Variable Sparsity Kernel Learning

This paper(1) presents novel algorithms and applications for a particular class of mixed-norm regularization based Multiple Kernel Learning (MKL) formulations. The formulations assume that the given kernels are grouped and employ l(1) norm regularization for promoting sparsity within RKHS norms of each group and l(s), s >= 2 norm regularization for promoting non-sparse combinations across groups. Various sparsity levels in combining the kernels can be achieved by varying the grouping of kernels-hence we name the formulations as Variable Sparsity Kernel Learning (VSKL) formulations. While previous attempts have a non-convex formulation, here we present a convex formulation which admits efficient Mirror-Descent (MD) based solving techniques. The proposed MD based algorithm optimizes over product of simplices and has a computational complexity of O (m(2)n(tot) log n(max)/epsilon(2)) where m is no. training data points, n(max), n(tot) are the maximum no. kernels in any group, total no. kernels respectively and epsilon is the error in approximating the objective. A detailed proof of convergence of the algorithm is also presented. Experimental results show that the VSKL formulations are well-suited for multi-modal learning tasks like object categorization. Results also show that the MD based algorithm outperforms state-of-the-art MKL solvers in terms of computational efficiency.

Chain folding and A:T pairing in human telomeric DNA: a model-building and molecular dynamics study

The various types of chain folding and possible intraloop as well as interloop base pairing in human telomeric DNA containing d(TTAG(3)) repeats have been investigated by model-building, molecular mechanics, and molecular dynamics techniques. Model-building and molecular mechanics studies indicate that it is possible to build a variety of energetically favorable folded-back structures with the two TTA loops on same side and the 5' end thymines in the two loops forming TATA tetrads involving a number of different intraloop as well as interloop A:T pairing schemes. In these folded-back structures, although both intraloop and interloop Watson-Crick pairing is feasible, no structure is possible with interloop Hoogsteen pairing. MD studies of representative structures indicate that the guanine-tetraplex stem is very rigid and, while the loop regions are relatively much more flexible, most of the hydrogen bonds remain intact throughout the 350-ps in vacuo simulation. The various possible TTA loop structures, although they are energetically similar, have characteristic inter proton distances, which could give rise to unique cross-peaks in two-dimensional nuclear Overhauser effect spectroscopy (NOESY) experiments. These folded-back structures with A:T pairings in the loop region help in rationalizing the data from chemical probing and other biochemical studies on human telomeric DNA.

Molecular dynamics studies on nucleoside 2',3'-cyclic phosphates

2',3'-cyclic nucleotides are intermediates and substrates of Ribonuclease (RNase)-catalysed reactions. The characterization of the equilibrium conformation as well as the flexibility inherent in these molecules helps in understanding the enzymatic action of RNases. The present study explores parameters like phase angle, glycosydic torsion angle and hydrogen bond to find possible interrelationship between them through Molecular Dynamics (MD) simulations on 3'-GMP, 3'-UMP, A>p, G>p, U>p, C>p, GpA>p and UpA>p. Interesting results of the effect of cyclisation and other constraints such as hydrogen bond between certain groups on the equilibrium ribose conformation have emerged from this study.

Diffusion Anomaly in Silicalite and VPI-5 from Molecular Dynamics Simulations

Molecular dynamics (MD) simulations on rigid and flexible framework models of silicalite and a rigid framework model of the aluminophosphate VPI-5 for different sorbate diameters are reported. The sorbate-host interactions are modeled in terms of simple atom-atom Lennard-Jones interactions. The results suggest that the diffusion coefficient exhibits an anomaly as gamma approaches unity. The MD results confirm the existence of a linear regime for sorbate diameters significantly smaller than the channel diameter and an anomalous regime observed for sorbate diameters comparable to the channel diameter. The power spectra obtained by Fourier transformation of the velocity autocorrelation function indicate that there is an increase in the intensity of the low-frequency component for the velocity component parallel to the direction of motion for the sorbate diameter in the anomalous regime. The present results suggest that the diffusion anomaly is observed irrespective of (1) the geometry and topology of the pore structure and (2) the nature of the host material. The results are compared with the work of Derouane and co-workers, who have suggested the existence of ''floating molecules'' on the basis of earlier theoretical and computational approaches.

Effect of Strain Rate on Ductile-to-Brittle Transition Temperature of a Free-Standing Pt-Aluminide Bond Coat

The ductile-to-brittle transition temperature (DBTT) of a free-standing Pt-aluminide (PtAl) bondcoat was determined using the microtensile testing method and the effect of strain rate variation, in the range 10(-5) to 10(-1) s(-1), on the DBTT studied. The DBTT increased appreciably with the increase in strain rate. The activation energy determined for brittle-to-ductile transition, suggested that such transition is most likely associated with vacancy diffusion. Climb of aOE (c) 100 > dislocations observed in analysis of dislocation structure using a transmission electron microscope (TEM) supported the preceding mechanism.

Modeling of angiogenin-3 '-NMP complex

Angiogenin belongs to the Ribonuclease superfamily and has a weak enzymatic activity that is crucial for its biological function of stimulating blood vessel growth. Structural studies on ligand bound Angiogenin will go a long way in understanding the mechanism of the protein as well as help in designing drugs against it. In this study we present the first available structure of nucleotide ligand bound Angiogenin obtained by computer modeling. The importance of this study in itself notwithstanding, is a precursor to modeling a full dinucleotide substrate onto Angiogenin. Bovine Angiogenin, the structure of which has been solved at a high resolution, was earlier subjected to Molecular Dynamics simulations for a nanosecond. The MD structures offer better starting points for docking as they offer lesser obstruction than the crystal structure to ligand binding. The MD structure with the least serious short contacts was modeled to obtain a steric free Angiogenin - 3' mononucleotide complex structure. The structures were energetically minimized and subjected to a brief spell of Molecular Dynamics. The results of the simulation show that all the li,ligand-Angiogenin interactions and hydrogen bonds are retained, redeeming the structure and docking procedure. Further, following ligand - protein interactions in the case of the ligands 3'-CMP and 3'-UMP we were able to speculate on how Angiogenin, a predominantly prymidine specific ribonuclease prefers Cytosine to Uracil in the first base position.

Absolute Entropy and Energy of Carbon Dioxide Using the Two-Phase Thermodynamic Model

The two-phase thermodynamic (2PT) model is used to determine the absolute entropy and energy of carbon dioxide over a wide range of conditions from molecular dynamics trajectories. The 2PT method determines the thermodynamic properties by applying the proper statistical mechanical partition function to the normal modes of a fluid. The vibrational density of state (DoS), obtained from the Fourier transform of the velocity autocorrelation function, converges quickly, allowing the free energy, entropy, and other thermodynamic properties to be determined from short 20-ps MD trajectories. The anharmonic effects in the vibrations are accounted for by the broadening of the normal modes into bands from sampling the velocities over the trajectory. The low frequency diffusive modes, which lead to finite DoS at zero frequency, are accounted for by considering the DoS as a superposition of gas-phase and solid-phase components (two phases). The analytical decomposition of the DoS allows for an evaluation of properties contributed by different types of molecular motions. We show that this 2PT analysis leads to accurate predictions of entropy and energy of CO2 over a wide range of conditions (from the triple point to the critical point of both the vapor and the liquid phases along the saturation line). This allows the equation of state of CO2 to be determined, which is limited only by the accuracy of the force field. We also validated that the 2PT entropy agrees with that determined from thermodynamic integration, but 2PT requires only a fraction of the time. A complication for CO2 is that its equilibrium configuration is linear, which would have only two rotational modes, but during the dynamics it is never exactly linear, so that there is a third mode from rotational about the axis. In this work, we show how to treat such linear molecules in the 2PT framework.

Structures of the catalytic site mutants D99A and H48Q and the calcium-loop mutant D49E of phospholipase A(2)

Crystal structures of the active-site mutants D99A and H48Q and the calcium-loop mutant D49E of bovine phospholipase A(2) have been determined at around 1.9 Angstrom resolution. The D99A mutant is isomorphous to the orthorhombic recombinant enzyme, space group P2(1)2(1)2(1), The H48Q and the calcium-loop mutant D49E are isomorphous to the trigonal recombinant enzyme, space group P3(1)21, The two active-site mutants show no major structural perturbations. The structural water is absent in D99A and, therefore, the hydrogen-bonding scheme is changed. In H48Q, the catalytic water is present and hydrogen bonded to Gln48 N, but the second water found in native His48 is absent. In the calcium-loop mutant D49E, the two water molecules forming the pentagonal bipyramid around calcium are absent and only one O atom of the Glu49 carboxylate group is coordinated to calcium, resulting in only four ligands.

The molecular dynamics study of lithium ion conduction in phosphate glasses and the role of non-bridging oxygen

Molecular dynamics (MD) simulation of lithium phosphate (Li2O-P2O5) glasses with varying Li2O content has been carried out. Two different P-O distances corresponding to phosphorus coordination with bridging oxygen (BO) and non-bridging oxygen (NBO) were identified in the simulated glasses. NBO-BO interconversion or bond switching was noted, which results in a dynamic equilibration of the tetrahedral phosphate units (P-n, n = 1,3 indicates the number of bridging oxygen atoms in the coordination of phosphorus). The NBO-BO bond switching is mildly activated with an effective activation barrier of 0.03-0.05 eV. Lithium ion jumps do not appear to be strongly coupled to bond switching. But the number of Li+ ions coordinated to an optimum number of NBOs and the number of Li+ ions jumping out of their sites appear to be correlated. Detailed analysis was made of the dynamics of P-n species and new insights have been obtained regarding ion migration in network-modified phosphate glasses.

Thermal expansion of LaBa2Cu2CoO7+delta

In this paper, we report high-temperature X-ray diffraction (HTXRD) and dilatometric studies on LaBa2Cu2CoO7+delta. Bulk and volume thermal expansion studies, along with a study of its phase transition, were carried out. The linear and volume thermal expansion coefficients were found to be 11.7 X 10(-6) K-1 and 42.3 X 10(-6) K-1, respectively. (C) 2000 Elsevier Science Ltd.

Comparison of the ultrafast to slow time scale dynamics of three liquid crystals in the isotropic phase

The dynamics of three liquid crystals, 4'(pentyloxy)-4-biphenylcarbonitrile (5-OCB), 4'-pentyl-4-biphenylcarbonitrile (5-CB), and 1-isothiocyanato-(4-propylcyclohexyl)benzene (3-CHBT), are investigated from very short time (similar to1 ps) to very long time (>100 ns) as a function of temperature using optical heterodyne detected optical Kerr effect experiments. For all three liquid crystals, the data decay exponentially only on the longest time scale (> several ns). The temperature dependence of the long time scale exponential decays is described well by the Landau-de Gennes theory of the randomization of pseudonematic domains that exist in the isotropic phase of liquid crystals near the isotropic to nematic phase transition. At short time, all three liquid crystals display power law decays. Over the full range of times, the data for all three liquid crystals are fit with a model function that contains a short time power law. The power law exponents for the three liquid crystals range between 0.63 and 0.76, and the power law exponents are temperature independent over a wide range of temperatures. Integration of the fitting function gives the empirical polarizability-polarizability (orientational) correlation function. A preliminary theoretical treatment of collective motions yields a correlation function that indicates that the data can decay as a power law at short times. The power law component of the decay reflects intradomain dynamics. (C) 2002 American Institute of Physics.

Efficient algorithms for learning kernels from multiple similarity matrices with general convex loss functions

In this paper we consider the problem of learning an n × n kernel matrix from m(1) similarity matrices under general convex loss. Past research have extensively studied the m = 1 case and have derived several algorithms which require sophisticated techniques like ACCP, SOCP, etc. The existing algorithms do not apply if one uses arbitrary losses and often can not handle m > 1 case. We present several provably convergent iterative algorithms, where each iteration requires either an SVM or a Multiple Kernel Learning (MKL) solver for m > 1 case. One of the major contributions of the paper is to extend the well knownMirror Descent(MD) framework to handle Cartesian product of psd matrices. This novel extension leads to an algorithm, called EMKL, which solves the problem in O(m2 log n 2) iterations; in each iteration one solves an MKL involving m kernels and m eigen-decomposition of n × n matrices. By suitably defining a restriction on the objective function, a faster version of EMKL is proposed, called REKL,which avoids the eigen-decomposition. An alternative to both EMKL and REKL is also suggested which requires only an SVMsolver. Experimental results on real world protein data set involving several similarity matrices illustrate the efficacy of the proposed algorithms.

An infrared spectroscopic study of the low-spin to intermediate-spin state (1A1–3T1) transition in rare earth cobaltates, LnCoO3 (Ln=La, Pr and Nd)

Low-spin (LS) to intermediate-spin (IS) state transitions in crystals of LnCoO3 (Ln=La, Pr and Nd) have been investigated by variable temperature infrared spectroscopy. The spectra reveal the occurrence of the transition around 120, 220 and 275 K, respectively, in LaCoO3,PrCoO3 and NdCoO3, at which temperatures the intensities of the stretching and the bending modes associated with the LS state decrease, accompanied by an increase in the intensities of the bands due to IS state. The characteristic frequencies of both the spin states decrease with increase in temperature, showing anomalies around the transition.

A molecular dynamics study based post facto free energy analysis of the binding of bovine angiogenin with UMP and CMP ligands

Angiogenin is a protein belonging to the superfamily of RNase A. The RNase activity of this protein is essential for its angiogenic activity. Although members of the RNase A family carry out RNase activity, they differ markedly in their strength and specificity. In this paper, we address the problem of higher specificity of angiogenin towards cytosine against uracil in the first base binding position. We have carried out extensive nano-second level molecular dynamics(MD) computer simulations on the native bovine angiogenin and on the CMP and UMP complexes of this protein in aqueous medium with explicit molecular solvent. The structures thus generated were subjected to a rigorous free energy component analysis to arrive at a plausible molecular thermodynamic explanation for the substrate specificity of angiogenin.