Derivative Kernels for Noise Robust ASR

Recently there has been interest in combined gen- erative/discriminative classifiers. In these classifiers features for the discriminative models are derived from generative kernels. One advantage of using generative kernels is that systematic approaches exist how to introduce complex dependencies beyond conditional independence assumptions. Furthermore, by using generative kernels model-based compensation/adaptation tech- niques can be applied to make discriminative models robust to noise/speaker conditions. This paper extends previous work with combined generative/discriminative classifiers in several directions. First, it introduces derivative kernels based on context- dependent generative models. Second, it describes how derivative kernels can be incorporated in continuous discriminative models. Third, it addresses the issues associated with large number of classes and parameters when context-dependent models and high- dimensional features of derivative kernels are used. The approach is evaluated on two noise-corrupted tasks: small vocabulary AURORA 2 and medium-to-large vocabulary AURORA 4 task.

POSTSYNAPTIC ALPHA-2 RECEPTOR STIMULATION IMPROVES MEMORY IN AGED MONKEYS - INDIRECT EFFECTS OF YOHIMBINE VERSUS DIRECT EFFECTS OF CLONIDINE

Very low doses (0.00001 mg/kg) of the alpha-2 adrenergic antagonist, yohimbine, improved working memory performance in a subset of aged monkeys. Improvement appeared to result from increased norepinephrine (NE) release onto postsynaptic alpha-2 adrenoceptors, as the response was blocked by the ''postsynaptic'' alpha-2 antagonist, SKF104078. Cognitive-enhancing effects of low dose yohimbine treatment may depend on aged animals retaining an intact, endogenous NE system. In contrast to yohimbine, the alpha-2 agonist, clonidine, has improved working memory in air aged animals examined. In the present study, clonidine's beneficial effects were also blocked by the postsynaptic antagonists SKF104078 and SKF104856, suggesting that clonidine acts by directly stimulating postsynaptic alpha-2 adrenoceptors. Beneficial doses of clonidine (0.01 mg/kg) and yohimbine (0.00001 mg/kg) were combined to see if they would produce additive effects on memory enhancement. This strategy was successful in young monkeys with intact NE systems but was not effective in the aged monkeys. These findings demonstrate that drugs that indirectly stimulate postsynaptic alpha-2 receptors by increasing NE release are not as reliable in aged monkeys as directly acting agonists that can replace NE at postsynaptic alpha-2 receptors.

DOPAMINE D-1 RECEPTOR MECHANISMS IN THE COGNITIVE PERFORMANCE OF YOUNG-ADULT AND AGED MONKEYS

Dopamine (DA) D-1 receptor compounds were examined in monkeys for effects on the working memory functions of the prefrontal cortex and on the fine motor abilities of the primary motor cortex. The D-1 antagonist, SCH23390, the partial D-1 agonist, SKF38393, and the full D-1 agonist, dihydrexidine, were characterized in young control monkeys, and in aged monkeys with naturally occurring catecholamine depletion. In addition, SKF38393 was tested in young monkeys experimentally depleted of catecholamines with chronic reserpine treatment. Injections of SCH23390 significantly impaired the memory performance of young control monkeys, but did not impair aged monkeys with presumed catecholamine depletion. Conversely, the partial agonist, SKF38393, improved the depleted monkeys (aged or reserpine-treated) but did not improve young control animals. The full agonist, dihydrexidine, did improve memory performance in young control monkeys, as well as in a subset of aged monkeys. Consistent with D, receptor mechanisms, agonist-induced improvements were blocked by SCH23390. Drug effects on memory performance occurred independently of effects on fine motor performance. These results underscore the importance of DA D-1 mechanisms in cognitive function, and provide functional evidence of DA system degeneration in aged monkeys. Finally, high doses of D-1 agonists impaired memory performance in aged monkeys, suggesting that excessive D-1 stimulation may be deleterious to cognitive function.

DOPAMINE D2 RECEPTOR MECHANISMS CONTRIBUTE TO AGE-RELATED COGNITIVE DECLINE - THE EFFECTS OF QUINPIROLE ON MEMORY AND MOTOR-PERFORMANCE IN MONKEYS

The D2 dopamine (DA) receptor agonist, quinpirole, was characterized in young adult monkeys, young reserpine-treated monkeys and aged monkeys to assess the contribution of DA to age-related loss of prefrontal cortical (PFC) cognitive function, Monkeys were tested on a delayed response memory task that depends on the PFC, and a fine motor task that taps the functions of the motor cortex, In young adult monkeys, low quinpirole doses impaired performance of the PFC and fine motor tasks, while higher doses improved memory performance and induced dyskinesias and ''hallucinatory-like'' behaviors. The pattern of the quinpirole response in reserpine-treated monkeys suggested that the impairments in delayed response and fine motor performance resulted from drug actions at D2 autoreceptors, while the improvement in delayed response performance, dyskinesias and ''hallucinatory-like'' behaviors resulted from actions at postsynaptic receptors. In aged monkeys, low doses of quinpirole continued to impair fine motor performance, but lost their ability to impair delayed response performance. The magnitude of cognitive improvement and the incidence of ''hallucinatory-like'' behaviors were also reduced in the aged animals, suggesting some loss of postsynaptic D2 receptor function, The pattern of results is consistent with the greater loss of DA from the PFC than from motor areas in aged monkey brain (Goldman-Rakic and Brown, 1981; Wenk et al., 1989), and indicates that DA depletion contributes significantly to age-related cognitive decline.

Dose-dependent effects of the dopamine D1 receptor agonists A77636 or SKF81297 on spatial working memory in aged monkeys

With advancing age, monkeys develop deficits in spatial working memory resembling those induced by lesions of the prefrontal cortex (PFC). Aged monkeys also exhibit marked loss of dopamine from the PFC, a transmitter known to be important for proper PFC cognitive function. Previous results suggest that D1 agonist treatment can improve spatial working memory abilities in aged monkeys. However, this research was limited by the use of drugs with either partial agonist actions or significant D2 receptor actions. In our study, the selective dopamine D1 receptor full agonists A77636 and SKF81297 were examined in aged monkeys for effects on the working memory functions of the PFC. Both compounds produced a significant, dose-related effect on delayed response performance without evidence of side effects: low doses improved performance although higher doses impaired or had no effect on performance. Both the improvement and impairment in performance were reversed by pretreatment with the D1 receptor antagonist, SCH23390. These findings are consistent with previous results demonstrating that there is a narrow range of D1 receptor stimulation for optimal PFC cognitive function, and suggest that very low doses of D1 receptor agonists may have cognitive-enhancing actions in the elderly.

Molecular modeling on some human interleukins sharing gamma c of interleukin-2 receptor, and structure-function relationships

Five models for human interleukin-7 (HIL-7), HIL-9, HIL-13, HIL-15 and HIL-17 have been generated by SYBYL software package. The primary models were optimized using molecular dynamics and molecular mechanics methods. The final models were optimized using a steepest descent algorithm and a subsequent conjugate gradient method. The complexes with these interleukins and the common gamma chain of interleukin-2 receptor (IL-2R) were constructed and subjected to energy minimization. We found residues, such as Gln127 and Tyr103, of the common gamma chain of IL-2R are very important. Other residues, e.g. Lys70, Asn128 and Glu162, are also significant. Four hydrophobic grooves and two hydrophilic sites converge at the active site triad of the gamma chain. The binding sites of these interleukins interaction with the common gamma chain exist in the first helical and/or the fourth helical domains. Therefore, we conclude that these interleukins binds to the common gamma chain of IL-2R by the first and the fourth helix domain. Especially at the binding sites of some residues (lysine, arginine, asparagine, glutamic acid and aspartic acid), with a discontinuous region of the common gamma chain of IL-2R, termed the interleukins binding sites (103-210). The study of these sites can be important for the development of new drugs. (C) 2000 Elsevier Science B.V. All rights reserved.

Interaction between human interleukin-16 and CD4 receptor of HIV-1

AIM: To study the interaction between human interleukin-16 (IL-16) and the receptor CD4 (T-lymphocyte differentiation antigen) of human immunodeficiency virus type 1 (HIV-1). METHODS: Two structurally con served regions (SCRs) of human IL-16 were built by the SYBYL/Biopolymer module using the corresponding transmembrane (TM) domain of human interleukin-1 (HIL-4) and HIL-2 as the templates. The coordinates for amino-terminal residue sequence, carboxyl-terminal residue sequences, and cytoplasm loops were generated using Biopolymer's LOOP SEARCH algorithm. RESULTS: HIL-16 first formed a homodimer, then contacted with CD4 dimer further forming a dimeric complex. Subsequently, the dimeric complex constructed the tetrameric complex by two disulfide bridges between the cysteines of HIL-16 (Cys31-Cys31). CONCLUSION: The interaction model is useful to propose the action mechanism of HIL-16 and is beneficial for rational designing of novel anti-HIV drugs.

Expression and regulation of orphan receptor TR2 mRNA in germ cells of cryptorchid testis in rat and rhesus monkey

Expression and cellular localization of orphan receptor TR2 mRNA in relation to germ cell apoptosis in cryptorchid testes of rat and rhesus monkey have been studied by using in situ hybridization and in situ 3'-end labeling of DNA fragments (TUNEL). The results show that: (i) TR2 mRNA is specifically expressed in the germ cells, mainly in the spermatocytes, round and elongated spermatids. The expression level of TR2 mRNA varies with the seminiferous cycle, (ii) In the rat cryptorchid testes on days 3 and 5 after the surgery, the germ cells began to undergo apoptosis with no evident decrease in TR2 mRNA level. On day 7.5, however, most germ cells underwent apoptosis, while the expression level of TR2 mRNA declined markedly, and TR2 mRNA was rarely expressed on day 10 thereafter. (iii) On days 15 and 20 of the cryptorchid testes of rhesus monkey, TR2 mRNA was only expressed in a few of primary spermatocytes and the mRNA was almost undetectable on days 30, 45, 60. These results suggest that TR2 mRNA probably plays an important role in spermatogenesis and germ cell apoptosis.

Structural and functional characterization of the human CCR5 receptor in complex with HIV gp120 envelope glycoprotein and CD4 receptor by molecular modeling studies

The entry of human immunodeficiency virus (HIV) into cells depends on a sequential interaction of the gp120 envelope glycoprotein with the cellular receptors CD4 and members of the chemokine receptor family. The CC chemokine receptor CCR5 is such a receptor for several chemokines and a major coreceptor for the entry of R5 HIV type-1 (HIV-1) into cells. Although many studies focus on the interaction of CCR5 with HIV-1, the corresponding interaction sites in CCR5 and gp120 have not been matched. Here we used an approach combining protein structure modeling, docking and molecular dynamics simulation to build a series of structural models of the CCR5 in complexes with gp120 and CD4. Interactions such as hydrogen bonds, salt bridges and van der Waals contacts between CCR5 and gp120 were investigated. Three snapshots of CCR5-gp120-CD4 models revealed that the initial interactions of CCR5 with gp120 are involved in the negatively charged N-terminus (Nt) region of CCR5 and positively charged bridging sheet region of gp120. Further interactions occurred between extracellular loop2 (ECL2) of CCR5 and the base of V3 loop regions of gp120. These interactions may induce the conformational changes in gp120 and lead to the final entry of HIV into the cell. These results not only strongly support the two-step gp120-CCR5 binding mechanism, but also rationalize extensive biological data about the role of CCR5 in HIV-1 gp120 binding and entry, and may guide efforts to design novel inhibitors.

A variational perspective on noise-robust speech recognition

Model compensation methods for noise-robust speech recognition have shown good performance. Predictive linear transformations can approximate these methods to balance computational complexity and compensation accuracy. This paper examines both of these approaches from a variational perspective. Using a matched-pair approximation at the component level yields a number of standard forms of model compensation and predictive linear transformations. However, a tighter bound can be obtained by using variational approximations at the state level. Both model-based and predictive linear transform schemes can be implemented in this framework. Preliminary results show that the tighter bound obtained from the state-level variational approach can yield improved performance over standard schemes. © 2011 IEEE.

Derivative kernels for noise robust ASR

Recently there has been interest in combining generative and discriminative classifiers. In these classifiers features for the discriminative models are derived from the generative kernels. One advantage of using generative kernels is that systematic approaches exist to introduce complex dependencies into the feature-space. Furthermore, as the features are based on generative models standard model-based compensation and adaptation techniques can be applied to make discriminative models robust to noise and speaker conditions. This paper extends previous work in this framework in several directions. First, it introduces derivative kernels based on context-dependent generative models. Second, it describes how derivative kernels can be incorporated in structured discriminative models. Third, it addresses the issues associated with large number of classes and parameters when context-dependent models and high-dimensional feature-spaces of derivative kernels are used. The approach is evaluated on two noise-corrupted tasks: small vocabulary AURORA 2 and medium-to-large vocabulary AURORA 4 task. © 2011 IEEE.