Daniel Svozil

Introduction to multi-layer feed-forward neural networks

Abstract Basic definitions concerning the multi-layer feed-forward neural networks are given. The back-propagation training algorithm is explained. Partial derivatives of the objective function with respect to the weight and threshold coefficients are derived. These derivatives are valuable for an adaptation process of the considered neural network. Training and generalisation of multi-layer feed-forward neural networks are discussed. Improvements of the standard back-propagation algorithm are reviewed. Example of the use of multi-layer feed-forward neural networks for prediction of carbon-13 NMR chemical shifts of alkanes is given. Further applications of neural networks in chemistry are reviewed. Advantages and disadvantages of multilayer feed-forward neural networks are discussed.

DNA conformations and their sequence preferences

The geometry of the phosphodiester backbone was analyzed for 7739 dinucleotides from 447 selected crystal structures of naked and complexed DNA. Ten torsion angles of a near-dinucleotide unit have been studied by combining Fourier averaging and clustering. Besides the known variants of the A-, B- and Z-DNA forms, we have also identified combined A + B backbone-deformed conformers, e.g. with α/γ switches, and a few conformers with a syn orientation of bases occurring e.g. in G-quadruplex structures. A plethora of A- and B-like conformers show a close relationship between the A- and B-form double helices. A comparison of the populations of the conformers occurring in naked and complexed DNA has revealed a significant broadening of the DNA conformational space in the complexes, but the conformers still remain within the limits defined by the A- and B- forms. Possible sequence preferences, important for sequence-dependent recognition, have been assessed for the main A and B conformers by means of statistical goodness-of-fit tests. The structural properties of the backbone in quadruplexes, junctions and histone-core particles are discussed in further detail.

Comparison of Intrinsic Stacking Energies of Ten Unique Dinucleotide Steps in A-RNA and B-DNA Duplexes. Can We Determine Correct Order of Stability by Quantum-Chemical Calculations?

High level ab initio methods have been used to study stacking interactions in ten unique base pair steps both in A-RNA and in B-DNA duplexes. The protocol for selection of geometries based on molecular dynamics (MD) simulations is proposed, and its suitability is demonstrated by comparison with stacking in steps at fiber diffraction geometries. It is shown that fiber diffraction geometries are not sufficiently accurate for interaction energy calculations. In addition, the protocol for selection of geometries based on MD simulations allows for the evaluation of the variability of the intrinsic stacking energies along the MD trajectories. The uncertainty in stacking energies (difference between the most and least stable geometry) due to the dynamical nature of systems can be, in some cases, as large as 3.0 kcal x mol(-1), which is almost 50% of the actual sequence dependence of base stacking energies (the energy difference between the most and least stable sequences). Thus, assessing the relative magnitude of the gas phase stacking energy using a single geometry for each sequence is insufficient to obtain an unambiguous order of gas phase stacking energies in canonical double helices. Though the ordering of ten unique dinucleotide steps cannot be definitive, some general conclusions were drawn. The stacking energies of base pair steps in A-RNA are more evenly separated compared to B-DNA, and their ordering is less sensitive to the dynamics of the system compared to be B-DNA. The most stable step both in B-DNA and A-RNA is the GC/GC [corrected] step that is well separated from the second most stable step CG/CG. [corrected] Also the least stable step (the CC/GG step) is well separated from the rest of the structures. The calculations further show that B-DNA stacking is favorable only marginally (on average by 1.14 kcal x mol(-1) per base pair step) over A-RNA stacking, and this difference vanishes after subtracting the stabilizing van der Waals effect of the thymine 5-methyl group that is absent in RNA. Basically, no correlation between the sequence dependence of gas phase stacking energies and the sequence dependence of DeltaG degrees(37) free energies used in nearest-neighbor models was found either for B-DNA or for A-RNA. This reflects the complexity of the balance of forces that are responsible for the sequence dependence of thermodynamics stability of nucleic acids, which masks the effect of the intrinsic interactions between the stacked base pairs.

The DNA and RNA sugar–phosphate backbone emerges as the key player. An overview of quantum-chemical, structural biology and simulation studies

Knowledge of geometrical and physico-chemical properties of the sugar-phosphate backbone substantially contributes to the comprehension of the structural dynamics, function and evolution of nucleic acids. We provide a side by side overview of structural biology/bioinformatics, quantum chemical and molecular mechanical/simulation studies of the nucleic acids backbone. We highlight main features, advantages and limitations of these techniques, with a special emphasis given to their synergy. The present status of the research is then illustrated by selected examples which include classification of DNA and RNA backbone families, benchmark structure-energy quantum chemical calculations, parameterization of the dihedral space of simulation force fields, incorporation of arsenate into DNA, sugar-phosphate backbone self-cleavage in small RNA enzymes, and intricate geometries of the backbone in recurrent RNA building blocks. Although not apparent from the current literature showing limited overlaps between the QM, simulation and bioinformatics studies of the nucleic acids backbone, there in fact should be a major cooperative interaction between these three approaches in studies of the sugar-phosphate backbone.

NEURAL NETWORK PREDICTION OF CARBON-13 NMR CHEMICAL SHIFTS OF ALKANES

Three-layer feed-forward neural networks for the prediction of I3C NMR chemical shifts of alkanes through nine carbon atoms are used. Carbon atoms in alkanes are determined by 13 descriptors that correspond to the so-called embedding frequencies of rooted subtrees. These descriptors are equal to numbers of appearance of smaller structural skeletons composed of two through five carbon atoms. It is demonstrated that the used descriptors offer a very useful formal tool for the proper and adequate description of environment of carbon atoms in alkanes. Neural networks with different numbers of hidden neurons have been examined. Best results are given by the neural network composed of three hidden neurons. Simultaneous calculations carried out by the standard linear regression analysis are compared with our neural network calculations.

Can We Accurately Describe the Structure of Adenine Tracts in B-DNA? Reference Quantum-Chemical Computations Reveal Overstabilization of Stacking by Molecular Mechanics.

Sequence-dependent local variations of helical parameters, structure, and flexibility are crucial for molecular recognition processes involving B-DNA. A-tracts, i.e., stretches of several consecutive adenines in one strand that are in phase with the DNA helical repeat, mediate significant DNA bending. During the past few decades, there have been intense efforts to understand the sequence dependence of helical parameters in DNA. Molecular dynamics (MD) simulations can provide valuable insights into the molecular mechanism behind the relationship between sequence and structure. However, although recent improvements in empirical force fields have helped to capture many sequence-dependent B-DNA properties, several problems remain, such as underestimation of the helical twist and suspected underestimation of the propeller twist in A-tracts. Here, we employ reference quantum mechanical (QM) calculations, explicit solvent MD, and bioinformatics to analyze the underestimation of propeller twisting of A-tracts in simulations. Although we did not identify a straightforward explanation, we discovered two imbalances in the empirical force fields. The first was overestimation of stacking interactions accompanied by underestimation of base-pairing energy, which we attribute to anisotropic polarizabilities that are not reflected by the isotropic force fields. This may lead to overstacking with potentially important consequences for MD simulations of nucleic acids. The second observed imbalance was steric clash between A(N1) and T(N3) nitrogens of AT base pairs in force-field descriptions, resulting in overestimation of the AT pair stretch in MD simulations. We also substantially extend the available set of benchmark estimated CCSD(T)/CBS data for B-DNA base stacking and provide a code that allows the generation of diverse base-stacking geometries suitable for QM computations with predefined intra- and interbase pair parameters.

Electric resistance in a Nafion® membrane exposed to air after a step change in the relative humidity

Solid polymer electrolytes in the form of membranes are often used in industrial electrolysers, fuel cells and solid-state chemical sensors. Perfluorosulphonic membranes known under the name Nation ® (DuPont) are the most common. The physico-chemical properties of solid polymer electrolytes are strongly dependent on the amount of water contained within the membrane [1]. The relationship between the water content in the membrane and the electric resistance or conductivity of the membrane is usually obtained from a measurement of the resistance of a membrane sample containing a defined amount of water, under conditions ensuring that Nafion ® does not exchange water with the environment for the time of measurement. The resultant membrane resistance values then correspond to certain equilibrium water contents [2,3]. The present paper follows the changes in the resistance of a Nation ® membrane during the process in which one equilibrium water content changes into another owing to exchange of water between Nafion ® and ambient air of a certain relative humidity (RH). The air R H is changed stepwise and the time dependence of the Nation ® resistance is monitored. The conductivity electrodes formed directly on the Nation ® membrane surface make possible continuous resistance monitoring.

Balance of Attraction and Repulsion in Nucleic-Acid Base Stacking: CCSD(T)/Complete-Basis-Set-Limit Calculations on Uracil Dimer and a Comparison with the Force-Field Description

We have carried out reference quantum-chemical calculations for about 100 geometries of the uracil dimer in stacked conformations. The calculations have been specifically aimed at geometries with unoptimized distances between the monomers including geometries with mutually tilted monomers. Such geometries are characterized by a delicate balance between local steric clashes and local unstacking and had until now not been investigated using reference quantum-mechanics (QM) methods. Nonparallel stacking geometries often occur in nucleic acids and are of decisive importance, for example, for local conformational variations in B-DNA. Errors in the short-range repulsion region would have a major impact on potential energy scans which were often used in the past to investigate local geometry variations in DNA. An incorrect description of such geometries may also partially affect molecular dynamics (MD) simulations in applications when quantitative accuracy is required. The reference QM calculations have been carried out using the MP2 method extrapolated to the complete basis-set limit and corrected for higher-order electron-correlation contributions using CCSD(T) calculations with a medium-sized basis set. These reference calculations have been used as benchmark data to test the performance of the DFT-D, SCS(MI)-MP2, and DFT-SAPT QM methods and of the AMBER molecular-mechanics (MM) force field. The QM methods show close to quantitative agreement with the reference data, albeit the DFT-D method tends to modestly exaggerate the repulsion of steric clashes. The force field in general also provides a good description of base stacking for the systems studied here. However, for geometries with close interatomic contacts and clashes, the repulsion effects are rather severely exaggerated. The discrepancy reported here should not affect the overall stability of MD simulations and qualitative applications of the force field. However, it may affect the description of subtle quantitative effects such as the local conformational variations in B-DNA. Preliminary calculations for two H-bonded uracil base pairs, including one with a C-H···O H-bond, indicate excellent performance of the tested QM methods for all intermonomer distances. The force field, on the other hand, is less satisfactory, especially in the repulsive regions.

Benchmark quantum-chemical calculations on a complete set of rotameric families of the DNA sugar–phosphate backbone and their comparison with modern density functional theory

The DNA sugar-phosphate backbone has a substantial influence on the DNA structural dynamics. Structural biology and bioinformatics studies revealed that the DNA backbone in experimental structures samples a wide range of distinct conformational substates, known as rotameric DNA backbone conformational families. Their correct description is essential for methods used to model nucleic acids and is known to be the Achilles heel of force field computations. In this study we report the benchmark database of MP2 calculations extrapolated to the complete basis set of atomic orbitals with aug-cc-pVTZ and aug-cc-pVQZ basis sets, MP2(T,Q), augmented by ΔCCSD(T)/aug-cc-pVDZ corrections. The calculations are performed in the gas phase as well as using a COSMO solvent model. This study includes a complete set of 18 established and biochemically most important families of DNA backbone conformations and several other salient conformations that we identified in experimental structures. We utilize an electronically sufficiently complete DNA sugar-phosphate-sugar (SPS) backbone model system truncated to prevent undesired intramolecular interactions. The calculations are then compared with other QM methods. The BLYP and TPSS functionals supplemented with Grimmes D3(BJ) dispersion term provide the best tradeoff between computational demands and accuracy and can be recommended for preliminary conformational searches as well as calculations on large model systems. Among the tested methods, the best agreement with the benchmark database has been obtained for the double-hybrid DSD-BLYP functional in combination with a quadruple-ζ basis set, which is, however, computationally very demanding. The new hybrid density functionals PW6B95-D3 and MPW1B95-D3 yield outstanding results and even slightly outperform the computationally more demanding PWPB95 double-hybrid functional. B3LYP-D3 is somewhat less accurate compared to the other hybrids. Extrapolated MP2(D,T) calculations are not as accurate as the less demanding DFT-D3 methods. Preliminary force field tests using several charge sets reveal an almost order of magnitude larger deviations from the reference QM data compared to modern DFT-D3, underlining the challenges facing force field simulations of nucleic acids. As expected, inclusion of the solvent environment approximated by a continuum approach has a large impact on the relative stabilities of different backbone substates and is important when comparing the QM data with structural bioinformatics and other experimental data.

InCHlib – interactive cluster heatmap for web applications

BackgroundHierarchical clustering is an exploratory data analysis method that reveals the groups (clusters) of similar objects. The result of the hierarchical clustering is a tree structure called dendrogram that shows the arrangement of individual clusters. To investigate the row/column hierarchical cluster structure of a data matrix, a visualization tool called ‘cluster heatmap’ is commonly employed. In the cluster heatmap, the data matrix is displayed as a heatmap, a 2-dimensional array in which the colour of each element corresponds to its value. The rows/columns of the matrix are ordered such that similar rows/columns are near each other. The ordering is given by the dendrogram which is displayed on the side of the heatmap.ResultsWe developed InCHlib (Interactive Cluster Heatmap Library), a highly interactive and lightweight JavaScript library for cluster heatmap visualization and exploration. InCHlib enables the user to select individual or clustered heatmap rows, to zoom in and out of clusters or to flexibly modify heatmap appearance. The cluster heatmap can be augmented with additional metadata displayed in a different colour scale. In addition, to further enhance the visualization, the cluster heatmap can be interconnected with external data sources or analysis tools. Data clustering and the preparation of the input file for InCHlib is facilitated by the Python utility script inchlib_clust.ConclusionsThe cluster heatmap is one of the most popular visualizations of large chemical and biomedical data sets originating, e.g., in high-throughput screening, genomics or transcriptomics experiments. The presented JavaScript library InCHlib is a client-side solution for cluster heatmap exploration. InCHlib can be easily deployed into any modern web application and configured to cooperate with external tools and data sources. Though InCHlib is primarily intended for the analysis of chemical or biological data, it is a versatile tool which application domain is not limited to the life sciences only.