Showing papers by "Lehigh University published in 2019"

StackGAN++: Realistic Image Synthesis with Stacked Generative Adversarial Networks

Abstract: Although Generative Adversarial Networks (GANs) have shown remarkable success in various tasks, they still face challenges in generating high quality images. In this paper, we propose Stacked Generative Adversarial Networks (StackGANs) aimed at generating high-resolution photo-realistic images. First, we propose a two-stage generative adversarial network architecture, StackGAN-v1, for text-to-image synthesis. The Stage-I GAN sketches the primitive shape and colors of a scene based on a given text description, yielding low-resolution images. The Stage-II GAN takes Stage-I results and the text description as inputs, and generates high-resolution images with photo-realistic details. Second, an advanced multi-stage generative adversarial network architecture, StackGAN-v2, is proposed for both conditional and unconditional generative tasks. Our StackGAN-v2 consists of multiple generators and multiple discriminators arranged in a tree-like structure; images at multiple scales corresponding to the same scene are generated from different branches of the tree. StackGAN-v2 shows more stable training behavior than StackGAN-v1 by jointly approximating multiple distributions. Extensive experiments demonstrate that the proposed stacked generative adversarial networks significantly outperform other state-of-the-art methods in generating photo-realistic images.

Molecular interactions underlying liquid−liquid phase separation of the FUS low-complexity domain

Abstract: The low-complexity domain of the RNA-binding protein FUS (FUS LC) mediates liquid-liquid phase separation (LLPS), but the interactions between the repetitive SYGQ-rich sequence of FUS LC that stabilize the liquid phase are not known in detail. By combining NMR and Raman spectroscopy, mutagenesis, and molecular simulation, we demonstrate that heterogeneous interactions involving all residue types underlie LLPS of human FUS LC. We find no evidence that FUS LC adopts conformations with traditional secondary structure elements in the condensed phase; rather, it maintains conformational heterogeneity. We show that hydrogen bonding, π/sp2, and hydrophobic interactions all contribute to stabilizing LLPS of FUS LC. In addition to contributions from tyrosine residues, we find that glutamine residues also participate in contacts leading to LLPS of FUS LC. These results support a model in which FUS LC forms dynamic, multivalent interactions via multiple residue types and remains disordered in the densely packed liquid phase.

CHARMM-GUI Membrane Builder for Complex Biological Membrane Simulations with Glycolipids and Lipoglycans.

Abstract: Glycolipids (such as glycoglycerolipids, glycosphingolipids, and glycosylphosphatidylinositol) and lipoglycans (such as lipopolysaccharides (LPS), lipooligosaccharides (LOS), mycobacterial lipoarabinomannan, and mycoplasma lipoglycans) are typically found on the surface of cell membranes and play crucial roles in various cellular functions. Characterizing their structure and dynamics at the molecular level is essential to understand their biological roles, but systematic generation of glycolipid and lipoglycan structures is challenging because of great variations in lipid structures and glycan sequences (i.e., carbohydrate types and their linkages). To facilitate the generation of all-atom glycolipid/LPS/LOS structures, we have developed Glycolipid Modeler and LPS Modeler in CHARMM-GUI ( http://www.charmm-gui.org ), a web-based interface that simplifies building of complex biological simulation systems. In addition, we have incorporated these modules into Membrane Builder so that users can readily build a complex symmetric or asymmetric biological membrane system with various glycolipids and LPS/LOS. These tools are expected to be useful in innovative and novel glycolipid/LPS/LOS modeling and simulation research by easing tedious and intricate steps in modeling complex biological systems and shall provide insight into structures, dynamics, and underlying mechanisms of complex glycolipid-/LPS-/LOS-containing biological membrane systems.

DAG-GNN: DAG Structure Learning with Graph Neural Networks

Abstract: Learning a faithful directed acyclic graph (DAG) from samples of a joint distribution is a challenging combinatorial problem, owing to the intractable search space superexponential in the number of graph nodes. A recent breakthrough formulates the problem as a continuous optimization with a structural constraint that ensures acyclicity (Zheng et al., 2018). The authors apply the approach to the linear structural equation model (SEM) and the least-squares loss function that are statistically well justified but nevertheless limited. Motivated by the widespread success of deep learning that is capable of capturing complex nonlinear mappings, in this work we propose a deep generative model and apply a variant of the structural constraint to learn the DAG. At the heart of the generative model is a variational autoencoder parameterized by a novel graph neural network architecture, which we coin DAG-GNN. In addition to the richer capacity, an advantage of the proposed model is that it naturally handles discrete variables as well as vector-valued ones. We demonstrate that on synthetic data sets, the proposed method learns more accurate graphs for nonlinearly generated samples; and on benchmark data sets with discrete variables, the learned graphs are reasonably close to the global optima. The code is available at \url{this https URL}.

CHARMM-GUI Glycan Modeler for modeling and simulation of carbohydrates and glycoconjugates

Abstract: Characterizing glycans and glycoconjugates in the context of three-dimensional structures is important in understanding their biological roles and developing efficient therapeutic agents. Computational modeling and molecular simulation have become an essential tool complementary to experimental methods. Here, we present a computational tool, Glycan Modeler for in silico N-/O-glycosylation of the target protein and generation of carbohydrate-only systems. In our previous study, we developed Glycan Reader, a web-based tool for detecting carbohydrate molecules from a PDB structure and generation of simulation system and input files. As integrated into Glycan Reader in CHARMM-GUI, Glycan Modeler (Glycan Reader & Modeler) enables to generate the structures of glycans and glycoconjugates for given glycan sequences and glycosylation sites using PDB glycan template structures from Glycan Fragment Database (http://glycanstructure.org/fragment-db). Our benchmark tests demonstrate the universal applicability of Glycan Reader & Modeler to various glycan sequences and target proteins. We also investigated the structural properties of modeled glycan structures by running 2-μs molecular dynamics simulations of HIV envelope protein. The simulations show that the modeled glycan structures built by Glycan Reader & Modeler have the similar structural features compared to the ones solved by X-ray crystallography. We also describe the representative examples of glycoconjugate modeling with video demos to illustrate the practical applications of Glycan Reader & Modeler. Glycan Reader & Modeler is freely available at http://charmm-gui.org/input/glycan.

Black phosphorus and its isoelectronic materials

Abstract: The family of 2D and layered materials has been expanding rapidly for more than a decade. Within this large family of hundreds of materials, black phosphorus and its isoelectronic group IV monochalcogenides have a unique place. These puckered materials have distinctive crystalline symmetries and exhibit various exciting properties, such as high carrier mobility, strong infrared responsivity, widely tunable bandgap, in-plane anisotropy and spontaneous electric polarization. Here, we review their basic properties, highlight new electronic and photonic device concepts and novel physical phenomena and discuss future directions. Layered black phosphorus and its isoelectronic group IV monochalcogenides have distinctive physical properties arising from their unusual crystal symmetries. This Review discusses some of the interesting physical phenomena, possible device applications and future research directions for this group of materials.

Temperature-Controlled Liquid-Liquid Phase Separation of Disordered Proteins

Abstract: The liquid-liquid phase separation (LLPS) of intrinsically disordered proteins (IDPs) is a commonly observed phenomenon within the cell, and such condensates are also highly attractive for applications in biomaterials and drug delivery. A better understanding of the sequence-dependent thermoresponsive behavior is of immense interest as it will aid in the design of protein sequences with desirable properties and in the understanding of cellular response to heat stress. In this work, we use a transferable coarse-grained model to directly probe the sequence-dependent thermoresponsive phase behavior of IDPs. To achieve this goal, we develop a unique knowledge-based amino acid potential that accounts for the temperature-dependent effects on solvent-mediated interactions for different types of amino acids. Remarkably, we are able to distinguish between more than 35 IDPs with upper or lower critical solution temperatures at experimental conditions, thus providing direct evidence that incorporating the temperature-dependent solvent-mediated interactions to IDP assemblies can capture the difference in the shape of the resulting phase diagrams. Given the success of the model in predicting experimental behavior, we use it as a high-throughput screening framework to scan through millions of disordered sequences to characterize the composition dependence of protein phase separation.

Distributed Learning with Compressed Gradient Differences.

Abstract: Training large machine learning models requires a distributed computing approach, with communication of the model updates being the bottleneck. For this reason, several methods based on the compression (e.g., sparsification and/or quantization) of updates were recently proposed, including QSGD (Alistarh et al., 2017), TernGrad (Wen et al., 2017), SignSGD (Bernstein et al., 2018), and DQGD (Khirirat et al., 2018). However, none of these methods are able to learn the gradients, which renders them incapable of converging to the true optimum in the batch mode, incompatible with non-smooth regularizers, and slows down their convergence. In this work we propose a new distributed learning method --- DIANA --- which resolves these issues via compression of gradient differences. We perform a theoretical analysis in the strongly convex and nonconvex settings and show that our rates are superior to existing rates. Our analysis of block-quantization and differences between $\ell_2$ and $\ell_\infty$ quantization closes the gaps in theory and practice. Finally, by applying our analysis technique to TernGrad, we establish the first convergence rate for this method.

GRIP: Graph-based Interaction-aware Trajectory Prediction

Abstract: Nowadays, autonomous driving cars have become commercially available. However, the safety of a self-driving car is still a challenging problem that has not been well studied. Motion prediction is one of the core functions of an autonomous driving car. In this paper, we propose a novel scheme called GRIP which is designed to predict trajectories for traffic agents around an autonomous car efficiently. GRIP uses a graph to represent the interactions of close objects, applies several graph convolutional blocks to extract features, and subsequently uses an encoder-decoder long short-term memory (LSTM) model to make predictions. The experimental results on two well-known public datasets show that our proposed model improves the prediction accuracy of the state-of-the-art solution by 30%. The prediction error of GRIP is one meter shorter than existing schemes. Such an improvement can help autonomous driving cars avoid many traffic accidents. In addition, the proposed GRIP runs 5x faster than the state-of-the-art schemes.

Tuning of catalytic sites in Pt/TiO2 catalysts for the chemoselective hydrogenation of 3-nitrostyrene

Abstract: The catalytic activities of supported metal nanoparticles can be tuned by appropriate design of synthesis strategies. Each step in a catalyst synthesis method can play an important role in preparing the most efficient catalyst. Here we report the careful manipulation of the post-synthetic heat treatment procedure—together with control over the metal loading—to prepare a highly efficient 0.2 wt% Pt/TiO2 catalyst for the chemoselective hydrogenation of 3-nitrostyrene. For Pt/TiO2 catalysts with 0.2 and 0.5 wt% loading levels, reduction at 450 °C induces the coverage of TiOx over Pt nanoparticles through a strong metal–support interaction, which is detrimental to their catalytic activities. However, this can be avoided by following calcination treatment with reduction (both at 450 °C), allowing us to prepare an exceptionally active catalyst. Detailed characterization has revealed that the peripheral sites at the Pt/TiO2 interface are the most likely active sites for this hydrogenation reaction

TESS Discovery of an Ultra-short-period Planet around the Nearby M Dwarf LHS 3844

Abstract: Data from the newly commissioned Transiting Exoplanet Survey Satellite has revealed a "hot Earth" around LHS 3844, an M dwarf located 15 pc away. The planet has a radius of 1.303 ± 0.022 R⊕ and orbits the star every 11 hr. Although the existence of an atmosphere around such a strongly irradiated planet is questionable, the star is bright enough (I = 11.9, K = 9.1) for this possibility to be investigated with transit and occultation spectroscopy. The star's brightness and the planet's short period will also facilitate the measurement of the planet's mass through Doppler spectroscopy.

DEEPSEC: A Uniform Platform for Security Analysis of Deep Learning Model

Abstract: Deep learning (DL) models are inherently vulnerable to adversarial examples – maliciously crafted inputs to trigger target DL models to misbehave – which significantly hinders the application of DL in security-sensitive domains. Intensive research on adversarial learning has led to an arms race between adversaries and defenders. Such plethora of emerging attacks and defenses raise many questions: Which attacks are more evasive, preprocessing-proof, or transferable? Which defenses are more effective, utility-preserving, or general? Are ensembles of multiple defenses more robust than individuals? Yet, due to the lack of platforms for comprehensive evaluation on adversarial attacks and defenses, these critical questions remain largely unsolved. In this paper, we present the design, implementation, and evaluation of DEEPSEC, a uniform platform that aims to bridge this gap. In its current implementation, DEEPSEC incorporates 16 state-of-the-art attacks with 10 attack utility metrics, and 13 state-of-the-art defenses with 5 defensive utility metrics. To our best knowledge, DEEPSEC is the first platform that enables researchers and practitioners to (i) measure the vulnerability of DL models, (ii) evaluate the effectiveness of various attacks/defenses, and (iii) conduct comparative studies on attacks/defenses in a comprehensive and informative manner. Leveraging DEEPSEC, we systematically evaluate the existing adversarial attack and defense methods, and draw a set of key findings, which demonstrate DEEPSEC’s rich functionality, such as (1) the trade-off between misclassification and imperceptibility is empirically confirmed; (2) most defenses that claim to be universally applicable can only defend against limited types of attacks under restricted settings; (3) it is not necessary that adversarial examples with higher perturbation magnitude are easier to be detected; (4) the ensemble of multiple defenses cannot improve the overall defense capability, but can improve the lower bound of the defense effectiveness of individuals. Extensive analysis on DEEPSEC demonstrates its capabilities and advantages as a benchmark platform which can benefit future adversarial learning research.

Polarization of Λ (Λ[over ¯]) Hyperons along the Beam Direction in Au+Au Collisions at sqrt[s_{NN}]=200 GeV.

Abstract: The Λ (Λ[over ¯]) hyperon polarization along the beam direction has been measured in Au+Au collisions at sqrt[s_{NN}]=200 GeV, for the first time in heavy-ion collisions. The polarization dependence on the hyperons' emission angle relative to the elliptic flow plane exhibits a second harmonic sine modulation, indicating a quadrupole pattern of the vorticity component along the beam direction, expected due to elliptic flow. The polarization is found to increase in more peripheral collisions, and shows no strong transverse momentum (p_{T}) dependence at p_{T} greater than 1 GeV/c. The magnitude of the signal is about 5 times smaller than those predicted by hydrodynamic and multiphase transport models; the observed phase of the emission angle dependence is also opposite to these model predictions. In contrast, the kinematic vorticity calculations in the blast-wave model tuned to reproduce particle spectra, elliptic flow, and the azimuthal dependence of the Gaussian source radii measured with the Hanbury Brown-Twiss intensity interferometry technique reproduce well the modulation phase measured in the data and capture the centrality and transverse momentum dependence of the polarization signal.

Oxidative Coupling of Methane (OCM) by SiO2-Supported Tungsten Oxide Catalysts Promoted with Mn and Na

Abstract: The literature for the oxidative coupling of methane (OCM) on supported Mn/Na2WO4/SiO2 catalysts is systematically and critically reviewed. The influence of the precursors, starting SiO2 support cr...

Materials informatics for the screening of multi-principal elements and high-entropy alloys

Abstract: The field of multi-principal element or (single-phase) high-entropy (HE) alloys has recently seen exponential growth as these systems represent a paradigm shift in alloy development, in some cases exhibiting unexpected structures and superior mechanical properties. However, the identification of promising HE alloys presents a daunting challenge given the associated vastness of the chemistry/composition space. We describe here a supervised learning strategy for the efficient screening of HE alloys that combines two complementary tools, namely: (1) a multiple regression analysis and its generalization, a canonical-correlation analysis (CCA) and (2) a genetic algorithm (GA) with a CCA-inspired fitness function. These tools permit the identification of promising multi-principal element alloys. We implement this procedure using a database for which mechanical property information exists and highlight new alloys having high hardnesses. Our methodology is validated by comparing predicted hardnesses with alloys fabricated by arc-melting, identifying alloys having very high measured hardnesses.

Who is the "Human" in Human-Centered Machine Learning: The Case of Predicting Mental Health from Social Media

Abstract: "Human-centered machine learning" (HCML) combines human insights and domain expertise with data-driven predictions to answer societal questions. This area's inherent interdisciplinarity causes tensions in the obligations researchers have to the humans whose data they use. This paper studies how scientific papers represent human research subjects in HCML. Using mental health status prediction on social media as a case study, we conduct thematic discourse analysis on 55 papers to examine these representations. We identify five discourses that weave a complex narrative of who the human subject is in this research: Disorder/Patient, Social Media, Scientific, Data/Machine Learning, and Person. We show how these five discourses create paradoxical subject and object representations of the human, which may inadvertently risk dehumanization. We also discuss the tensions and impacts of interdisciplinary research; the risks of this work to scientific rigor, online communities, and mental health; and guidelines for stronger HCML research in this nascent area.

Mechanism by which Tungsten Oxide Promotes the Activity of Supported V 2 O 5 /TiO 2 Catalysts for NO X Abatement: Structural Effects Revealed by 51 V MAS NMR Spectroscopy.

Abstract: The selective catalytic reduction (SCR) of NOx with NH3 to N2 with supported V2 O5 (-WO3 )/TiO2 catalysts is an industrial technology used to mitigate toxic emissions. Long-standing uncertainties in the molecular structures of surface vanadia are clarified, whereby progressive addition of vanadia to TiO2 forms oligomeric vanadia structures and reveals a proportional relationship of SCR reaction rate to [surface VOx concentration]2 , implying a 2-site mechanism. Unreactive surface tungsta (WO3 ) also promote the formation of oligomeric vanadia (V2 O5 ) sites, showing that promoter incorporation enhances the SCR reaction by a structural effect generating adjacent surface sites and not from electronic effects as previously proposed. The findings outline a method to assess structural effects of promoter incorporation on catalysts and reveal both the dual-site requirement for the SCR reaction and the important structural promotional effect that tungsten oxide offers for the SCR reaction by V2 O5 /TiO2 catalysts.

Intrinsically reversible superglues via shape adaptation inspired by snail epiphragm

Abstract: Adhesives are ubiquitous in daily life and industrial applications. They usually fall into one of two classes: strong but irreversible (e.g., superglues) or reversible/reusable but weak (e.g., pressure-sensitive adhesives and biological and biomimetic surfaces). Achieving both superstrong adhesion and reversibility has been challenging. This task is particularly difficult for hydrogels that, because their major constituent is liquid water, typically do not adhere strongly to any material. Here, we report a snail epiphragm-inspired adhesion mechanism where a polymer gel system demonstrates superglue-like adhesion strength (up to 892 N⋅cm-2) that is also reversible. It is applicable to both flat and rough target surfaces. In its hydrated state, the softened gel conformally adapts to the target surface by low-energy deformation, which is locked upon drying as the elastic modulus is raised from hundreds of kilopascals to ∼2.3 GPa, analogous to the action of the epiphragm of snails. We show that in this system adhesion strength is based on the material's intrinsic, especially near-surface, properties and not on any near-surface structure, providing reversibility and ease of scaling up for practical applications.

Balancing Communication and Computation in Distributed Optimization

Abstract: Methods for distributed optimization have received significant attention in recent years owing to their wide applicability in various domains including machine learning, robotics, and sensor networks. A distributed optimization method typically consists of two key components: communication and computation. More specifically, at every iteration (or every several iterations) of a distributed algorithm, each node in the network requires some form of information exchange with its neighboring nodes (communication) and the computation step related to a (sub)-gradient (computation). The standard way of judging an algorithm via only the number of iterations overlooks the complexity associated with each iteration. Moreover, various applications deploying distributed methods may prefer a different composition of communication and computation. Motivated by this discrepancy, in this paper, we propose an adaptive cost framework that adjusts the cost measure depending on the features of various applications. We present a flexible algorithmic framework, where communication and computation steps are explicitly decomposed to enable algorithm customization for various applications. We apply this framework to the well-known distributed gradient descent (DGD) method, and show that the resulting customized algorithms, which we call DGD $^t$ , NEAR-DGD $^t$ , and NEAR-DGD $^+$ , compare favorably to their base algorithms, both theoretically and empirically. The proposed NEAR-DGD $^+$ algorithm is an exact first-order method where the communication and computation steps are nested, and when the number of communication steps is adaptively increased, the method converges to the optimal solution. We test the performance and illustrate the flexibility of the methods, as well as practical variants, on quadratic functions and classification problems that arise in machine learning, in terms of iterations, gradient evaluations, communications, and the proposed cost framework.

Convergence Rate Analysis of a Stochastic Trust-Region Method via Supermartingales

Abstract: We propose a novel framework for analyzing convergence rates of stochastic optimization algorithms with adaptive step sizes. This framework is based on analyzing properties of an underlying generic...

Simulation methods for liquid-liquid phase separation of disordered proteins.

Abstract: Liquid-liquid phase separation of intrinsically disordered proteins (IDPs) and other biomolecules is a highly complex but robust process used by living systems. Drawing inspiration from biology, phase separating proteins have been successfully utilized for promising applications in fields of materials design and drug delivery. These protein-based materials are advantageous due to the ability to finely tune their stimulus-responsive phase behavior and material properties, and the ability to encode biologically active motifs directly into the sequence. The number of possible protein sequences is virtually endless, which makes sequence-based design a rather daunting task, but also attractive due to the amount of control coming from exploration of this variable space. The use of computational methods in this field of research have come to the aid in several aspects, including interpreting experimental results, identifying important structural features and molecular mechanisms capable of explaining the phase behavior, and ultimately providing predictive frameworks for rational design of protein sequences. Here we provide an overview of computational studies focused on phase separating biomolecules and the tools that are available to researchers interested in this topic.

Mental health care among blacks in America: Confronting racism and constructing solutions.

Abstract: Objectives To describe reasons for unmet need for mental health care among blacks, identify factors associated with causes of unmet need, examine racism as a context of unmet need, and construct ways to improve service use. Data sources Data from the 2011-2015 National Survey on Drug Use and Health were pooled to create an analytic sample of black adults with unmet mental health need (N = 1237). Qualitative data came from focus groups (N = 30) recruited through purposive sampling. Study design Using sequential mixed methods, reasons for unmet need were regressed on sociodemographic, economic, and health characteristics of respondents. Findings were further explored in focus groups. Principal findings Higher education was associated with greater odds of reporting stigma and minimization of symptoms as reasons for unmet need. The fear of discrimination based on race and on mental illness was exacerbated among college-educated blacks. Racism causes mistrust in mental health service systems. Participants expressed the importance of anti-racism education and community-driven practice in reducing unmet need. Conclusion Mental health systems should confront racism and engage the historical and contemporary racial contexts within which black people experience mental health problems. Critical self-reflection at the individual level and racial equity analysis at the organizational level are critical.

Disentangling loosening from softening: insights into primary cell wall structure

Abstract: How cell wall elasticity, plasticity, and time-dependent extension (creep) relate to one another, to plant cell wall structure and to cell growth remain unsettled topics. To examine these issues without the complexities of living tissues, we treated cell-free strips of onion epidermal walls with various enzymes and other agents to assess which polysaccharides bear mechanical forces in-plane and out-of-plane of the cell wall. This information is critical for integrating concepts of wall structure, wall material properties, tissue mechanics and mechanisms of cell growth. With atomic force microscopy we also monitored real-time changes in the wall surface during treatments. Driselase, a potent cocktail of wall-degrading enzymes, removed cellulose microfibrils in superficial lamellae sequentially, layer-by-layer, and softened the wall (reduced its mechanical stiffness), yet did not induce wall loosening (creep). In contrast Cel12A, a bifunctional xyloglucanase/cellulase, induced creep with only subtle changes in wall appearance. Both Driselase and Cel12A increased the tensile compliance, but differently for elastic and plastic components. Homogalacturonan solubilization by pectate lyase and calcium chelation greatly increased the indentation compliance without changing tensile compliances. Acidic buffer induced rapid cell wall creep via endogenous α-expansins, with negligible effects on wall compliances. We conclude that these various wall properties are not tightly coupled and therefore reflect distinctive aspects of wall structure. Cross-lamellate networks of cellulose microfibrils influenced creep and tensile stiffness whereas homogalacturonan influenced indentation mechanics. This information is crucial for constructing realistic molecular models that define how wall mechanics and growth depend on primary cell wall structure.