Abstract: Despite an abundance of new studies about topologically associating domains (TADs), the role of genetic information in TAD formation is still not fully understood. Here we use our software, HiCExplorer (hicexplorer.readthedocs.io) to annotate >2800 high-resolution (570 bp) TAD boundaries in Drosophila melanogaster. We identify eight DNA motifs enriched at boundaries, including a motif bound by the M1BP protein, and two new boundary motifs. In contrast to mammals, the CTCF motif is only enriched on a small fraction of boundaries flanking inactive chromatin while most active boundaries contain the motifs bound by the M1BP or Beaf-32 proteins. We demonstrate that boundaries can be accurately predicted using only the motif sequences at open chromatin sites. We propose that DNA sequence guides the genome architecture by allocation of boundary proteins in the genome. Finally, we present an interactive online database to access and explore the spatial organization of fly, mouse and human genomes, available at http://chorogenome.ie-freiburg.mpg.de . Although topologically associating domains (TADs) have been extensively investigated, it is not clear to what extent DNA sequence contributes to their formation. Here the authors develop software to identify high-resolution TAD boundaries and reveal their relationship to underlying DNA motifs.

Abstract: Cancer cells exploit the expression of the programmed death-1 (PD-1) ligand 1 (PD-L1) to subvert T-cell-mediated immunosurveillance. The success of therapies that disrupt PD-L1-mediated tumour tolerance has highlighted the need to understand the molecular regulation of PD-L1 expression. Here we identify the uncharacterized protein CMTM6 as a critical regulator of PD-L1 in a broad range of cancer cells, by using a genome-wide CRISPR-Cas9 screen. CMTM6 is a ubiquitously expressed protein that binds PD-L1 and maintains its cell surface expression. CMTM6 is not required for PD-L1 maturation but co-localizes with PD-L1 at the plasma membrane and in recycling endosomes, where it prevents PD-L1 from being targeted for lysosome-mediated degradation. Using a quantitative approach to profile the entire plasma membrane proteome, we find that CMTM6 displays specificity for PD-L1. Notably, CMTM6 depletion decreases PD-L1 without compromising cell surface expression of MHC class I. CMTM6 depletion, via the reduction of PD-L1, significantly alleviates the suppression of tumour-specific T cell activity in vitro and in vivo. These findings provide insights into the biology of PD-L1 regulation, identify a previously unrecognized master regulator of this critical immune checkpoint and highlight a potential therapeutic target to overcome immune evasion by tumour cells.

Abstract: The ambitious goals set for 5G wireless networks, which are expected to be introduced around 2020, require dramatic changes in the design of different layers for next generation communications systems. Massive MIMO systems, filter bank multi-carrier modulation, relaying technologies, and millimeter-wave communications have been considered as some of the strong candidates for the physical layer design of 5G networks. In this article, we shed light on the potential and implementation of IM techniques for MIMO and multi-carrier communications systems, which are expected to be two of the key technologies for 5G systems. Specifically, we focus on two promising applications of IM: spatial modulation and orthogonal frequency-division multiplexing with IM, and discuss the recent advances and future research directions in IM technologies toward spectrum- and energy-efficient 5G wireless networks.

Abstract: The epidermal growth factor receptor (EGFR)-directed tyrosine kinase inhibitors (TKIs) gefitinib, erlotinib and afatinib are approved treatments for non-small cell lung cancers harbouring activating mutations in the EGFR kinase, but resistance arises rapidly, most frequently owing to the secondary T790M mutation within the ATP site of the receptor. Recently developed mutant-selective irreversible inhibitors are highly active against the T790M mutant, but their efficacy can be compromised by acquired mutation of C797, the cysteine residue with which they form a key covalent bond. All current EGFR TKIs target the ATP-site of the kinase, highlighting the need for therapeutic agents with alternative mechanisms of action. Here we describe the rational discovery of EAI045, an allosteric inhibitor that targets selected drug-resistant EGFR mutants but spares the wild-type receptor. The crystal structure shows that the compound binds an allosteric site created by the displacement of the regulatory C-helix in an inactive conformation of the kinase. The compound inhibits L858R/T790M-mutant EGFR with low-nanomolar potency in biochemical assays. However, as a single agent it is not effective in blocking EGFR-driven proliferation in cells owing to differential potency on the two subunits of the dimeric receptor, which interact in an asymmetric manner in the active state. We observe marked synergy of EAI045 with cetuximab, an antibody therapeutic that blocks EGFR dimerization, rendering the kinase uniformly susceptible to the allosteric agent. EAI045 in combination with cetuximab is effective in mouse models of lung cancer driven by EGFR(L858R/T790M) and by EGFR(L858R/T790M/C797S), a mutant that is resistant to all currently available EGFR TKIs. More generally, our findings illustrate the utility of purposefully targeting allosteric sites to obtain mutant-selective inhibitors.

Abstract: Microbiota represents the entire microbial community present in the gut host. It serves several functions establishing a mutualistic relation with the host. Latest years have seen a burst in the number of studies focusing on this topic, in particular on intestinal diseases. In this scenario, Proteobacteria are one of the most abundant phyla, comprising several known human pathogens. This review highlights the latest findings on the role of Proteobacteria not only in intestinal but also in extraintestinal diseases. Indeed, an increasing amount of data identifies Proteobacteria as a possible microbial signature of disease. Several studies demonstrate an increased abundance of members belonging to this phylum in such conditions. Major evidences currently involve metabolic disorders and inflammatory bowel disease. However, more recent studies suggest a role also in lung diseases, such as asthma and chronic obstructive pulmonary disease, but evidences are still scant. Notably, all these conditions are sustained by various degree of inflammation, which thus represents a core aspect of Proteobacteria-related diseases.

Abstract: We recently developed base editing, the programmable conversion of target C:G base pairs to T:A without inducing double-stranded DNA breaks (DSBs) or requiring homology-directed repair using engineered fusions of Cas9 variants and cytidine deaminases. Over the past year, the third-generation base editor (BE3) and related technologies have been successfully used by many researchers in a wide range of organisms. The product distribution of base editing-the frequency with which the target C:G is converted to mixtures of undesired by-products, along with the desired T:A product-varies in a target site-dependent manner. We characterize determinants of base editing outcomes in human cells and establish that the formation of undesired products is dependent on uracil N-glycosylase (UNG) and is more likely to occur at target sites containing only a single C within the base editing activity window. We engineered CDA1-BE3 and AID-BE3, which use cytidine deaminase homologs that increase base editing efficiency for some sequences. On the basis of these observations, we engineered fourth-generation base editors (BE4 and SaBE4) that increase the efficiency of C:G to T:A base editing by approximately 50%, while halving the frequency of undesired by-products compared to BE3. Fusing BE3, BE4, SaBE3, or SaBE4 to Gam, a bacteriophage Mu protein that binds DSBs greatly reduces indel formation during base editing, in most cases to below 1.5%, and further improves product purity. BE4, SaBE4, BE4-Gam, and SaBE4-Gam represent the state of the art in C:G-to-T:A base editing, and we recommend their use in future efforts.

Abstract: Automatic segmentation in MR brain images is important for quantitative analysis in large-scale studies with images acquired at all ages. This paper presents a method for the automatic segmentation of MR brain images into a number of tissue classes using a convolutional neural network. To ensure that the method obtains accurate segmentation details as well as spatial consistency, the network uses multiple patch sizes and multiple convolution kernel sizes to acquire multi-scale information about each voxel. The method is not dependent on explicit features, but learns to recognise the information that is important for the classification based on training data. The method requires a single anatomical MR image only. The segmentation method is applied to five different data sets: coronal $ {\rm T}_{2}$ -weighted images of preterm infants acquired at 30 weeks postmenstrual age (PMA) and 40 weeks PMA, axial $ {\rm T}_{2}$ -weighted images of preterm infants acquired at 40 weeks PMA, axial $ {\rm T}_{1}$ -weighted images of ageing adults acquired at an average age of 70 years, and $ {\rm T}_{1}$ -weighted images of young adults acquired at an average age of 23 years. The method obtained the following average Dice coefficients over all segmented tissue classes for each data set, respectively: 0.87, 0.82, 0.84, 0.86, and 0.91. The results demonstrate that the method obtains accurate segmentations in all five sets, and hence demonstrates its robustness to differences in age and acquisition protocol.

Abstract: We present a comprehensive study of evaluation methods for unsupervised embedding techniques that obtain meaningful representations of words from text. Different evaluations result in different orderings of embedding methods, calling into question the common assumption that there is one single optimal vector representation. We present new evaluation techniques that directly compare embeddings with respect to specific queries. These methods reduce bias, provide greater insight, and allow us to solicit data-driven relevance judgments rapidly and accurately through crowdsourcing.

Abstract: Key questions in COVID-19 are the duration and determinants of infectious virus shedding. Here, we report that infectious virus shedding is detected by virus cultures in 23 of the 129 patients (17.8%) hospitalized with COVID-19. The median duration of shedding infectious virus is 8 days post onset of symptoms (IQR 5–11) and drops below 5% after 15.2 days post onset of symptoms (95% confidence interval (CI) 13.4–17.2). Multivariate analyses identify viral loads above 7 log10 RNA copies/mL (odds ratio [OR] of 14.7 (CI 3.57-58.1; p < 0.001) as independently associated with isolation of infectious SARS-CoV-2 from the respiratory tract. A serum neutralizing antibody titre of at least 1:20 (OR of 0.01 (CI 0.003-0.08; p < 0.001) is independently associated with non-infectious SARS-CoV-2. We conclude that quantitative viral RNA load assays and serological assays could be used in test-based strategies to discontinue or de-escalate infection prevention and control precautions.

Abstract: β-Lactams are the most widely used class of antibiotics. Since the discovery of benzylpenicillin in the 1920s, thousands of new penicillin derivatives and related β-lactam classes of cephalosporins, cephamycins, monobactams, and carbapenems have been discovered. Each new class of β-lactam has been developed either to increase the spectrum of activity to include additional bacterial species or to address specific resistance mechanisms that have arisen in the targeted bacterial population. Resistance to β-lactams is primarily because of bacterially produced β-lactamase enzymes that hydrolyze the β-lactam ring, thereby inactivating the drug. The newest effort to circumvent resistance is the development of novel broad-spectrum β-lactamase inhibitors that work against many problematic β-lactamases, including cephalosporinases and serine-based carbapenemases, which severely limit therapeutic options. This work provides a comprehensive overview of β-lactam antibiotics that are currently in use, as well as a look ahead to several new compounds that are in the development pipeline.

Abstract: The third catalog of active galactic nuclei (AGNs) detected by the Fermi-LAT (3LAC) is presented. It is based on the third Fermi-LAT catalog (3FGL) of sources detected between 100 MeV and 300 GeV with a Test Statistic (TS) greater than 25, between 2008 August 4 and 2012 July 31. The 3LAC includes 1591 AGNs located at high Galactic latitudes (|b|>10°), a 71% increase over the second catalog based on 2 years of data. There are 28 duplicate associations, thus 1563 of the 2192 high-latitude gamma-ray sources of the 3FGL catalog are AGNs. Most of them (98%) are blazars. About half of the newly detected blazars are of unknown type, i.e., they lack spectroscopic information of sufficient quality to determine the strength of their emission lines. Based on their gamma-ray spectral properties, these sources are evenly split between flat-spectrum radio quasars (FSRQs) and BL~Lacs. The most abundant detected BL~Lacs are of the high-synchrotron-peaked (HSP) type. About 50% of the BL~Lacs have no measured redshifts. A few new rare outliers (HSP-FSRQs and high-luminosity HSP BL~Lacs) are reported. The general properties of the 3LAC sample confirm previous findings from earlier catalogs. The fraction of 3LAC blazars in the total population of blazars listed in BZCAT remains non-negligible even at the faint ends of the BZCAT-blazar radio, optical and X-ray flux distributions, which is a clue that even the faintest known blazars could eventually shine in gamma rays at LAT-detection levels. The energy-flux distributions of the different blazar populations are in good agreement with extrapolation from earlier catalogs.

Abstract: The U.S. Geological Survey (USGS), in partnership with several federal agencies, has developed and released four National Land Cover Database (NLCD) products over the past two decades: NLCD 1992, 2001, 2006, and 2011. These products provide spatially explicit and reliable information on the Nation’s land cover and land cover change. To continue the legacy of NLCD and further establish a long-term monitoring capability for the Nation’s land resources, the USGS has designed a new generation of NLCD products named NLCD 2016. The NLCD 2016 design aims to provide innovative, consistent, and robust methodologies for production of a multi-temporal land cover and land cover change database from 2001 to 2016 at 2–3-year intervals. Comprehensive research was conducted and resulted in developed strategies for NLCD 2016: a streamlined process for assembling and preprocessing Landsat imagery and geospatial ancillary datasets; a multi-source integrated training data development and decision-tree based land cover classifications; a temporally, spectrally, and spatially integrated land cover change analysis strategy; a hierarchical theme-based post-classification and integration protocol for generating land cover and change products; a continuous fields biophysical parameters modeling method; and an automated scripted operational system for the NLCD 2016 production. The performance of the developed strategies and methods were tested in twenty World Reference System-2 path/row throughout the conterminous U.S. An overall agreement ranging from 71% to 97% between land cover classification and reference data was achieved for all tested area and all years. Results from this study confirm the robustness of this comprehensive and highly automated procedure for NLCD 2016 operational mapping.

Abstract: In this paper we provide a comprehensive introduction to knowledge graphs, which have recently garnered significant attention from both industry and academia in scenarios that require exploiting diverse, dynamic, large-scale collections of data. After some opening remarks, we motivate and contrast various graph-based data models and query languages that are used for knowledge graphs. We discuss the roles of schema, identity, and context in knowledge graphs. We explain how knowledge can be represented and extracted using a combination of deductive and inductive techniques. We summarise methods for the creation, enrichment, quality assessment, refinement, and publication of knowledge graphs. We provide an overview of prominent open knowledge graphs and enterprise knowledge graphs, their applications, and how they use the aforementioned techniques. We conclude with high-level future research directions for knowledge graphs.

Abstract: Low oxygen tension (hypoxia) is a pervasive physiological and pathophysiological stimulus that metazoan organisms have contended with since they evolved from their single-celled ancestors. The effect of hypoxia on a tissue can be either positive or negative, depending on the severity, duration and context. Over the long-term, hypoxia is not usually consistent with normal function and so multicellular organisms have had to evolve both systemic and cellular responses to hypoxia. Our reliance on oxygen for efficient adenosine triphosphate (ATP) generation has meant that the cellular metabolic network is particularly sensitive to alterations in oxygen tension. Metabolic changes in response to hypoxia are elicited through both direct mechanisms, such as the reduction in ATP generation by oxidative phosphorylation or inhibition of fatty-acid desaturation, and indirect mechanisms including changes in isozyme expression through hypoxia-responsive transcription factor activity. Significant regions of cancers often grow in hypoxic conditions owing to the lack of a functional vasculature. As hypoxic tumour areas contain some of the most malignant cells, it is important that we understand the role metabolism has in keeping these cells alive. This review will outline our current understanding of many of the hypoxia-induced changes in cancer cell metabolism, how they are affected by other genetic defects often present in cancers, and how these metabolic alterations support the malignant hypoxic phenotype.

Abstract: Background Recent guidelines recommend consideration of the use of oral edoxaban or rivaroxaban for the treatment of venous thromboembolism in patients with cancer. However, the benefit of...

Abstract: We report the first electronic structure calculation performed on a quantum computer without exponentially costly precompilation. We use a programmable array of superconducting qubits to compute the energy surface of molecular hydrogen using two distinct quantum algorithms. First, we experimentally execute the unitary coupled cluster method using the variational quantum eigensolver. Our efficient implementation predicts the correct dissociation energy to within chemical accuracy of the numerically exact result. Second, we experimentally demonstrate the canonical quantum algorithm for chemistry, which consists of Trotterization and quantum phase estimation. We compare the experimental performance of these approaches to show clear evidence that the variational quantum eigensolver is robust to certain errors. This error tolerance inspires hope that variational quantum simulations of classically intractable molecules may be viable in the near future.

Abstract: Federated learning (FL), as a type of distributed machine learning, is capable of significantly preserving clients’ private data from being exposed to adversaries. Nevertheless, private information can still be divulged by analyzing uploaded parameters from clients, e.g., weights trained in deep neural networks. In this paper, to effectively prevent information leakage, we propose a novel framework based on the concept of differential privacy (DP), in which artificial noise is added to parameters at the clients’ side before aggregating, namely, noising before model aggregation FL (NbAFL). First, we prove that the NbAFL can satisfy DP under distinct protection levels by properly adapting different variances of artificial noise. Then we develop a theoretical convergence bound on the loss function of the trained FL model in the NbAFL. Specifically, the theoretical bound reveals the following three key properties: 1) there is a tradeoff between convergence performance and privacy protection levels, i.e., better convergence performance leads to a lower protection level; 2) given a fixed privacy protection level, increasing the number $N$ of overall clients participating in FL can improve the convergence performance; and 3) there is an optimal number aggregation times (communication rounds) in terms of convergence performance for a given protection level. Furthermore, we propose a $K$ -client random scheduling strategy, where $K$ ( $1\leq K ) clients are randomly selected from the $N$ overall clients to participate in each aggregation. We also develop a corresponding convergence bound for the loss function in this case and the $K$ -client random scheduling strategy also retains the above three properties. Moreover, we find that there is an optimal $K$ that achieves the best convergence performance at a fixed privacy level. Evaluations demonstrate that our theoretical results are consistent with simulations, thereby facilitating the design of various privacy-preserving FL algorithms with different tradeoff requirements on convergence performance and privacy levels.

Abstract: Recent genomic studies challenge the conventional model that each metastasis must arise from a single tumor cell and instead reveal that metastases can be composed of multiple genetically distinct clones. These intriguing observations raise the question: How do polyclonal metastases emerge from the primary tumor? In this study, we used multicolor lineage tracing to demonstrate that polyclonal seeding by cell clusters is a frequent mechanism in a common mouse model of breast cancer, accounting for >90% of metastases. We directly observed multicolored tumor cell clusters across major stages of metastasis, including collective invasion, local dissemination, intravascular emboli, circulating tumor cell clusters, and micrometastases. Experimentally aggregating tumor cells into clusters induced a >15-fold increase in colony formation ex vivo and a >100-fold increase in metastasis formation in vivo. Intriguingly, locally disseminated clusters, circulating tumor cell clusters, and lung micrometastases frequently expressed the epithelial cytoskeletal protein, keratin 14 (K14). RNA-seq analysis revealed that K14(+) cells were enriched for desmosome and hemidesmosome adhesion complex genes, and were depleted for MHC class II genes. Depletion of K14 expression abrogated distant metastases and disrupted expression of multiple metastasis effectors, including Tenascin C (Tnc), Jagged1 (Jag1), and Epiregulin (Ereg). Taken together, our findings reveal K14 as a key regulator of metastasis and establish the concept that K14(+) epithelial tumor cell clusters disseminate collectively to colonize distant organs.

Abstract: With the ever-growing volume of online information, recommender systems have been an effective strategy to overcome such information overload. The utility of recommender systems cannot be overstated, given its widespread adoption in many web applications, along with its potential impact to ameliorate many problems related to over-choice. In recent years, deep learning has garnered considerable interest in many research fields such as computer vision and natural language processing, owing not only to stellar performance but also the attractive property of learning feature representations from scratch. The influence of deep learning is also pervasive, recently demonstrating its effectiveness when applied to information retrieval and recommender systems research. Evidently, the field of deep learning in recommender system is flourishing. This article aims to provide a comprehensive review of recent research efforts on deep learning based recommender systems. More concretely, we provide and devise a taxonomy of deep learning based recommendation models, along with providing a comprehensive summary of the state-of-the-art. Finally, we expand on current trends and provide new perspectives pertaining to this new exciting development of the field.

Abstract: We consider the status of black hole (BH) solutions with nontrivial scalar fields but no gauge fields, in four-dimensional asymptotically flat spacetimes, reviewing both classical results and recent developments. We start by providing a simple illustration on the physical difference between BHs in electro-vacuum and scalar-vacuum. Next, we review no-scalar-hair theorems. In particular, we detail an influential theorem by Bekenstein and stress three key assumptions: (1) The type of scalar field equation; (2) the spacetime symmetry inheritance by the scalar field and (3) an energy condition. Then, we list regular (on and outside the horizon), asymptotically flat BH solutions with scalar hair, organizing them by the assumption which is violated in each case and distinguishing primary from secondary hair. We provide a table summary of the state-of-the-art.

Abstract: New challenges in remote sensing impose the necessity of designing pixel classification methods that, once trained on a certain dataset, generalize to other areas of the earth. This may include regions where the appearance of the same type of objects is significantly different. In the literature it is common to use a single image and split it into training and test sets to train a classifier and assess its performance, respectively. However, this does not prove the generalization capabilities to other inputs. In this paper, we propose an aerial image labeling dataset that covers a wide range of urban settlement appearances, from different geographic locations. Moreover, the cities included in the test set are different from those of the training set. We also experiment with convolutional neural networks on our dataset.

Abstract: Anomaly detection, a.k.a. outlier detection or novelty detection, has been a lasting yet active research area in various research communities for several decades. There are still some unique problem complexities and challenges that require advanced approaches. In recent years, deep learning enabled anomaly detection, i.e., deep anomaly detection, has emerged as a critical direction. This article surveys the research of deep anomaly detection with a comprehensive taxonomy, covering advancements in 3 high-level categories and 11 fine-grained categories of the methods. We review their key intuitions, objective functions, underlying assumptions, advantages, and disadvantages and discuss how they address the aforementioned challenges. We further discuss a set of possible future opportunities and new perspectives on addressing the challenges.

Abstract: In recent years, vector-based machine learning algorithms, such as random forests, support vector machines, and 1-D convolutional neural networks, have shown promising results in hyperspectral image classification. Such methodologies, nevertheless, can lead to information loss in representing hyperspectral pixels, which intrinsically have a sequence-based data structure. A recurrent neural network (RNN), an important branch of the deep learning family, is mainly designed to handle sequential data. Can sequence-based RNN be an effective method of hyperspectral image classification? In this paper, we propose a novel RNN model that can effectively analyze hyperspectral pixels as sequential data and then determine information categories via network reasoning. As far as we know, this is the first time that an RNN framework has been proposed for hyperspectral image classification. Specifically, our RNN makes use of a newly proposed activation function, parametric rectified tanh (PRetanh), for hyperspectral sequential data analysis instead of the popular tanh or rectified linear unit. The proposed activation function makes it possible to use fairly high learning rates without the risk of divergence during the training procedure. Moreover, a modified gated recurrent unit, which uses PRetanh for hidden representation, is adopted to construct the recurrent layer in our network to efficiently process hyperspectral data and reduce the total number of parameters. Experimental results on three airborne hyperspectral images suggest competitive performance in the proposed mode. In addition, the proposed network architecture opens a new window for future research, showcasing the huge potential of deep recurrent networks for hyperspectral data analysis.

Abstract: Despite the recent advance of Generative Adversarial Networks (GANs) in high-fidelity image synthesis, there lacks enough understanding of how GANs are able to map a latent code sampled from a random distribution to a photo-realistic image. Previous work assumes the latent space learned by GANs follows a distributed representation but observes the vector arithmetic phenomenon. In this work, we propose a novel framework, called InterFaceGAN, for semantic face editing by interpreting the latent semantics learned by GANs. In this framework, we conduct a detailed study on how different semantics are encoded in the latent space of GANs for face synthesis. We find that the latent code of well-trained generative models actually learns a disentangled representation after linear transformations. We explore the disentanglement between various semantics and manage to decouple some entangled semantics with subspace projection, leading to more precise control of facial attributes. Besides manipulating gender, age, expression, and the presence of eyeglasses, we can even vary the face pose as well as fix the artifacts accidentally generated by GAN models. The proposed method is further applied to achieve real image manipulation when combined with GAN inversion methods or some encoder-involved models. Extensive results suggest that learning to synthesize faces spontaneously brings a disentangled and controllable facial attribute representation.

Abstract: The ubiquitin–proteasome system (UPS) and autophagy are two distinct and interacting proteolytic systems. They play critical roles in cell survival under normal conditions and during stress. An increasing body of evidence indicates that ubiquitinated cargoes are important markers of degradation. p62, a classical receptor of autophagy, is a multifunctional protein located throughout the cell and involved in many signal transduction pathways, including the Keap1–Nrf2 pathway. It is involved in the proteasomal degradation of ubiquitinated proteins. When the cellular p62 level is manipulated, the quantity and location pattern of ubiquitinated proteins change with a considerable impact on cell survival. Altered p62 levels can even lead to some diseases. The proteotoxic stress imposed by proteasome inhibition can activate autophagy through p62 phosphorylation. A deficiency in autophagy may compromise the ubiquitin–proteasome system, since overabundant p62 delays delivery of the proteasomal substrate to the proteasome despite proteasomal catalytic activity being unchanged. In addition, p62 and the proteasome can modulate the activity of HDAC6 deacetylase, thus influencing the autophagic degradation.

Abstract: Fully convolutional deep neural networks carry out excellent potential for fast and accurate image segmentation. One of the main challenges in training these networks is data imbalance, which is particularly problematic in medical imaging applications such as lesion segmentation where the number of lesion voxels is often much lower than the number of non-lesion voxels. Training with unbalanced data can lead to predictions that are severely biased towards high precision but low recall (sensitivity), which is undesired especially in medical applications where false negatives are much less tolerable than false positives. Several methods have been proposed to deal with this problem including balanced sampling, two step training, sample re-weighting, and similarity loss functions. In this paper, we propose a generalized loss function based on the Tversky index to address the issue of data imbalance and achieve much better trade-off between precision and recall in training 3D fully convolutional deep neural networks. Experimental results in multiple sclerosis lesion segmentation on magnetic resonance images show improved \(F_2\) score, Dice coefficient, and the area under the precision-recall curve in test data. Based on these results we suggest Tversky loss function as a generalized framework to effectively train deep neural networks.