A random-sampling high dimensional model representation neural network for building potential energy surfaces

We combine the high dimensional model representation (HDMR) idea of Rabitz and co-workers [J. Phys. Chem. 110, 2474 (2006)] with neural network (NN) fits to obtain an effective means of building multidimensional potentials. We verify that it is possible to determine an accurate many-dimensional potential by doing low dimensional fits. The final potential is a sum of terms each of which depends on a subset of the coordinates. This form facilitates quantum dynamics calculations. We use NNs to represent HDMR component functions that minimize error mode term by mode term. This NN procedure makes it possible to construct high-order component functions which in turn enable us to determine a good potential. It is shown that the number of available potential points determines the order of the HDMR which should be used.     

Localized bond model for molecular energy to fourth order in perturbation theory

The localized bond model of Malrieu, Diner, and Claverie is extended to fourth order in perturbation theory. Single, double, triple, and quadruple replacements from the doubly occupied bonding reference function are included utilizing a symmetric form of diagrammatic perturbation theory. The fourth order theory derived executes on a computer as quickly as does the third order theory. Results are examined utilizing the Pariser–Parr–Pople and CNDO/2 model Hamiltonians, and are compared with third order results and with either exact results where they are known, or with a configuration interaction of all singles and doubles. The influence of the initial hybridization, localization, and bond polarization is discussed. In general, the fourth order corrections are of comparable size to third order. Improvement in results appears to be marginal in the Nesbet–Epstein scheme in passing to fourth order because of the oscillating nature of the series; for Moller–Plesset theory errors are approximately halved. The relative energies as a function of modest geometry change about minima is about the same at third order as it is at fourth for most cases examined.     

GlyNet: a multi-task neural network for predicting protein–glycan interactions

Advances in diagnostics, therapeutics, vaccines, transfusion, and organ transplantation build on a fundamental understanding of glycan–protein interactions. To aid this, we developed GlyNet, a model that accurately predicts interactions (relative binding strengths) between mammalian glycans and 352 glycan-binding proteins, many at multiple concentrations. For each glycan input, our model produces 1257 outputs, each representing the relative interaction strength between the input glycan and a particular protein sample. GlyNet learns these continuous values using relative fluorescence units (RFUs) measured on 599 glycans in the Consortium for Functional Glycomics glycan arrays and extrapolates these to RFUs from additional, untested glycans. GlyNet''s output of continuous values provides more detailed results than the standard binary classification models. After incorporating a simple threshold to transform such continuous outputs the resulting GlyNet classifier outperforms those standard classifiers. GlyNet is the first multi-output regression model for predicting protein–glycan interactions and serves as an important benchmark, facilitating development of quantitative computational glycobiology.

GlyNet, a neural net model of glycan-protein binding strengths. Given a glycan it outputs binding to each of several protein samples. Reproducing glycan array data, it extrapolates the binding of untested glycans against the protein samples.     

分子轨道成键性质及原子间化学键强度的成键能判据

本文定义了成键能Eb并用作分子轨道成键性质和分子中原子间化学键强度的判据。与Mulliken重叠布居Pb不同, 在成键能Eb中同时包含了原子轨道间的重叠因素和原子轨道的能量因素。对一些分子所作计算结果表明, 成键能判据较Mulliken重叠布居判据所得结论与实验更相符。     

Estimation of intramolecular hydrogen bond energy via molecular tailoring approach

A novel method, based on the molecular tailoring approach for estimating intramolecular hydrogen bond energies, is proposed. Here, as a case study, the O-H...O bond energy is directly estimated by addition/subtraction of the single point individual fragment energies. This method is tested on polyhydroxy molecules at MP2 and B3LYP levels of theory. It is seen to be able to distinguish between weak ( approximately 1 kcal mol(-1)) and moderately strong ( approximately 5 kcal mol(-1)) hydrogen bonds in polyhydroxy molecules.     

Calculation of molecular energy indices by the bond interaction method

Conclusions Parameters were determined for an interacting bond model for hydrocarbons and oxygen- and sulfur-containing organic molecules. The parameters obtained were used to calculate the homolytic bond dissociation energies and molecular formation energies.Translated from Izvestiya Akademii Nauk SSSR, Seriya Khimicheskaya, No. 7, pp. 1680–1683, July, 1988.     

Design and training of a neural network for predicting the solvent accessibility of proteins

A feed-forward neural network has been developed to predict the solvent accessibility/accessible surface area (ASA) of proteins using improved design and training methods. Several network issues ranging from the coding of ASA states to the problem of local minima of learning curve, have been addressed. Successful new approaches to overcome these problems are presented. Set of trained network weights for each ASA threshold is provided. It has been established that the prediction accuracy results with neural network are better than other reported results of ASA prediction, despite a high test to training data ratio.     

预测烷烃密度的新方法: 基团键贡献法

根据分子结构的特点,通过用染色矩阵和邻接矩阵对分子结构进行矩阵化表征,发展了一种根据分子结构信息烷烃密度的新方法---基团键贡献法。该方法有机地将基团贡献法和化学键贡献法结合在一起,既考虑了分子中基团的特性,又考虑了基团之间的连接性(化学键),具有基团贡献法和化学键贡献法的特点。对658种烷烃的计算结果表明,密度预测值十分接近实验值,平均误差0.245%,进一步外推对聚乙烯、聚丙烯和聚1-丁烯等聚合物的密度进行预测,也取得了令人满意的结果。     

BonDNet: a graph neural network for the prediction of bond dissociation energies for charged molecules

A broad collection of technologies, including e.g. drug metabolism, biofuel combustion, photochemical decontamination of water, and interfacial passivation in energy production/storage systems rely on chemical processes that involve bond-breaking molecular reactions. In this context, a fundamental thermodynamic property of interest is the bond dissociation energy (BDE) which measures the strength of a chemical bond. Fast and accurate prediction of BDEs for arbitrary molecules would lay the groundwork for data-driven projections of complex reaction cascades and hence a deeper understanding of these critical chemical processes and, ultimately, how to reverse design them. In this paper, we propose a chemically inspired graph neural network machine learning model, BonDNet, for the rapid and accurate prediction of BDEs. BonDNet maps the difference between the molecular representations of the reactants and products to the reaction BDE. Because of the use of this difference representation and the introduction of global features, including molecular charge, it is the first machine learning model capable of predicting both homolytic and heterolytic BDEs for molecules of any charge. To test the model, we have constructed a dataset of both homolytic and heterolytic BDEs for neutral and charged (−1 and +1) molecules. BonDNet achieves a mean absolute error (MAE) of 0.022 eV for unseen test data, significantly below chemical accuracy (0.043 eV). Besides the ability to handle complex bond dissociation reactions that no previous model could consider, BonDNet distinguishes itself even in only predicting homolytic BDEs for neutral molecules; it achieves an MAE of 0.020 eV on the PubChem BDE dataset, a 20% improvement over the previous best performing model. We gain additional insight into the model''s predictions by analyzing the patterns in the features representing the molecules and the bond dissociation reactions, which are qualitatively consistent with chemical rules and intuition. BonDNet is just one application of our general approach to representing and learning chemical reactivity, and it could be easily extended to the prediction of other reaction properties in the future.

Prediction of bond dissociation energies for charged molecules with a graph neural network enabled by global molecular features and reaction difference features between products and reactants.     

A molecular view of bond rupture

Molecular mechanisms of rupture will be discussed in the light of recent computational studies. Bonds formed by materials as diverse as simple crystalline solids, glassy polymers, and biological ligands and receptors, reveal similar behavior. When these bonds are pulled apart at constant velocity, they rupture through a series of sudden yield events during which the material reorganizes. Yield events are separated by periods of elastic deformation where the stress builds until the system becomes unstable. The nature of the structural change at yield events varies from system to system. Small cavities form in the polymer film, an additional atomic layer is formed in the crystal, and hydrogen binding sites rearrange in the biological system. The work required to rupture these bonds is determined by the full sequence of yield events.     

A single layer artificial neural network type architecture with molecular engineered bacteria for reversible and irreversible computing

Here, we adapted the basic concept of artificial neural networks (ANNs) and experimentally demonstrate a broadly applicable single layer ANN type architecture with molecular engineered bacteria to perform complex irreversible computing like multiplexing, de-multiplexing, encoding, decoding, majority functions, and reversible computing like Feynman and Fredkin gates. The encoder and majority functions and reversible computing were experimentally implemented within living cells for the first time. We created cellular devices, which worked as artificial neuro-synapses in bacteria, where input chemical signals were linearly combined and processed through a non-linear activation function to produce fluorescent protein outputs. To create such cellular devices, we established a set of rules by correlating truth tables, mathematical equations of ANNs, and cellular device design, which unlike cellular computing, does not require a circuit diagram and the equation directly correlates the design of the cellular device. To our knowledge this is the first adaptation of ANN type architecture with engineered cells. This work may have significance in establishing a new platform for cellular computing, reversible computing and in transforming living cells as ANN-enabled hardware.

We created artificial neural network type architecture with engineered bacteria to perform reversible and irreversible computation. This may work as new computing system for performing complex cellular computation.     

Hopfield neural network model for calculating the potential energy function from second virial data

The calculation of the intermolecular potential from the second virial coefficient is treated here by using a Hopfield neural network model. From simulated data for the prototype system HeNe, the repulsive potential was obtained with a desired accuracy. The algorithm used here is general, as it can handle noise in the experimental data and, a neural network of higher dimension can be easily constructed. Although the inversion of the short-range part of the potential was obtained in the present work, the Hopfield neural network under consideration can equally be used to invert virial data to give the long-range part of the potential. The convergence of the states of the neuron and the accuracy of the inverted potential is also discussed.     

Relationship between the energy of donor–acceptor bond and the reorganization energy in molecular complexes

Donor–acceptor complexes of silicon halides with ammonia, pyridine, and 2,2′bipyridine SiX₄ · nD (X = F, Cl, Br) have been studied at the B3LYP/pVDZ level of theory. Energies of the donor–acceptor bond have been estimated taking into account the reorganization energy of the donor and acceptor fragments and basis set superposition error correction. Despite of the very low (or even negative) dissociation energy of SiX₄ · nD into free fragments, the Si–N bonding in all complexes is rather strong (75–220 kJ mol^?1). It is the reorganization energy of the acceptor SiX₄ (75–280 kJ mol^?1) that governs the dissociation energy of the complex. Thus, in contrast to the complexes of group 13 halides, the reorganization effects are crucial for the complexes of group 14 halides, and their neglecting leads to erroneous conclusions about the strength of the donor–acceptor bond in these systems. © 2002 Wiley Periodicals, Inc. Int J Quantum Chem, 2002     

A constant denominator perturbation theory for molecular energy

A constant denominator perturbation theory is developed based on a zeroth order Hamiltonian characterized by degenerate subsets of orbitals. Such a formulation allows for a decoupling of the numerators of the perturbation sequence, allowing for much more rapid evaluation of the resultant sums. For example, the full fourth order theory can be evaluated as an N ⁶ step rather than N ⁷, where N is proportional to the basis set.Although the theory is general, a constant denominator is chosen for this study as the difference between the average occupied and average virtual orbital energies scaled so that the first order wavefunction yields the lowest possible variational bound. The third order correction then appears naturally as a scaled Langhoff-Davidson correction. The full fourth order with this partitioning is developed. Results are presented within the localized bond model utilizing both the Pariser-Parr-Pople and CNDO/2 model Hamiltonians. The second order theory presents a useful bound, usually containing a good deal of the basis set correlation. In all cases examined the fourth order theory shows remarkable stability, even in those cases in which the Nesbet-Epstein partitioning seems poorly convergent, and the Moller-Plesset theory uncertain.     

A self‐assembling neural network for modeling polymers

A central problem in modeling protein and other polymer structures is the generation of self‐avoiding chains which obey a priori distance restraint information which could include a folding potential function. This problem is usually addressed with a lattice model or a torsion space model of the polymer. Exhaustive searches in these spaces are of necessity exponentially complex. A new computer algorithm for modeling polymers and polymeric systems is described. This algorithm is a randomized algorithm based on a self‐assembling or Kohonen neural network. Given a defined chain topology, a defined spatial extent, and a prior probability distribution, it finds a set of coordinates which reproduce these properties. The convergence rate of the algorithm is linear with respect to the number of distance terms included. Modifications to the standard Kohonen algorithm to include a defined spatial metric, and a modified update rule improve the convergence of the standard algorithm and result in a highly efficient algorithm for building polymer models which are self avoiding and consistent with prior probability information and interatomic distance restraints. This revised version was published online in July 2006 with corrections to the Cover Date.     

Cluster analysis from molecular similarity matrices using a non-linear neural network

A non-linear neural network model to perform cluster analysis is presented. It provides an efficient parallel algorithm for solving this pattern recognition task, consisting, from the mathematical point of view, of a combinatorial optimization problem. A new classification technique is discussed in order to visualize clustering patterns within a molecular set, by means of numerical analysis of the similarity matrix. As an example of the application of the reported neural network model, a quantum molecular similarity study in the field of structure-activity relationships is reported. A molecular set made of eighteen quinolones is used as an example. The resultant cluster distribution showed a good qualitative correlation between similarity data and biological activity.