Huy Nguyen

Named entity recognition with character-level models

We discuss two named-entity recognition models which use characters and character n-grams either exclusively or as an important part of their data representation. The first model is a character-level HMM with minimal context information, and the second model is a maximum-entropy conditional markov model with substantially richer context features. Our best model achieves an overall F1 of 86.07% on the English test data (92.31% on the development data). This number represents a 25% error reduction over the same model without word-internal (substring) features.

Exploiting context for biomedical entity recognition: from syntax to the web

We describe a machine learning system for the recognition of names in biomedical texts. The system makes extensive use of local and syntactic features within the text, as well as external resources including the web and gazetteers. It achieves an F-score of 70% on the Coling 2004 NLPBA/BioNLP shared task of identifying five biomedical named entities in the GENIA corpus.

CNS bioavailability and radiation protection of normal hippocampal neurogenesis by a lipophilic Mn porphyrin-based superoxide dismutase mimic, MnTnBuOE-2-PyP5+

Although radiation therapy can be effective against cancer, potential damage to normal tissues limits the amount that can be safely administered. In central nervous system (CNS), radiation damage to normal tissues is presented, in part, as suppressed hippocampal neurogenesis and impaired cognitive functions. Mn porphyrin (MnP)-based redox active drugs have demonstrated differential effects on cancer and normal tissues in experimental animals that lead to protection of normal tissues and radio- and chemo-sensitization of cancers. To test the efficacy of MnPs in CNS radioprotection, we first examined the tissue levels of three different MnPs – MnTE-2-PyP5+(MnE), MnTnHex-2-PyP5+(MnHex), and MnTnBuOE-2-PyP5+(MnBuOE). Nanomolar concentrations of MnHex and MnBuOE were detected in various brain regions after daily subcutaneous administration, and MnBuOE was well tolerated at a daily dose of 3 mg/kg. Administration of MnBuOE for one week before cranial irradiation and continued for one week afterwards supported production and long-term survival of newborn neurons in the hippocampal dentate gyrus. MnP-driven S-glutathionylation in cortex and hippocampus showed differential responses to MnP administration and radiation in these two brain regions. A better understanding of how preserved hippocampal neurogenesis correlates with cognitive functions following cranial irradiation will be helpful in designing better MnP-based radioprotection strategies.

Peptide library synthesis on spectrally encoded beads for multiplexed protein/peptide bioassays

Protein-peptide interactions are essential for cellular responses. Despite their importance, these interactions remain largely uncharacterized due to experimental challenges associated with their measurement. Current techniques (e.g. surface plasmon resonance, fluorescence polarization, and isothermal calorimetry) either require large amounts of purified material or direct fluorescent labeling, making high-throughput measurements laborious and expensive. In this report, we present a new technology for measuring antibody-peptide interactions in vitro that leverages spectrally encoded beads for biological multiplexing. Specific peptide sequences are synthesized directly on encoded beads with a 1:1 relationship between peptide sequence and embedded code, thereby making it possible to track many peptide sequences throughout the course of an experiment within a single small volume. We demonstrate the potential of these bead-bound peptide libraries by: (1) creating a set of 46 peptides composed of 3 commonly used epitope tags (myc, FLAG, and HA) and single amino-acid scanning mutants; (2) incubating with a mixture of fluorescently-labeled antimyc, anti-FLAG, and anti-HA antibodies; and (3) imaging these bead-bound libraries to simultaneously identify the embedded spectral code (and thus the sequence of the associated peptide) and quantify the amount of each antibody bound. To our knowledge, these data demonstrate the first customized peptide library synthesized directly on spectrally encoded beads. While the implementation of the technology provided here is a high-affinity antibody/protein interaction with a small code space, we believe this platform can be broadly applicable to any range of peptide screening applications, with the capability to multiplex into libraries of hundreds to thousands of peptides in a single assay.

Automatic semantic grouping in a spoken language user interface toolkit

With the rapid growth of real application domains for NLP systems, there is a genuine demand for a general toolkit from which programmers with no linguistic knowledge can build specific NLP systems. Such a toolkit should provide an interface to accept sample sentences and convert them into semantic representations so as to allow programmers to map them to domain actions. In order to reduce the workload of managing a large number of semantic forms individually, the toolkit will perform what we call semantic grouping to organize the forms into meaningful groups. In this paper, we present three semantic grouping methods: similarity-based, verb-based and category-based grouping, and their implementation in the SLUI toolkit. We also discuss the pros and cons of each method and how they can be utilized according to the different domain needs.

Discovering novel calcineurin inhibitors through quantitative mapping of protein-peptide affinity landscapes

Transient, regulated binding of globular protein domains to Short Linear Motifs (SLiMs) in disordered regions of other proteins drives cellular signaling. Mapping the energy landscapes of these interactions is essential for deciphering and therapeutically perturbing signaling networks, but is challenging due to their weak affinities. We present a powerful technology, MRBLE-pep, that simultaneously quantifies protein binding to a library of peptides directly synthesized on beads containing unique spectral codes. Using computational modeling and MRBLE-pep, we systematically probe binding of calcineurin (CN), a conserved protein phosphatase essential for the immune response and target of immunosuppressants, to the PxIxIT SLiM. We establish that flanking residues and post-translational modifications critically contribute to PxIxIT-CN affinity, and discover CN-inhibitory peptides with unprecedented affinity and therapeutic potential. The quantitative measurements provided by this approach will improve computational modeling efforts, elucidate a broad range of weak protein-SLiM interactions, and revolutionize our understanding of signaling networks.Weak, transient binding of globular protein domains to Short Linear Motifs (SLiMs) in disordered regions of other proteins drive cellular signaling. Mapping the energy landscape of such interactions is essential for deciphering signaling networks and developing therapeutic inhibitors, but is complicated by technical challenges associated with quantitatively measuring weak interactions in high-throughput. Here, we synthesize peptide libraries on spectrally encoded beads with each peptide sequence uniquely linked to a spectral code (MRBLE-pep), allowing high-throughput quantification of protein-peptide binding via imaging. Using computational modeling and MRBLE-pep assays, we map the affinity landscape for human calcineurin (CN), a conserved protein phosphatase essential for the immune response and target of immunosuppressants, interacting with a known SLiM (PxIxIT). We find that PxIxIT recognition depends critically on flanking residues and is regulated by post-translational modifications. Using this information, we designed PxIxIT peptides with unprecedented affinity and therapeutic potential that strongly inhibit CN function in vivo.Weak, transient binding of globular protein domains to Short Linear Motifs (SLiMs) in disordered regions of other proteins drive cellular signaling. Mapping the energy landscape of such interactions is essential for deciphering signaling networks and developing therapeutic inhibitors, but is complicated by technical challenges associated with quantitatively measuring weak interactions in high-throughput. Here, we synthesize peptide libraries on spectrally encoded beads with each peptide sequence uniquely linked to a spectral code (MRBLE-pep), allowing high-throughput quantification of protein-peptide binding via imaging. Using computational modeling and MRBLE-pep assays, we map the affinity landscape for human calcineurin (CN), a conserved protein phosphatase essential for the immune response and target of immunosuppressants, interacting with a known SLiM (PxIxIT). We find that PxIxIT recognition depends critically on flanking residues and is regulated by post-translational modifications. Using this information, we designed PxIxIT peptides with unprecedented affinity and therapeutic potential that strongly inhibit CN function in vivo.

An Open-Source Software Analysis Package for Microspheres with Ratiometric Barcode Lanthanide Encoding (MRBLEs)

Multiplexed bioassays, in which multiple analytes of interest are probed in parallel within a single small volume, have greatly accelerated the pace of biological discovery. Bead-based multiplexed bioassays have many technical advantages, including near solution-phase kinetics, small sample volume requirements, many within-assay replicates to reduce measurement error, and, for some bead materials, the ability to synthesize analytes directly on beads via solid-phase synthesis. To allow bead-based multiplexing, analytes can be synthesized on spectrally encoded beads with a 1:1 linkage between analyte identity and embedded codes. Bead-bound analyte libraries can then be pooled and incubated with a fluorescently-labeled macromolecule of interest, allowing downstream quantification of interactions between the macromolecule and all analytes simultaneously via imaging alone. Extracting quantitative binding data from these images poses several computational image processing challenges, requiring the ability to identify all beads in each image, quantify bound fluorescent material associated with each bead, and determine their embedded spectral code to reveal analyte identities. Here, we present a novel open-source Python software package (the mrbles analysis package) that provides the necessary tools to: (1) find encoded beads in a bright-field microscopy image; (2) quantify bound fluorescent material associated with bead perimeters; (3) identify embedded ratiometric spectral codes within beads; and (4) return data aggregated by embedded code and for each individual bead. We demonstrate the utility of this package by applying it towards analyzing data generated via multiplexed measurement of calcineurin protein binding to MRBLEs (Microspheres with Ratiometric Barcode Lanthanide Encoding) containing known and mutant binding peptide motifs. We anticipate that this flexible package should be applicable to a wide variety of assays, including simple bead or droplet finding analysis, quantification of binding to non-encoded beads, and analysis of multiplexed assays that use ratiometric, spectrally encoded beads.

Walking along chromosomes with super-resolution imaging, contact maps, and integrative modeling

Chromosome structure is thought to be crucial for proper functioning of the nucleus. Here, we present a method for visualizing chromosomal DNA at super-resolution and then integrating Hi-C data to produce three-dimensional models of chromosome organization. We begin by applying Oligopaint probes and the single-molecule localization microscopy methods of OligoSTORM and OligoDNA-PAINT to image 8 megabases of human chromosome 19, discovering that chromosomal regions contributing to compartments can form distinct structures. Intriguingly, our data also suggest that homologous maternal and paternal regions may be differentially organized. Finally, we integrate imaging data with Hi-C and restraint-based modeling using a method called integrative modeling of genomic regions (IMGR) to increase the genomic resolution of our traces to 10 kb. One Sentence Summary Super-resolution genome tracing, contact maps, and integrative modeling enable 10 kb resolution glimpses of chromosome folding.

The hippocampal extracellular matrix regulates pain and memory after injury

Chronic pain poses a heavy burden for the individual and society, comprising personal suffering, comorbid psychiatric symptoms, cognitive decline, and disability. Treatment options are poor due in large part to pain centralization, where an initial injury can result in lasting CNS maladaptations. Hippocampal cellular plasticity in chronic pain has become a focus of study due to its roles in cognition, memory, and the experience of pain itself. However, the extracellular alterations that parallel and facilitate changes in hippocampal function have not been addressed to date. Here we show structural and biochemical plasticity in the hippocampal extracellular matrix (ECM) that is linked to behavioral, cellular, and synaptic changes in a mouse model of chronic pain. Specifically, we report deficits in working location memory that are associated with decreased hippocampal dendritic complexity, altered ECM microarchitecture, decreased ECM rigidity, and changes in the levels of key ECM components and enzymes, including increased levels of MMP8. We also report aberrations in long-term potentiation (LTP) and a loss of inhibitory interneuron perineuronal ECM nets, potentially accounting for the aberrations in LTP. Finally, we demonstrate that MMP8 is upregulated after injury and that its genetic downregulation normalizes the behavioral, electrophysiological, and extracellular alterations. By linking specific extracellular changes to the chronic pain phenotype, we provide a novel mechanistic understanding of pain centralization that provides new targets for the treatment of chronic pain.

MicroRNA-Mediated Therapy Modulating Blood–Brain Barrier Disruption Improves Vascular Cognitive Impairment

Objective— There are currently no effective treatments for the prevention of dementia associated with vascular cognitive impairment. MicroRNAs regulate gene expression at the post-transcriptional level and play key roles in vascular disorders. TNF&agr; (tumor necrosis factor-&agr;) regulates blood–brain barrier breakdown through modification of cerebral tight junctions. Here, we sought key TNF&agr;-responsive microRNAs that might influence blood–brain barrier breakdown via cerebral tight junction disruption in vascular cognitive impairment. Approach and Results— Using a mouse model of vascular cognitive impairment, chronic cerebral hypoperfusion within the white matter was induced with bilateral common carotid artery stenosis (BCAS) surgery. TNF&agr; gene expression was increased in white matter post-BCAS surgery, and TNF&agr; stimulation decreased claudin-5, ZO-1 (tight-junction protein 1), and occludin gene expression in murine brain endothelial cells. In silico analysis predicted 8 candidate microRNAs as regulators of claudin-5, ZO-1, and occludin gene expression. Of these, only miR-501-3p was upregulated by TNF&agr; in vitro and was upregulated in the white matter after BCAS surgery. Further, miR-501-3p directly bound to the 3′-untranslated region of human ZO-1 and downregulated transendothelial electric resistance. In vivo administration of a locked nucleic acid –modified antisense oligonucleotide versus miR-501-3p suppressed BCAS-induced reduction of ZO-1 gene expression and blood–brain barrier disruption within the white matter and significantly ameliorated working memory deficits after BCAS surgery. Conclusions— We here provide the first evidence that the TNF&agr;–miR-501-3p–ZO-1 axis plays an important role in the pathogenesis of cerebral hypoperfusion–induced working memory deficits and white matter lesions, as a result of blood–brain barrier breakdown via tight junction disruption. Therapeutic manipulation of miR-501-3p holds promise for limiting vascular cognitive impairment progression.