Transfer Learning With Singular Value Decomposition of Multichannel Convolution Matrices.

The task of transfer learning using pretrained convolutional neural networks is considered. We propose a convolution-SVD layer to analyze the convolution operators with a singular value decomposition computed in the Fourier domain. Singular vectors extracted from the source domain are transferred to the target domain, whereas the singular values are fine-tuned with a target data set. In this way, dimension reduction is achieved to avoid overfitting, while some flexibility to fine-tune the convolution kernels is maintained. We extend an existing convolution kernel reconstruction algorithm to allow for a reconstruction from an arbitrary set of learned singular values. A generalization bound for a single convolution-SVD layer is devised to show the consistency between training and testing errors. We further introduce a notion of transfer learning gap. We prove that the testing error for a single convolution-SVD layer is bounded in terms of the gap, which motivates us to develop a regularization model with the gap as the regularizer. Numerical experiments are conducted to demonstrate the superiority of the proposed model in solving classification problems and the influence of various parameters. In particular, the regularization is shown to yield a significantly higher prediction accuracy.

Discovery of Dengue Virus NS4B Inhibitors.

The four serotypes of dengue virus (DENV-1 to -4) represent the most prevalent mosquito-borne viral pathogens in humans. No clinically approved vaccine or antiviral is currently available for DENV. Here we report a spiropyrazolopyridone compound that potently inhibits DENV both in vitro and in vivo. The inhibitor was identified through screening of a 1.8-million-compound library by using a DENV-2 replicon assay. The compound selectively inhibits DENV-2 and -3 (50% effective concentration [EC50], 10 to 80 nM) but not DENV-1 and -4 (EC50,>20 M). Resistance analysis showed that a mutation at amino acid 63 of DENV-2 NS4B (a nonenzymatic transmembrane protein and a component of the viral replication complex) could confer resistance to compound inhibition. Genetic studies demonstrate that variations at amino acid 63 of viral NS4B are responsible for the selective inhibition of DENV-2 and -3. Medicinal chemistry improved the physicochemical properties of the initial "hit" (compound 1), leading to compound 14a, which has good in vivo pharmacokinetics. Treatment of DENV-2-infected AG129 mice with compound 14a suppressed viremia, even when the treatment started after viral infection. The results have proven the concept that inhibitors of NS4B could potentially be developed for clinical treatment of DENV infection. Compound 14a represents a potential preclinical candidate for treatment of DENV-2- and -3-infected patients.

NMR analysis of a novel enzymatically active unlinked dengue NS2B-NS3 protease complex.

The dengue virus (DENV) is a mosquito-borne pathogen responsible for an estimated 100 million human infections annually. The viral genome encodes a two-component trypsin-like protease that contains the cofactor region from the nonstructural protein NS2B and the protease domain from NS3 (NS3pro). The NS2B-NS3pro complex plays a crucial role in viral maturation and has been identified as a potential drug target. Using a DENV protease construct containing NS2B covalently linked to NS3pro via a Gly4-Ser-Gly4 linker ("linked protease"), previous x-ray crystal structures show that the C-terminal fragment of NS2B is remote from NS3pro and exists in an open state in the absence of an inhibitor; however, in the presence of an inhibitor, NS2B complexes with NS3pro to form a closed state. This linked enzyme produced NMR spectra with severe signal overlap and line broadening. To obtain a protease construct with a resolved NMR spectrum, we expressed and purified an unlinked protease complex containing a 50-residue segment of the NS2B cofactor region and NS3pro without the glycine linker using a coexpression system. This unlinked protease complex was catalytically active at neutral pH in the absence of glycerol and produced dispersed cross-peaks in a (1)H-(15)N heteronuclear single quantum correlation spectrum that enabled us to conduct backbone assignments using conventional techniques. In addition, titration with an active-site peptide aldehyde inhibitor and paramagnetic relaxation enhancement studies demonstrated that the unlinked DENV protease exists predominantly in a closed conformation in solution. This protease complex can serve as a useful tool for drug discovery against DENV.

Multi-relational graph convolutional networks: Generalization guarantees and experiments.

The class of multi-relational graph convolutional networks (MRGCNs) is a recent extension of standard graph convolutional networks (GCNs) to handle heterogenous graphs with multiple types of relationships. MRGCNs have been shown to yield results superior than traditional GCNs in various machine learning tasks. The key idea is to introduce a new kind of convolution operated on tensors that can effectively exploit correlations exhibited in multiple relationships. The main objective of this paper is to analyze the algorithmic stability and generalization guarantees of MRGCNs to confirm the usefulness of MRGCNs. Our contributions are of three folds. First, we develop a matrix representation of various tensor operations underneath MRGCNs to simplify the analysis significantly. Next, we prove the uniform stability of MRGCNs and deduce the convergence of the generalization gap to support the usefulness of MRGCNs. The analysis sheds lights on the design of MRGCNs, for instance, how the data should be scaled to achieve the uniform stability of the learning process. Finally, we provide experimental results to demonstrate the stability results.

A single amino acid in nonstructural protein NS4B confers virulence to dengue virus in AG129 mice through enhancement of viral RNA synthesis.

Dengue (DEN) is a mosquito-borne viral disease that has become an increasing economic and health burden for the tropical and subtropical world. The lack of an appropriate animal model of DEN has greatly impeded the study of its pathogenesis and the development of vaccines/antivirals. We recently reported a DEN virus 2 (DENV-2) strain (D2Y98P) that lethally infects immunocompromised AG129 mice, resulting in organ damage or dysfunction and increased vascular permeability, hallmarks of severe DEN in patients (G. K. Tan et al., PLoS Negl. Trop. Dis. 4:e672, 2010). Here we report the identification of one critical virulence determinant of strain D2Y98P. By mutagenesis, we showed that a Phe-to-Leu alteration at amino acid position 52 in nonstructural protein NS4B completely abolished the pathogenicity of the D2Y98P virus, as evidenced by a lack of lethality and the absence of histological signs of disease, which correlated with reduced viral titers and intact vascular permeability. Conversely, a Leu-to-Phe alteration at position 52 of NS4B in nonvirulent DENV-2 strain TSV01 led to 80% lethality and increased viremia. The NS4B(Phe52) viruses displayed enhanced RNA synthesis in mammalian cells but not in mosquito cells. The increased viral RNA synthesis was independent of the ability of NS4B to interfere with the host type I interferon response. Overall, our results demonstrate that Phe at position 52 in NS4B confers virulence in mice on two independent DENV-2 strains through enhancement of viral RNA synthesis. In addition to providing further insights into the functional role of NS4B protein, our findings further support a direct relationship between viral loads and DEN pathogenesis in vivo, consistent with observations in DEN patients.

An adenosine nucleoside inhibitor of dengue virus.

Dengue virus (DENV), a mosquito-borne flavivirus, is a major public health threat. The virus poses risk to 2.5 billion people worldwide and causes 50 to 100 million human infections each year. Neither a vaccine nor an antiviral therapy is currently available for prevention and treatment of DENV infection. Here, we report a previously undescribed adenosine analog, NITD008, that potently inhibits DENV both in vitro and in vivo. In addition to the 4 serotypes of DENV, NITD008 inhibits other flaviviruses, including West Nile virus, yellow fever virus, and Powassan virus. The compound also suppresses hepatitis C virus, but it does not inhibit nonflaviviruses, such as Western equine encephalitis virus and vesicular stomatitis virus. A triphosphate form of NITD008 directly inhibits the RNA-dependent RNA polymerase activity of DENV, indicating that the compound functions as a chain terminator during viral RNA synthesis. NITD008 has good in vivo pharmacokinetic properties and is biologically available through oral administration. Treatment of DENV-infected mice with NITD008 suppressed peak viremia, reduced cytokine elevation, and completely prevented the infected mice from death. No observed adverse effect level (NOAEL) was achieved when rats were orally dosed with NITD008 at 50 mg/kg daily for 1 week. However, NOAEL could not be accomplished when rats and dogs were dosed daily for 2 weeks. Nevertheless, our results have proved the concept that a nucleoside inhibitor could be developed for potential treatment of flavivirus infections.

Subspace learning for Mumford-Shah-model-based texture segmentation through texture patches.

In this paper, we develop a robust and effective algorithm for texture segmentation and feature selection. The approach is to incorporate a patch-based subspace learning technique into the subspace Mumford-Shah (SMS) model to make the minimization of the SMS model robust and accurate. The proposed method is fully unsupervised in that it removes the need to specify training data, which is required by existing methods for the same model. We further propose a novel (to our knowledge) pairwise dissimilarity measure for pixels. Its novelty lies in the use of the relevance scores of the features of each pixel to improve its discriminating power. Some superior results are obtained compared to existing unsupervised algorithms, which do not use a subspace approach. This confirms the usefulness of the subspace approach and the proposed unsupervised algorithm.

Inhibition of dengue virus by an ester prodrug of an adenosine analog.

Dengue virus (DENV) is the most prevalent mosquito-borne viral pathogen that infects humans. Neither a vaccine nor an antiviral therapy is currently available for DENV. Here, we report an adenosine nucleoside prodrug that potently inhibits DENV replication both in cell culture and in a DENV mouse model. NITD449 (2'-C-acetylene-7-deaza-7-carbamoyladenosine) was initially identified as a parental compound that inhibits all four serotypes of DENV with low cytotoxicity. However, in vivo pharmacokinetic studies indicated that NITD449 had a low level of exposure in plasma when dosed orally. To increase the oral bioavailability, we covalently linked isobutyric acids to the 3'- and 5'-hydroxyl groups of ribose via ester linkage to NITD449, leading to the prodrug NITD203 (3',5'-O-diisobutyryl-2'-C-acetylene-7-deaza-7-carbamoyl-adenosin). Pharmacokinetic analysis showed that upon oral dosing of the prodrug, NITD203 was readily converted to NITD449, resulting in improved exposure of the parental compound in plasma in both mouse and rat. In DENV-infected AG129 mice, oral dosing of the prodrug at 25 mg/kg of body weight reduced peak viremia by 30-fold. Antiviral spectrum analysis showed that NITD203 inhibited various flaviviruses (DENV, yellow fever virus, and West Nile virus) and hepatitis C virus but not Chikungunya virus (an alphavirus). Mode-of-action analysis, using a luciferase-reporting replicon, indicated that NITD203 inhibited DENV RNA synthesis. Although NITD203 exhibited potent in vitro and in vivo efficacies, the compound could not reach a satisfactory no-observable-adverse-effect level (NOAEL) in a 2-week in vivo toxicity study. Nevertheless, our results demonstrate that a prodrug approach using a nucleoside analog could potentially be developed for flavivirus antiviral therapy.

Inhibition of dengue virus RNA synthesis by an adenosine nucleoside.

We recently reported that (2R,3R,4R,5R)-2-(4-amino-pyrrolo[2,3-d]pyrimidin-7-yl)-3-ethynyl-5-hydroxy-methyl-tetrahydro-furan-3,4-diol is a potent inhibitor of dengue virus (DENV), with 50% effective concentration (EC(50)) and cytotoxic concentration (CC(50)) values of 0.7 microM and >100 microM, respectively. Here we describe the synthesis, structure-activity relationship, and antiviral characterization of the inhibitor. In an AG129 mouse model, a single-dose treatment of DENV-infected mice with the compound suppressed peak viremia and completely prevented death. Mode-of-action analysis using a DENV replicon indicated that the compound blocks viral RNA synthesis. Recombinant adenosine kinase could convert the compound to a monophosphate form. Suppression of host adenosine kinase, using a specific inhibitor (iodotubercidin) or small interfering RNA (siRNA), abolished or reduced the compound's antiviral activity in cell culture. Studies of rats showed that (14)C-labeled compound was converted to mono-, di-, and triphosphate metabolites in vivo. Collectively, the results suggest that this adenosine inhibitor is phosphorylated to an active (triphosphate) form which functions as a chain terminator for viral RNA synthesis.

Semi-supervised subspace learning for Mumford-Shah model based texture segmentation.

We propose a novel image segmentation model which incorporates subspace clustering techniques into a Mumford-Shah model to solve texture segmentation problems. While the natural unsupervised approach to learn a feature subspace can easily be trapped in a local solution, we propose a novel semi-supervised optimization algorithm that makes use of information derived from both the intermediate segmentation results and the regions-of-interest (ROI) selected by the user to determine the optimal subspaces of the target regions. Meanwhile, these subspaces are embedded into a Mumford-Shah objective function so that each segment of the optimal partition is homogeneous in its own subspace. The method outperforms standard Mumford-Shah models since it can separate textures which are less separated in the full feature space. Experimental results are presented to confirm the usefulness of subspace clustering in texture segmentation.

Fast splitting algorithm for multiframe total variation blind video deconvolution.

We consider the recovery of degraded videos without complete knowledge about the degradation. A spatially shift-invariant but temporally shift-varying video formation model is used. This leads to a simple multiframe degradation model that relates each original video frame with multiple observed frames and point spread functions (PSFs). We propose a variational method that simultaneously reconstructs each video frame and the associated PSFs from the corresponding observed frames. Total variation (TV) regularization is used on both the video frames and the PSFs to further reduce the ill-posedness and to better preserve edges. In order to make TV minimization practical for video sequences, we propose an efficient splitting method that generalizes some recent fast single-image TV minimization methods to the multiframe case. Both synthetic and real videos are used to show the performance of the proposed method.

A fast optimization transfer algorithm for image inpainting in wavelet domains.

A wavelet inpainting problem refers to the problem of filling in missing wavelet coefficients in an image. A variational approach was used by Chan et al. The resulting functional was minimized by the gradient descent method. In this paper, we use an optimization transfer technique which involves replacing their univariate functional by a bivariate functional by adding an auxiliary variable. Our bivariate functional can be minimized easily by alternating minimization: for the auxiliary variable, the minimum has a closed form solution, and for the original variable, the minimization problem can be formulated as a classical total variation (TV) denoising problem and, hence, can be solved efficiently using a dual formulation. We show that our bivariate functional is equivalent to the original univariate functional. We also show that our alternating minimization is convergent. Numerical results show that the proposed algorithm is very efficient and outperforms that of Chan et al.

The Plasmodium liver-specific protein 2 (LISP2) is an early marker of liver stage development.

Plasmodium vivax hypnozoites persist in the liver, cause malaria relapse and represent a major challenge to malaria elimination. Our previous transcriptomic study provided a novel molecular framework to enhance our understanding of the hypnozoite biology (Voorberg-van der Wel A, et al., 2017). In this dataset, we identified and characterized the Liver-Specific Protein 2 (LISP2) protein as an early molecular marker of liver stage development. Immunofluorescence analysis of hepatocytes infected with relapsing malaria parasites, in vitro (P. cynomolgi) and in vivo (P. vivax), reveals that LISP2 expression discriminates between dormant hypnozoites and early developing parasites. We further demonstrate that prophylactic drugs selectively kill all LISP2-positive parasites, while LISP2-negative hypnozoites are only sensitive to anti-relapse drug tafenoquine. Our results provide novel biological insights in the initiation of liver stage schizogony and an early marker suitable for the development of drug discovery assays predictive of anti-relapse activity.

Hepatic spheroids used as an in vitro model to study malaria relapse.

Hypnozoites are the liver stage non-dividing form of the malaria parasite that are responsible for relapse and acts as a natural reservoir for human malaria Plasmodium vivax and P. ovale as well as a phylogenetically related simian malaria P. cynomolgi. Our understanding of hypnozoite biology remains limited due to the technical challenge of requiring the use of primary hepatocytes and the lack of robust and predictive in vitro models. In this study, we developed a malaria liver stage model using 3D spheroid-cultured primary hepatocytes. The infection of primary hepatocytes in suspension led to increased infectivity of both P. cynomolgi and P. vivax infections. We demonstrated that this hepatic spheroid model was capable of maintaining long term viability, hepatocyte specific functions and cell polarity which enhanced permissiveness and thus, permitting for the complete development of both P. cynomolgi and P. vivax liver stage parasites in the infected spheroids. The model described here was able to capture the full liver stage cycle starting with sporozoites and ending in the release of hepatic merozoites capable of invading simian erythrocytes in vitro. Finally, we showed that this system can be used for compound screening to discriminate between causal prophylactic and cidal antimalarials activity in vitro for relapsing malaria.

Robust continuous in vitro culture of the Plasmodium cynomolgi erythrocytic stages.

The ability to culture pathogenic organisms substantially enhances the quest for fundamental knowledge and the development of vaccines and drugs. Thus, the elaboration of a protocol for the in vitro cultivation of the erythrocytic stages of Plasmodium falciparum revolutionized research on this important parasite. However, for P. vivax, the most widely distributed and difficult to treat malaria parasite, a strict preference for reticulocytes thwarts efforts to maintain it in vitro. Cultivation of P. cynomolgi, a macaque-infecting species phylogenetically close to P. vivax, was briefly reported in the early 1980s, but not pursued further. Here, we define the conditions under which P. cynomolgi can be adapted to long term in vitro culture to yield parasites that share many of the morphological and phenotypic features of P. vivax. We further validate the potential of this culture system for high-throughput screening to prime and accelerate anti-P. vivax drug discovery efforts.

A multiresolution stochastic level set method for Mumford-Shah image segmentation.

The Mumford-Shah model is one of the most successful image segmentation models. However, existing algorithms for the model are often very sensitive to the choice of the initial guess. To make use of the model effectively, it is essential to develop an algorithm which can compute a global or near global optimal solution efficiently. While gradient descent based methods are well-known to find a local minimum only, even many stochastic methods do not provide a practical solution to this problem either. In this paper, we consider the computation of a global minimum of the multiphase piecewise constant Mumford-Shah model. We propose a hybrid approach which combines gradient based and stochastic optimization methods to resolve the problem of sensitivity to the initial guess. At the heart of our algorithm is a well-designed basin hopping scheme which uses global updates to escape from local traps in a way that is much more effective than standard stochastic methods. In our experiments, a very high-quality solution is obtained within a few stochastic hops whereas the solutions obtained with simulated annealing are incomparable even after thousands of steps. We also propose a multiresolution approach to reduce the computational cost and enhance the search for a global minimum. Furthermore, we derived a simple but useful theoretical result relating solutions at different spatial resolutions.

Gene network interconnectedness and the generalized topological overlap measure.

BACKGROUND: Network methods are increasingly used to represent the interactions of genes and/or proteins. Genes or proteins that are directly linked may have a similar biological function or may be part of the same biological pathway. Since the information on the connection (adjacency) between 2 nodes may be noisy or incomplete, it can be desirable to consider alternative measures of pairwise interconnectedness. Here we study a class of measures that are proportional to the number of neighbors that a pair of nodes share in common. For example, the topological overlap measure by Ravasz et al. 1 can be interpreted as a measure of agreement between the m = 1 step neighborhoods of 2 nodes. Several studies have shown that two proteins having a higher topological overlap are more likely to belong to the same functional class than proteins having a lower topological overlap. Here we address the question whether a measure of topological overlap based on higher-order neighborhoods could give rise to a more robust and sensitive measure of interconnectedness. RESULTS: We generalize the topological overlap measure from m = 1 step neighborhoods to m > or = 2 step neighborhoods. This allows us to define the m-th order generalized topological overlap measure (GTOM) by (i) counting the number of m-step neighbors that a pair of nodes share and (ii) normalizing it to take a value between 0 and 1. Using theoretical arguments, a yeast co-expression network application, and a fly protein network application, we illustrate the usefulness of the proposed measure for module detection and gene neighborhood analysis. CONCLUSION: Topological overlap can serve as an important filter to counter the effects of spurious or missing connections between network nodes. The m-th order topological overlap measure allows one to trade-off sensitivity versus specificity when it comes to defining pairwise interconnectedness and network modules.

Strategies for identifying statistically significant dense regions in microarray data.

We propose and study the notion of dense regions for the analysis of categorized gene expression data and present some searching algorithms for discovering them. The algorithms can be applied to any categorical data matrices derived from gene expression level matrices. We demonstrate that dense regions are simple but useful and statistically significant patterns that can be used to 1) identify genes and/or samples of interest and 2) eliminate genes and/or samples corresponding to outliers, noise, or abnormalities. Some theoretical studies on the properties of the dense regions are presented which allow us to characterize dense regions into several classes and to derive tailor-made algorithms for different classes of regions. Moreover, an empirical simulation study on the distribution of the size of dense regions is carried out which is then used to assess the significance of dense regions and to derive effective pruning methods to speed up the searching algorithms. Real microarray data sets are employed to test our methods. Comparisons with six other well-known clustering algorithms using synthetic and real data are also conducted which confirm the superiority of our methods in discovering dense regions. The DRIFT code and a tutorial are available as supplemental material, which can be found on the Computer Society Digital Library at http://computer.org/tcbb/archives.htm.

A primal-dual active-set method for non-negativity constrained total variation deblurring problems.

This paper studies image deblurring problems using a total variation-based model, with a non-negativity constraint. The addition of the non-negativity constraint improves the quality of the solutions, but makes the solution process a difficult one. The contribution of our work is a fast and robust numerical algorithm to solve the non-negatively constrained problem. To overcome the nondifferentiability of the total variation norm, we formulate the constrained deblurring problem as a primal-dual program which is a variant of the formulation proposed by Chan, Golub, and Mulet for unconstrained problems. Here, dual refers to a combination of the Lagrangian and Fenchel duals. To solve the constrained primal-dual program, we use a semi-smooth Newton's method. We exploit the relationship between the semi-smooth Newton's method and the primal-dual active set method to achieve considerable simplification of the computations. The main advantages of our proposed scheme are: no parameters need significant adjustment, a standard inverse preconditioner works very well, quadratic rate of local convergence (theoretical and numerical), numerical evidence of global convergence, and high accuracy of solving the optimality system. The scheme shows robustness of performance over a wide range of parameters. A comprehensive set of numerical comparisons are provided against other methods to solve the same problem which show the speed and accuracy advantages of our scheme.

A genomics approach to understanding host response during dengue infection.

Dengue infection results in a wide clinical spectrum, ranging from asymptomatic, through fever (DF), to the life threatening complications haemorrhagic fever (DHF) and shock syndrome (DSS). Although we now understand that factors such as repeat infections and the type or magnitude of the host response are important in determining severity, the mechanisms of these actions remain largely unknown. Understanding this host-pathogen interaction may enable outcome prediction and new therapy options. Developments in biology now allow a 'systems approach' to be applied to this problem, utilizing whole genomes of both human and virus, in vitro and in vivo to enable a more complete picture of their interplay to be built up. We have developed a chip-based approach to viral sequencing, to increase efficiency and enable large numbers ofgenomes to be completed, together with a web-based interpretation tool. We have also applied human whole genome expression arrays (24000 genes) to characterize the types of host response made to infection and plan to investigate the role of host variation using human whole genome genetic association studies in the future. These technologies have identified novel host pathways involved in viral replication in vitro, and also host immune responses, such as the interferon signalling pathway, that are influenced by viral sequence. This data will be further refined through interlinking with similar data obtained from a large study of dengue patients, initiated in Singapore, that is able to look at the early host response to infection.