validity of this definition have emerged. Using a consensus process, a panel of experts convened in 2011 (an initiative of the European Society of Intensive Care Medicine endorsed by the American Thoracic Society and the Society of Critical Care Medicine) developed the Berlin Definition, focusing on feasibility, reliability, validity, and objective evaluation of its performance. A draft definition proposed 3 mutually exclusive categories of ARDS based on degree of hypoxemia: mild (200 mm Hg < PaO2/FIO2 ≤ 300 mm Hg), moderate (100 mm Hg < PaO2/FIO2 ≤ 200 mm Hg), and severe (PaO2/FIO2 ≤ 100 mm Hg) and 4 ancillary variables for severe ARDS: radiographic severity, respiratory system compliance (≤40 mL/cm H2O), positive end-expiratory pressure (≥10 cm H2O), and corrected expired volume per minute (≥10 L/min). The draft Berlin Definition was empirically evaluated using patient-level meta-analysis of 4188 patients with ARDS from 4 multicenter clinical data sets and 269 patients with ARDS from 3 single-center data sets containing physiologic information. The 4 ancillary variables did not contribute to the predictive validity of severe ARDS for mortality and were removed from the definition. Using the Berlin Definition, stages of mild, moderate, and severe ARDS were associated with increased mortality (27%; 95% CI, 24%-30%; 32%; 95% CI, 29%-34%; and 45%; 95% CI, 42%-48%, respectively; P < .001) and increased median duration of mechanical ventilation in survivors (5 days; interquartile [IQR], 2-11; 7 days; IQR, 4-14; and 9 days; IQR, 5-17, respectively; P < .001). Compared with the AECC definition, the final Berlin Definition had better predictive validity for mortality, with an area under the receiver operating curve of 0.577 (95% CI, 0.561-0.593) vs 0.536 (95% CI, 0.520-0.553; P < .001). This updated and revised Berlin Definition for ARDS addresses a number of the limitations of the AECC definition. The approach of combining consensus discussions with empirical evaluation may serve as a model to create more accurate, evidence-based, critical illness syndrome definitions and to better inform clinical care, research, and health services planning.

Improving neural networks by preventing co-adaptation of feature detectors

Abstract: When a large feedforward neural network is trained on a small training set, it typically performs poorly on held-out test data. This "overfitting" is greatly reduced by randomly omitting half of the feature detectors on each training case. This prevents complex co-adaptations in which a feature detector is only helpful in the context of several other specific feature detectors. Instead, each neuron learns to detect a feature that is generally helpful for producing the correct answer given the combinatorially large variety of internal contexts in which it must operate. Random "dropout" gives big improvements on many benchmark tasks and sets new records for speech and object recognition.

Practical Bayesian Optimization of Machine Learning Algorithms

Abstract: The use of machine learning algorithms frequently involves careful tuning of learning parameters and model hyperparameters. Unfortunately, this tuning is often a "black art" requiring expert experience, rules of thumb, or sometimes brute-force search. There is therefore great appeal for automatic approaches that can optimize the performance of any given learning algorithm to the problem at hand. In this work, we consider this problem through the framework of Bayesian optimization, in which a learning algorithm's generalization performance is modeled as a sample from a Gaussian process (GP). We show that certain choices for the nature of the GP, such as the type of kernel and the treatment of its hyperparameters, can play a crucial role in obtaining a good optimizer that can achieve expertlevel performance. We describe new algorithms that take into account the variable cost (duration) of learning algorithm experiments and that can leverage the presence of multiple cores for parallel experimentation. We show that these proposed algorithms improve on previous automatic procedures and can reach or surpass human expert-level optimization for many algorithms including latent Dirichlet allocation, structured SVMs and convolutional neural networks.

Audit and feedback: effects on professional practice and healthcare outcomes

Abstract: Background Audit and feedback continues to be widely used as a strategy to improve professional practice. It appears logical that healthcare professionals would be prompted to modify their practice if given feedback that their clinical practice was inconsistent with that of their peers or accepted guidelines. Yet, audit and feedback has not been found to be consistently effective. Objectives To assess the effects of audit and feedback on the practice of healthcare professionals and patient outcomes. Search strategy We searched the Cochrane Effective Practice and Organisation of Care Group's register up to January 2001. This was supplemented with searches of MEDLINE and reference lists, which did not yield additional relevant studies. Selection criteria Randomised trials of audit and feedback (defined as any summary of clinical performance over a specified period of time) that reported objectively measured professional practice in a healthcare setting or healthcare outcomes. Data collection and analysis Two reviewers independently extracted data and assessed study quality. Quantitative (meta-regression), visual and qualitative analyses were undertaken. Main results We included 85 studies, 48 of which have been added to the previous version of this review. There were 52 comparisons of dichotomous outcomes from 47 trials with over 3500 health professionals that compared audit and feedback to no intervention. The adjusted RDs of non-compliance with desired practice varied from 0.09 (a 9% absolute increase in non-compliance) to 0.71 (a 71% decrease in non-compliance) (median = 0.07, inter-quartile range = 0.02 to 0.11). The one factor that appeared to predict the effectiveness of audit and feedback across studies was baseline non-compliance with recommended practice. Reviewer's conclusions Audit and feedback can be effective in improving professional practice. When it is effective, the effects are generally small to moderate. The absolute effects of audit and feedback are more likely to be larger when baseline adherence to recommended practice is low.

Guidelines for the use and interpretation of assays for monitoring autophagy

Abstract: In 2008 we published the first set of guidelines for standardizing research in autophagy. Since then, research on this topic has continued to accelerate, and many new scientists have entered the field. Our knowledge base and relevant new technologies have also been expanding. Accordingly, it is important to update these guidelines for monitoring autophagy in different organisms. Various reviews have described the range of assays that have been used for this purpose. Nevertheless, there continues to be confusion regarding acceptable methods to measure autophagy, especially in multicellular eukaryotes. A key point that needs to be emphasized is that there is a difference between measurements that monitor the numbers or volume of autophagic elements (e.g., autophagosomes or autolysosomes) at any stage of the autophagic process vs. those that measure flux through the autophagy pathway (i.e., the complete process); thus, a block in macroautophagy that results in autophagosome accumulation needs to be differentiated from stimuli that result in increased autophagic activity, defined as increased autophagy induction coupled with increased delivery to, and degradation within, lysosomes (in most higher eukaryotes and some protists such as Dictyostelium) or the vacuole (in plants and fungi). In other words, it is especially important that investigators new to the field understand that the appearance of more autophagosomes does not necessarily equate with more autophagy. In fact, in many cases, autophagosomes accumulate because of a block in trafficking to lysosomes without a concomitant change in autophagosome biogenesis, whereas an increase in autolysosomes may reflect a reduction in degradative activity. Here, we present a set of guidelines for the selection and interpretation of methods for use by investigators who aim to examine macroautophagy and related processes, as well as for reviewers who need to provide realistic and reasonable critiques of papers that are focused on these processes. These guidelines are not meant to be a formulaic set of rules, because the appropriate assays depend in part on the question being asked and the system being used. In addition, we emphasize that no individual assay is guaranteed to be the most appropriate one in every situation, and we strongly recommend the use of multiple assays to monitor autophagy. In these guidelines, we consider these various methods of assessing autophagy and what information can, or cannot, be obtained from them. Finally, by discussing the merits and limits of particular autophagy assays, we hope to encourage technical innovation in the field.

Nuclear ribosomal internal transcribed spacer (ITS) region as a universal DNA barcode marker for Fungi

Abstract: Six DNA regions were evaluated as potential DNA barcodes for Fungi, the second largest kingdom of eukaryotic life, by a multinational, multilaboratory consortium. The region of the mitochondrial cytochrome c oxidase subunit 1 used as the animal barcode was excluded as a potential marker, because it is difficult to amplify in fungi, often includes large introns, and can be insufficiently variable. Three subunits from the nuclear ribosomal RNA cistron were compared together with regions of three representative protein-coding genes (largest subunit of RNA polymerase II, second largest subunit of RNA polymerase II, and minichromosome maintenance protein). Although the protein-coding gene regions often had a higher percent of correct identification compared with ribosomal markers, low PCR amplification and sequencing success eliminated them as candidates for a universal fungal barcode. Among the regions of the ribosomal cistron, the internal transcribed spacer (ITS) region has the highest probability of successful identification for the broadest range of fungi, with the most clearly defined barcode gap between inter- and intraspecific variation. The nuclear ribosomal large subunit, a popular phylogenetic marker in certain groups, had superior species resolution in some taxonomic groups, such as the early diverging lineages and the ascomycete yeasts, but was otherwise slightly inferior to the ITS. The nuclear ribosomal small subunit has poor species-level resolution in fungi. ITS will be formally proposed for adoption as the primary fungal barcode marker to the Consortium for the Barcode of Life, with the possibility that supplementary barcodes may be developed for particular narrowly circumscribed taxonomic groups.

Host-microbe interactions have shaped the genetic architecture of inflammatory bowel disease

Abstract: Crohn's disease and ulcerative colitis, the two common forms of inflammatory bowel disease (IBD), affect over 2.5 million people of European ancestry, with rising prevalence in other populations. Genome-wide association studies and subsequent meta-analyses of these two diseases as separate phenotypes have implicated previously unsuspected mechanisms, such as autophagy, in their pathogenesis and showed that some IBD loci are shared with other inflammatory diseases. Here we expand on the knowledge of relevant pathways by undertaking a meta-analysis of Crohn's disease and ulcerative colitis genome-wide association scans, followed by extensive validation of significant findings, with a combined total of more than 75,000 cases and controls. We identify 71 new associations, for a total of 163 IBD loci, that meet genome-wide significance thresholds. Most loci contribute to both phenotypes, and both directional (consistently favouring one allele over the course of human history) and balancing (favouring the retention of both alleles within populations) selection effects are evident. Many IBD loci are also implicated in other immune-mediated disorders, most notably with ankylosing spondylitis and psoriasis. We also observe considerable overlap between susceptibility loci for IBD and mycobacterial infection. Gene co-expression network analysis emphasizes this relationship, with pathways shared between host responses to mycobacteria and those predisposing to IBD.

Derivation and validation of the Systemic Lupus International Collaborating Clinics classification criteria for systemic lupus erythematosus.

Abstract: Objective The Systemic Lupus International Collaborating Clinics (SLICC) group revised and validated the American College of Rheumatology (ACR) systemic lupus erythematosus (SLE) classification criteria in order to improve clinical relevance, meet stringent methodology requirements, and incorporate new knowledge regarding the immunology of SLE. Methods The classification criteria were derived from a set of 702 expert-rated patient scenarios. Recursive partitioning was used to derive an initial rule that was simplified and refined based on SLICC physician consensus. The SLICC group validated the classification criteria in a new validation sample of 690 new expert-rated patient scenarios. Results Seventeen criteria were identified. In the derivation set, the SLICC classification criteria resulted in fewer misclassifications compared with the current ACR classification criteria (49 versus 70; P = 0.0082) and had greater sensitivity (94% versus 86%; P < 0.0001) and equal specificity (92% versus 93%; P = 0.39). In the validation set, the SLICC classification criteria resulted in fewer misclassifications compared with the current ACR classification criteria (62 versus 74; P = 0.24) and had greater sensitivity (97% versus 83%; P < 0.0001) but lower specificity (84% versus 96%; P < 0.0001). Conclusion The new SLICC classification criteria performed well in a large set of patient scenarios rated by experts. According to the SLICC rule for the classification of SLE, the patient must satisfy at least 4 criteria, including at least one clinical criterion and one immunologic criterion OR the patient must have biopsy-proven lupus nephritis in the presence of antinuclear antibodies or antidouble-stranded DNA antibodies. (Less)

Context-Dependent Pre-Trained Deep Neural Networks for Large-Vocabulary Speech Recognition

Abstract: We propose a novel context-dependent (CD) model for large-vocabulary speech recognition (LVSR) that leverages recent advances in using deep belief networks for phone recognition. We describe a pre-trained deep neural network hidden Markov model (DNN-HMM) hybrid architecture that trains the DNN to produce a distribution over senones (tied triphone states) as its output. The deep belief network pre-training algorithm is a robust and often helpful way to initialize deep neural networks generatively that can aid in optimization and reduce generalization error. We illustrate the key components of our model, describe the procedure for applying CD-DNN-HMMs to LVSR, and analyze the effects of various modeling choices on performance. Experiments on a challenging business search dataset demonstrate that CD-DNN-HMMs can significantly outperform the conventional context-dependent Gaussian mixture model (GMM)-HMMs, with an absolute sentence accuracy improvement of 5.8% and 9.2% (or relative error reduction of 16.0% and 23.2%) over the CD-GMM-HMMs trained using the minimum phone error rate (MPE) and maximum-likelihood (ML) criteria, respectively.

The effect of nanoparticle size, shape, and surface chemistry on biological systems.

Abstract: An understanding of the interactions between nanoparticles and biological systems is of significant interest. Studies aimed at correlating the properties of nanomaterials such as size, shape, chemical functionality, surface charge, and composition with biomolecular signaling, biological kinetics, transportation, and toxicity in both cell culture and animal experiments are under way. These fundamental studies will provide a foundation for engineering the next generation of nanoscale devices. Here, we provide rationales for these studies, review the current progress in studies of the interactions of nanomaterials with biological systems, and provide a perspective on the long-term implications of these findings.

A Practical Guide to Training Restricted Boltzmann Machines

Abstract: Restricted Boltzmann machines (RBMs) have been used as generative models of many different types of data. RBMs are usually trained using the contrastive divergence learning procedure. This requires a certain amount of practical experience to decide how to set the values of numerical meta-parameters. Over the last few years, the machine learning group at the University of Toronto has acquired considerable expertise at training RBMs and this guide is an attempt to share this expertise with other machine learning researchers.

Trastuzumab Emtansine for HER2-Positive Advanced Breast Cancer

Abstract: Background Trastuzumab emtansine (T-DM1) is an antibody–drug conjugate incorporating the human epidermal growth factor receptor 2 (HER2)–targeted antitumor properties of tras tuz u mab with the cytotoxic activity of the microtubule-inhibitory agent DM1. The antibody and the cytotoxic agent are conjugated by means of a stable linker. Methods We randomly assigned patients with HER2-positive advanced breast cancer, who had previously been treated with tras tuz u mab and a taxane, to T-DM1 or la pa ti nib plus cap e ci ta bine. The primary end points were progression-free survival (as assessed by independent review), overall survival, and safety. Secondary end points included progression-free survival (investigator-assessed), the objective response rate, and the time to symptom progression. Two interim analyses of overall survival were conducted. Results Among 991 randomly assigned patients, median progression-free survival as assessed by independent review was 9.6 months with T-DM1 versus 6.4 months with la pa ti nib plus cap e ci ta bine (hazard ratio for progression or death from any cause, 0.65; 95% confidence interval [CI], 0.55 to 0.77; P<0.001), and median overall survival at the second interim analysis crossed the stopping boundary for efficacy (30.9 months vs. 25.1 months; hazard ratio for death from any cause, 0.68; 95% CI, 0.55 to 0.85; P<0.001). The objective response rate was higher with T-DM1 (43.6%, vs. 30.8% with la pa ti nib plus cap e ci ta bine; P<0.001); results for all additional secondary end points favored T-DM1. Rates of adverse events of grade 3 or above were higher with la pati nib plus cap e ci ta bine than with T-DM1 (57% vs. 41%). The incidences of thrombocytopenia and increased serum aminotransferase levels were higher with T-DM1, whereas the incidences of diarrhea, nausea, vomiting, and palmar–plantar erythro dysesthesia were higher with la pa ti nib plus cap e ci ta bine. Conclusions T-DM1 significantly prolonged progression-free and overall survival with less toxicity than la pa ti nib plus cap e ci ta bine in patients with HER2-positive advanced breast cancer previously treated with tras tuz u mab and a taxane. (Funded by F. Hoffmann– La Roche/Genentech; EMILIA ClinicalTrials.gov number, NCT00829166.)

Everolimus in Postmenopausal Hormone-Receptor–Positive Advanced Breast Cancer

Abstract: A b s t r ac t Background Resistance to endocrine therapy in breast cancer is associated with activation of the mammalian target of rapamycin (mTOR) intracellular signaling pathway. In early studies, the mTOR inhibitor everolimus added to endocrine therapy showed antitumor activity. Methods In this phase 3, randomized trial, we compared everolimus and exemestane versus exemestane and placebo (randomly assigned in a 2:1 ratio) in 724 patients with hormone-receptor–positive advanced breast cancer who had recurrence or progression while receiving previous therapy with a nonsteroidal aromatase inhibitor in the adjuvant setting or to treat advanced disease (or both). The primary end point was progression-free survival. Secondary end points included survival, response rate, and safety. A preplanned interim analysis was performed by an independent data and safety monitoring committee after 359 progression-free survival events were observed. Results Baseline characteristics were well balanced between the two study groups. The median age was 62 years, 56% had visceral involvement, and 84% had hormone-sensitive disease. Previous therapy included letrozole or anastrozole (100%), tamoxifen (48%), fulvestrant (16%), and chemotherapy (68%). The most common grade 3 or 4 adverse events were stomatitis (8% in the everolimus-plus-exemestane group vs. 1% in the placebo-plus-exemestane group), anemia (6% vs. <1%), dyspnea (4% vs. 1%), hyperglycemia (4% vs. <1%), fatigue (4% vs. 1%), and pneumonitis (3% vs. 0%). At the interim analysis, median progression-free survival was 6.9 months with everolimus plus exemestane and 2.8 months with placebo plus exemestane, according to assessments by local investigators (hazard ratio for progression or death, 0.43; 95% confi dence interval [CI], 0.35 to 0.54; P<0.001). Median progression-free survival was 10.6 months and 4.1 months, respectively, according to central assessment (hazard ratio, 0.36; 95% CI, 0.27 to 0.47; P<0.001). Conclusions Everolimus combined with an aromatase inhibitor improved progression-free survival in patients with hormone-receptor–positive advanced breast cancer previously treated with nonsteroidal aromatase inhibitors. (Funded by Novartis; BOLERO-2 ClinicalTrials .gov number, NCT00863655.)

Driver mutations in histone H3.3 and chromatin remodelling genes in paediatric glioblastoma

Abstract: Glioblastoma multiforme (GBM) is a lethal brain tumour in adults and children. However, DNA copy number and gene expression signatures indicate differences between adult and paediatric cases. To explore the genetic events underlying this distinction, we sequenced the exomes of 48 paediatric GBM samples. Somatic mutations in the H3.3-ATRX-DAXX chromatin remodelling pathway were identified in 44% of tumours (21/48). Recurrent mutations in H3F3A, which encodes the replication-independent histone 3 variant H3.3, were observed in 31% of tumours, and led to amino acid substitutions at two critical positions within the histone tail (K27M, G34R/G34V) involved in key regulatory post-translational modifications. Mutations in ATRX (α-thalassaemia/mental retardation syndrome X-linked) and DAXX (death-domain associated protein), encoding two subunits of a chromatin remodelling complex required for H3.3 incorporation at pericentric heterochromatin and telomeres, were identified in 31% of samples overall, and in 100% of tumours harbouring a G34R or G34V H3.3 mutation. Somatic TP53 mutations were identified in 54% of all cases, and in 86% of samples with H3F3A and/or ATRX mutations. Screening of a large cohort of gliomas of various grades and histologies (n = 784) showed H3F3A mutations to be specific to GBM and highly prevalent in children and young adults. Furthermore, the presence of H3F3A/ATRX-DAXX/TP53 mutations was strongly associated with alternative lengthening of telomeres and specific gene expression profiles. This is, to our knowledge, the first report to highlight recurrent mutations in a regulatory histone in humans, and our data suggest that defects of the chromatin architecture underlie paediatric and young adult GBM pathogenesis.

Fairness through awareness

Abstract: We study fairness in classification, where individuals are classified, e.g., admitted to a university, and the goal is to prevent discrimination against individuals based on their membership in some group, while maintaining utility for the classifier (the university). The main conceptual contribution of this paper is a framework for fair classification comprising (1) a (hypothetical) task-specific metric for determining the degree to which individuals are similar with respect to the classification task at hand; (2) an algorithm for maximizing utility subject to the fairness constraint, that similar individuals are treated similarly. We also present an adaptation of our approach to achieve the complementary goal of "fair affirmative action," which guarantees statistical parity (i.e., the demographics of the set of individuals receiving any classification are the same as the demographics of the underlying population), while treating similar individuals as similarly as possible. Finally, we discuss the relationship of fairness to privacy: when fairness implies privacy, and how tools developed in the context of differential privacy may be applied to fairness.

Leveled) fully homomorphic encryption without bootstrapping

Abstract: We present a novel approach to fully homomorphic encryption (FHE) that dramatically improves performance and bases security on weaker assumptions. A central conceptual contribution in our work is a new way of constructing leveled fully homomorphic encryption schemes (capable of evaluating arbitrary polynomial-size circuits), without Gentry's bootstrapping procedure.Specifically, we offer a choice of FHE schemes based on the learning with error (LWE) or ring-LWE (RLWE) problems that have 2λ security against known attacks. For RLWE, we have:• A leveled FHE scheme that can evaluate L-level arithmetic circuits with O(λ · L3) per-gate computation -- i.e., computation quasi-linear in the security parameter. Security is based on RLWE for an approximation factor exponential in L. This construction does not use the bootstrapping procedure.• A leveled FHE scheme that uses bootstrapping as an optimization, where the per-gate computation (which includes the bootstrapping procedure) is O(λ2), independent of L. Security is based on the hardness of RLWE for quasi-polynomial factors (as opposed to the sub-exponential factors needed in previous schemes).We obtain similar results to the above for LWE, but with worse performance.Based on the Ring LWE assumption, we introduce a number of further optimizations to our schemes. As an example, for circuits of large width -- e.g., where a constant fraction of levels have width at least λ -- we can reduce the per-gate computation of the bootstrapped version to O(λ), independent of L, by batching the bootstrapping operation. Previous FHE schemes all required Ω(λ3.5) computation per gate.At the core of our construction is a much more effective approach for managing the noise level of lattice-based ciphertexts as homomorphic operations are performed, using some new techniques recently introduced by Brakerski and Vaikuntanathan (FOCS 2011).

The Hubble Space Telescope Cluster Supernova Survey. V. Improving the Dark-energy Constraints above z > 1 and Building an Early-type-hosted Supernova Sample

Abstract: We present Advanced Camera for Surveys, NICMOS, and Keck adaptive-optics-assisted photometry of 20 Type Ia supernovae (SNe Ia) from the Hubble Space Telescope (HST) Cluster Supernova Survey. The SNe Ia were discovered over the redshift interval 0.623 1 SNe Ia. We describe how such a sample could be efficiently obtained by targeting cluster fields with WFC3 on board HST. The updated supernova Union2.1 compilation of 580 SNe is available at http://supernova.lbl.gov/Union.

Acoustic Modeling Using Deep Belief Networks

Abstract: Gaussian mixture models are currently the dominant technique for modeling the emission distribution of hidden Markov models for speech recognition. We show that better phone recognition on the TIMIT dataset can be achieved by replacing Gaussian mixture models by deep neural networks that contain many layers of features and a very large number of parameters. These networks are first pre-trained as a multi-layer generative model of a window of spectral feature vectors without making use of any discriminative information. Once the generative pre-training has designed the features, we perform discriminative fine-tuning using backpropagation to adjust the features slightly to make them better at predicting a probability distribution over the states of monophone hidden Markov models.

Pancreatic cancer genomes reveal aberrations in axon guidance pathway genes

Abstract: Pancreatic cancer is a highly lethal malignancy with few effective therapies. We performed exome sequencing and copy number analysis to define genomic aberrations in a prospectively accrued clinical cohort (n = 142) of early (stage I and II) sporadic pancreatic ductal adenocarcinoma. Detailed analysis of 99 informative tumours identified substantial heterogeneity with 2,016 non-silent mutations and 1,628 copy-number variations. We define 16 significantly mutated genes, reaffirming known mutations (KRAS, TP53, CDKN2A, SMAD4, MLL3, TGFBR2, ARID1A and SF3B1), and uncover novel mutated genes including additional genes involved in chromatin modification (EPC1 and ARID2), DNA damage repair (ATM) and other mechanisms (ZIM2, MAP2K4, NALCN, SLC16A4 and MAGEA6). Integrative analysis with in vitro functional data and animal models provided supportive evidence for potential roles for these genetic aberrations in carcinogenesis. Pathway-based analysis of recurrently mutated genes recapitulated clustering in core signalling pathways in pancreatic ductal adenocarcinoma, and identified new mutated genes in each pathway. We also identified frequent and diverse somatic aberrations in genes described traditionally as embryonic regulators of axon guidance, particularly SLIT/ROBO signalling, which was also evident in murine Sleeping Beauty transposon-mediated somatic mutagenesis models of pancreatic cancer, providing further supportive evidence for the potential involvement of axon guidance genes in pancreatic carcinogenesis.

Measurement-Device-Independent Quantum Key Distribution

Abstract: How to remove detector side channel attacks has been a notoriously hard problem in quantum cryptography. Here, we propose a simple solution to this problem--measurement-device-independent quantum key distribution (QKD). It not only removes all detector side channels, but also doubles the secure distance with conventional lasers. Our proposal can be implemented with standard optical components with low detection efficiency and highly lossy channels. In contrast to the previous solution of full device independent QKD, the realization of our idea does not require detectors of near unity detection efficiency in combination with a qubit amplifier (based on teleportation) or a quantum nondemolition measurement of the number of photons in a pulse. Furthermore, its key generation rate is many orders of magnitude higher than that based on full device independent QKD. The results show that long-distance quantum cryptography over say 200 km will remain secure even with seriously flawed detectors.

A Double-Blind, Placebo-Controlled Trial of Ruxolitinib for Myelofibrosis

Abstract: A B S T R AC T background Ruxolitinib, a selective inhibitor of Janus kinase (JAK) 1 and 2, has clinically significant activity in myelofibrosis. methodS In this double-blind trial, we randomly assigned patients with intermediate-2 or highrisk myelofibrosis to twice-daily oral ruxolitinib (155 patients) or placebo (154 patients). The primary end point was the proportion of patients with a reduction in spleen volume of 35% or more at 24 weeks, assessed by means of magnetic resonance imaging. Secondary end points included the durability of response, changes in symptom burden (assessed by the total symptom score), and overall survival. resulTS The primary end point was reached in 41.9% of patients in the ruxolitinib group as compared with 0.7% in the placebo group (P<0.001). A reduction in spleen volume was maintained in patients who received ruxolitinib; 67.0% of the patients with a response had the response for 48 weeks or more. There was an improvement of 50% or more in the total symptom score at 24 weeks in 45.9% of patients who received ruxolitinib as compared with 5.3% of patients who received placebo (P<0.001). Thirteen deaths occurred in the ruxolitinib group as compared with 24 deaths in the placebo group (hazard ratio, 0.50; 95% confidence interval, 0.25 to 0.98; P = 0.04). The rate of discontinuation of the study drug because of adverse events was 11.0% in the ruxolitinib group and 10.6% in the placebo group. Among patients who received ruxolitinib, anemia and thrombocytopenia were the most common adverse events, but they rarely led to discontinuation of the drug (in one patient for each event). Two patients had transformation to acute myeloid leukemia; both were in the ruxolitinib group. conclusionS Ruxolitinib, as compared with placebo, provided significant clinical benefits in patients with myelofibrosis by reducing spleen size, ameliorating debilitating myelofibrosis-related symptoms, and improving overall survival. These benefits came at the cost of more frequent anemia and thrombocytopenia in the early part of the treatment period. (Funded by Incyte; COMFORT-I ClinicalTrials.gov number, NCT00952289.)

Nanoparticle size and surface chemistry determine serum protein adsorption and macrophage uptake.

Abstract: Delivery and toxicity are critical issues facing nanomedicine research. Currently, there is limited understanding and connection between the physicochemical properties of a nanomaterial and its interactions with a physiological system. As a result, it remains unclear how to optimally synthesize and chemically modify nanomaterials for in vivo applications. It has been suggested that the physicochemical properties of a nanomaterial after synthesis, known as its “synthetic identity”, are not what a cell encounters in vivo. Adsorption of blood components and interactions with phagocytes can modify the size, aggregation state, and interfacial composition of a nanomaterial, giving it a distinct “biological identity”. Here, we investigate the role of size and surface chemistry in mediating serum protein adsorption to gold nanoparticles and their subsequent uptake by macrophages. Using label-free liquid chromatography tandem mass spectrometry, we find that over 70 different serum proteins are heterogeneously adso...