Tuning molecule-surface interactions with sub-nanometer-thick covalently bound organic monolayers

Measurements of the thermal desorption of methyl bromide (MeBr) from bare and RS-functionalized GaAs(110), where R = CH3 and CH3CH2, reveal marked systematic changes in molecule-surface interactions. As the thickness of the organic spacer layer is increased, the electrostatic MeBr-GaAs(110) interaction decreases, lowering the activation energy for desorption, Ed, as well as decreasing the critical coverage required for nucleation of bulklike MeBr. On the CH3CH2S-functionalized surface, Ed is lowered to a value roughly equal to that for desorption from three-dimensional (3-D) clusters; because the kinetics of desorption of isolated molecules differs from that for desorption from clusters, desorption of isolated molecules from the organic surface occurs at a lower temperature than desorption from the clusters. Thus, the "monolayer" desorption wave occurs at a lower temperature than the "multilayer" desorption wave. These results illustrate the role that organic chain length in nanometer-scale thin films can play in alteration of the delicate balance of interfacial interactions.     

Third-order interactions in symmetry-adapted perturbation theory

We present an extension of many-body symmetry-adapted perturbation theory (SAPT) by including all third-order polarization and exchange contributions obtained with the neglect of intramonomer correlation effects. The third-order polarization energy, which naturally decomposes into the induction, dispersion, and mixed, induction-dispersion components, is significantly quenched at short range by electron exchange effects. We propose a decomposition of the total third-order exchange energy into the exchange-induction, exchange-dispersion, and exchange-induction-dispersion contributions which provide the quenching for the corresponding individual polarization contributions. All components of the third-order energy have been expressed in terms of molecular integrals and orbital energies. The obtained formulas, valid for both dimer- and monomer-centered basis sets, have been implemented within the general closed-shell many-electron SAPT program. Test calculations for several small dimers have been performed and their results are presented. For dispersion-bound dimers, the inclusion of the third-order effects eliminates the need for a hybrid SAPT approach, involving supermolecular Hartree-Fock calculations. For dimers consisting of strongly polar monomers, the hybrid approach remains more accurate. It is shown that, due to the extent of the quenching, the third-order polarization effects should be included only together with their exchange counterparts. Furthermore, the latter have to be calculated exactly, rather than estimated by scaling the second-order values.     

Predicting chemical shifts with graph neural networks

Inferring molecular structure from Nuclear Magnetic Resonance (NMR) measurements requires an accurate forward model that can predict chemical shifts from 3D structure. Current forward models are limited to specific molecules like proteins and state-of-the-art models are not differentiable. Thus they cannot be used with gradient methods like biased molecular dynamics. Here we use graph neural networks (GNNs) for NMR chemical shift prediction. Our GNN can model chemical shifts accurately and capture important phenomena like hydrogen bonding induced downfield shift between multiple proteins, secondary structure effects, and predict shifts of organic molecules. Previous empirical NMR models of protein NMR have relied on careful feature engineering with domain expertise. These GNNs are trained from data alone with no feature engineering yet are as accurate and can work on arbitrary molecular structures. The models are also efficient, able to compute one million chemical shifts in about 5 seconds. This work enables a new category of NMR models that have multiple interacting types of macromolecules.

This model can predict chemical shifts on proteins and small molecules purely from atom elements and coordinates. It can capture important phenomena like hydrogen bonding induced downfield shift, thus can be used to infer intermolecular interactions.     

Confidence intervals for calibration with neural networks

Based on neural network calibration the confidence intervals of aromaticity determination from infrared reflectance spectra of raw brown coals were estimated by means of the bootstrap method, a simplified Monte Carlo Simulation. The standard deviations and the confidence intervals were estimated to characterise the analysis error.It is shown that confidence intervals of non-linear analysis methods like Back Propagation Neural Networks (BPNN) can be estimated by the bootstrap method. The estimated confidence intervals of the calibration confirm the analysis by BPNN.     

A journey into low-dimensional spaces with autoassociative neural networks

The compression and the visualization of the data have been always a subject of a great deal of excitement. Since multidimensional data sets are difficult to interpret and visualize, much of the attention is drawn how to compress them efficiently. Usually, the compression of dimensionality is considered as the first step of exploratory data analysis. Here, we focus our attention on autoassociative neural networks (ANNs), which in a very elegant manner provide data compression and visualization. ANNs can deal with linear and nonlinear correlation among variables, what makes them a very powerful tool in exploratory data analysis. In the literature, ANNs are often referred as nonlinear principal component analysis (PCA), and due to their specific structure they are also known as bottleneck neural networks. In this paper, ANNs are discussed in details. Different training modes are described and illustrated on real example. The usefulness of ANNs for nonlinear data compression and visualization purposes is proven with the aid of chemical data sets, being the subject of analysis. The comparison of ANNs with well-known PCA is also presented.     

Data modelling with neural networks: Advantages and limitations

The origins and operation of artificial neural networks are briefly described and their early application to data modelling in drug design is reviewed. Four problems in the use of neural networks in data modelling are discussed, namely overfitting, chance effects, overtraining and interpretation, and examples are given of the means by which the first three of these may be avoided. The use of neural networks as a variable selection tool is shown and the advantage of networks as a nonlinear data modelling device is discussed. The display of multivariate data in two dimensions employing a neural network is illustrated using experimental and theoretical data for a set of charge transfer complexes.     

Supramolecular isomerism in spin crossover networks with aurophilic interactions

Assembly of Fe(II), 3-cyanopyridine and [Au(CN)2](-) affords, in one-pot reaction, three coordination polymers that represent a genuine example of supramolecular isomerism with strong influence in the spin crossover regime of the Fe(II) ions.     

Investigating charge transport in molecular switches with neural networks

The control of charge transport on a polymer chain by impurity molecules working as switches is studied. Charge propagation is estimated using a backpropagation neural network approach. The supervised learning is accomplished using theoretical results in which the chain is modeled by a tight-binding Hamiltonian extended to include the effects of an external electric field. The charge transport through the sites that work like a switch is analyzed by the numerical integration of the equations of motion. For a donor–acceptor pair of impurities, we found that the chain offers a wide range of devices, from simple switches to perfect molecular rectifiers. The influence of the parameters of the molecules on the charge transport, the role of the length of separation between the sites where the impurity molecules bond, as well as the changes they must undergo to characterize each kind of molecular switch are determined. ©1999 John Wiley & Sons, Inc. J Comput Chem 20: 1060–1066, 1999     

Rapid computation of intermolecular interactions in molecular and ionic clusters: self-consistent polarization plus symmetry-adapted perturbation theory

A method that we have recently introduced for rapid computation of intermolecular interaction energies is reformulated and subjected to further tests. The method employs monomer-based self-consistent field calculations with an electrostatic embedding designed to capture many-body polarization (the "XPol" procedure), augmented by pairwise symmetry-adapted perturbation theory (SAPT) to capture dispersion and exchange interactions along with any remaining induction effects. A rigorous derivation of the XPol+SAPT methodology is presented here, which demonstrates that the method is systematically improvable, and moreover introduces some additional intermolecular interactions as compared to the more heuristic derivation that was presented previously. Applications to various non-covalent complexes and clusters are presented, including geometry optimizations and one-dimensional potential energy scans. The performance of the XPol+SAPT methodology in its present form (based on second-order intermolecular perturbation theory and neglecting intramolecular electron correlation) is qualitatively acceptable across a wide variety of systems-and quantitatively quite good in certain cases-but the quality of the results is rather sensitive to the choice of one-particle basis set. Basis sets that work well for dispersion-bound systems offer less-than-optimal performance for clusters dominated by induction and electrostatic interactions, and vice versa. A compromise basis set is identified that affords good results for both induction and dispersion interactions, although this favorable performance ultimately relies on error cancellation, as in traditional low-order SAPT. Suggestions for future improvements to the methodology are discussed.     

Estimating correlation energy of diatomic molecules and atoms with neural networks

The electronic correlation energy of diatomic molecules and heavy atoms is estimated using a back propagation neural network approach. The supervised learning is accomplished using known exact results of the electronic correlation energy. The recall rate, that is, the performance of the net in recognizing the training set, is about 96%. The correctness of values given to the test set and prediction rate is at the 90% level. We generate tables for the electronic correlation energy of several diatomic molecules and all the neutral atoms up to radon (Rn). © 1997 by John Wiley & Sons, Inc. J Comput Chem 18 : 1407–1414, 1997     

Evaluation of chemical equilibria with the use of artificial neural networks

Multivariate calibration with experimental design (ED) and artificial neural networks (ANN) modeling can be used to estimate equilibria constants from any kind of protonation or metal–ligand equilibrium data like potentiometry, polarography, spectrophotometry, extraction, etc. The method was tested on evenly or randomly distributed experimental error-free data and data with random noise and the results show that even rather higher experimental errors do not influence significantly the prediction power and correctness of ANN prediction. ANN with appropriate ED can provide accurate prediction of stability constants with the relative errors in the range of ±4% or smaller while the approach is very robust. Comparison with a hard model evaluation based on non-linear regression techniques shows excellent agreement. Proposed ANN method is of a general nature and, in principal, can be adopted to any analytical technique used in equilibria studies.     

SyntaLinker: automatic fragment linking with deep conditional transformer neural networks

Linking fragments to generate a focused compound library for a specific drug target is one of the challenges in fragment-based drug design (FBDD). Hereby, we propose a new program named SyntaLinker, which is based on a syntactic pattern recognition approach using deep conditional transformer neural networks. This state-of-the-art transformer can link molecular fragments automatically by learning from the knowledge of structures in medicinal chemistry databases (e.g. ChEMBL database). Conventionally, linking molecular fragments was viewed as connecting substructures that were predefined by empirical rules. In SyntaLinker, however, the rules of linking fragments can be learned implicitly from known chemical structures by recognizing syntactic patterns embedded in SMILES notations. With deep conditional transformer neural networks, SyntaLinker can generate molecular structures based on a given pair of fragments and additional restrictions. Case studies have demonstrated the advantages and usefulness of SyntaLinker in FBDD.

Linking fragments to generate a focused compound library for a specific drug target is one of the challenges in fragment-based drug design (FBDD).     

Large-scale symmetry-adapted perturbation theory computations via density fitting and Laplace transformation techniques: investigating the fundamental forces of DNA-intercalator interactions

Symmetry-adapted perturbation theory (SAPT) provides a means of probing the fundamental nature of intermolecular interactions. Low-orders of SAPT (here, SAPT0) are especially attractive since they provide qualitative (sometimes quantitative) results while remaining tractable for large systems. The application of density fitting and Laplace transformation techniques to SAPT0 can significantly reduce the expense associated with these computations and make even larger systems accessible. We present new factorizations of the SAPT0 equations with density-fitted two-electron integrals and the first application of Laplace transformations of energy denominators to SAPT. The improved scalability of the DF-SAPT0 implementation allows it to be applied to systems with more than 200 atoms and 2800 basis functions. The Laplace-transformed energy denominators are compared to analogous partial Cholesky decompositions of the energy denominator tensor. Application of our new DF-SAPT0 program to the intercalation of DNA by proflavine has allowed us to determine the nature of the proflavine-DNA interaction. Overall, the proflavine-DNA interaction contains important contributions from both electrostatics and dispersion. The energetics of the intercalator interaction are are dominated by the stacking interactions (two-thirds of the total), but contain important contributions from the intercalator-backbone interactions. It is hypothesized that the geometry of the complex will be determined by the interactions of the intercalator with the backbone, because by shifting toward one side of the backbone, the intercalator can form two long hydrogen-bonding type interactions. The long-range interactions between the intercalator and the next-nearest base pairs appear to be negligible, justifying the use of truncated DNA models in computational studies of intercalation interaction energies.     

Application of high-order neural networks in chemistry

Summary The effectiveness and usefulness of the so-called high-order neural networks for classification of chemical objects is demonstrated. The high-order neural networks usually do not need hidden neurons for correct interpretation of patterns. A simple formula for partial derivatives of the minimized objective (error) function is derived, which is used for production of weight coefficients during the adaptation process. An illustrative example dealing with inductive and resonance effects of functional groups by the second-order neural network is presented.     

Degenerate symmetry-adapted perturbation theory of weak interactions between closed- and open-shell monomers: application to Rydberg states of helium hydride

Symmetry-adapted perturbation theory is extended to the (quasi) degenerate, open-shell case. The new formalism is tested in calculations of the interaction energies for a helium atom in the ground state interacting with an excited hydrogen atom. It is shown that the method gives satisfactory results if the coupling with higher Rydberg states of the dimer is small, as is the case for the A²Σ⁺,B²Π,E²Π,3²Π, and 1²Δ states of HeH. For the C²Σ⁺ state convergence of the method is very slow, but it can be improved by including the n=3 states in the model space. Received: 3 June 1998 / Accepted: 9 September 1998 / Published online: 7 December 1998     

Use of neural networks in on-line analysis

The evaluation principles of neural networks are presented and compared with common known techniques. The concepts in data processing, introduced by neural nets are explained and processing types implemented by neural networks are presented. The evaluation of gas sensitive sensors will be an example for the special features of neural nets, with a focus to the self-organizing map.@peanuts.informatik.uni-tuebingen.de     

QSAR modelling using combined simple competitive learning networks and RBF neural networks

The aim of this study was to propose a QSAR modelling approach based on the combination of simple competitive learning (SCL) networks with radial basis function (RBF) neural networks for predicting the biological activity of chemical compounds. The proposed QSAR method consisted of two phases. In the first phase, an SCL network was applied to determine the centres of an RBF neural network. In the second phase, the RBF neural network was used to predict the biological activity of various phenols and Rho kinase (ROCK) inhibitors. The predictive ability of the proposed QSAR models was evaluated and compared with other QSAR models using external validation. The results of this study showed that the proposed QSAR modelling approach leads to better performances than other models in predicting the biological activity of chemical compounds. This indicated the efficiency of simple competitive learning networks in determining the centres of RBF neural networks.