Can Machine Learning Predict the Phase Behavior of Surfactants?

We explore the prediction of surfactant phase behavior using state-of-the-art machine learning methods, using a data set for twenty-three nonionic surfactants. Most machine learning classifiers we tested are capable of filling in missing data in a partially complete data set. However, strong data bias and a lack of chemical space information generally lead to poorer results for entire de novo phase diagram prediction. Although some machine learning classifiers perform better than others, these observations are largely robust to the particular choice of algorithm. Finally, we explore how de novo phase diagram prediction can be improved by the inclusion of observations from state points sampled by an analogy to commonly used experimental protocols. Our results indicate what factors should be considered when preparing for machine learning prediction of surfactant phase behavior in future studies.

A FFLUX Water Model: Flexible, Polarizable and with a Multipolar Description of Electrostatics.

Key to progress in molecular simulation is the development of advanced models that go beyond the limitations of traditional force fields that employ a fixed, point charge-based description of electrostatics. Taking water as an example system, the FFLUX framework is shown capable of producing models that are flexible, polarizable and have a multipolar description of the electrostatics. The kriging machine-learning methods used in FFLUX are able to reproduce the intramolecular potential energy surface and multipole moments of a single water molecule with chemical accuracy using as few as 50 training configurations. Molecular dynamics simulations of water clusters (25-216 molecules) using the new FFLUX model reveal that incorporating charge-quadrupole, dipole-dipole, and quadrupole-charge interactions into the description of the electrostatics results in significant changes to the intermolecular structuring of the water molecules. © 2019 The Authors. Journal of Computational Chemistry published by Wiley Periodicals, Inc.

Directionality of Halogen Bonds: An Interacting Quantum Atoms (IQA) and Relative Energy Gradient (REG) Study.

Interacting Quantum Atoms (IQA) and Interacting Quantum Fragments (IQF) analyses are used to study F 3 C - X ⋯ N H 3 (X=Cl and Br) model complexes in order to determine the origin of halogen bond directionality. IQA allows for the calculation of intra- and interatomic classical and exchange-correlation energies, which can be used to determine the energetic nature of the changes that occur when deviating from the preferred halogen bond approach. The Relative Energy Gradient (REG) method is also applied to rank the IQA energies and reveal which energy contributions best describe the total behavior of the system. Indeed, all the pairwise interactions and atomic self-energies are angularly dependent; some terms favor the linear structure and some tend toward nonlinear arrangements. For instance, when the C-X-N angle is altered, the halogen-nitrogen interaction energy behaves like the total energy of the system while the carbon-nitrogen interaction works against the total energy profile. Furthermore, the REG values reveal that the contribution of the halogen-nitrogen interaction to the total behavior of the system is small. Instead, the secondary interactions (e. g., fluorine-nitrogen and carbon-hydrogen interactions) and atomic self-energies are mainly responsible for the angular preference of these halogen bonds. Finally, IQF calculations followed by REG analysis reveal the importance of the self-energy of the fragments.

A relative energy gradient (REG) study of the planar and perpendicular torsional energy barriers in biphenyl.

Biphenyl is a prototype molecule, the study of which is important for a proper understanding of stereo-electronic effects. In the gas phase it has an equilibrium central torsion angle of ~ 45° and shows both a planar (0°) and a perpendicular (90°) torsional energy barrier. The latter is analysed for the first time. We use the newly proposed REG method, which is an exhaustive procedure that automatically ranks atomic energy contributions according to their importance in explaining the energy profile of a total system. Here, the REG method operates on energy contributions computed by the interacting quantum atoms method. This method is minimal in architecture and provides a crisp picture of well-defined and well-separated electrostatic, steric and exchange (covalent) energies at atomistic level. It is shown that the bond critical point occurring between the ortho-hydrogens in the planar geometry has been wrongly interpreted as a sign of repulsive interaction. A convenient metaphor of analysing football matches is introduced to clarify the role of a REG analysis.

A Re-evaluation of Factors Controlling the Nature of Complementary Hydrogen-Bonded Networks.

The energy profiles of hydrogen-bonded heterocyclic aromatics have been decomposed into atomistic energy contributions using the Interacting Quantum Atoms (IQA) method. The resulting energy contributions have been sequenced by the Relative Energy Gradient (REG) approach to determine their influence upon the shape of these energy profiles. The results show inadequacies in Jorgensen's secondary interaction hypothesis (SIH). A novel method of finding a condensed analogy for the interaction between the molecules is presented. The findings of this work further doubt the validity of the SIH, and reinforce previous warnings against its misguided use.

Description of Potential Energy Surfaces of Molecules Using FFLUX Machine Learning Models.

A new type of model, FFLUX, to describe the interaction between atoms has been developed as an alternative to traditional force fields. FFLUX models are constructed by applying the kriging machine learning method to the topological energy partitioning method, interacting quantum atoms (IQA). The effect of varying parameters in the construction of the FFLUX models is analyzed, with the most dominant effects found to be the structure of the molecule and the number of conformations used to build the model. Using these models, the optimization of a variety of small organic molecules is performed, with sub kJ mol-1 accuracy in the energy of the optimized molecules. The FFLUX models are also evaluated in terms of their performance in describing the potential energy surfaces (PESs) associated with specific degrees of freedoms within molecules. While the accurate description of PESs presents greater challenges than individual minima, FFLUX models are able to achieve errors of <2.5 kJ mol-1 across the full C-C-C-C dihedral PES of n-butane, indicating the future possibilities of the technique.

Using the Relative Energy Gradient Method with Interacting Quantum Atoms to Determine the Reaction Mechanism and Catalytic Effects in the Peptide Hydrolysis in HIV-1 Protease.

The reaction mechanism in an active site is of the utmost importance when trying to understand the role that an enzyme plays in biological processes. In a recently published paper [Theor. Chem. Acc. 2017, 136, 86], we formalised the Relative Energy Gradient (REG) method for automating an Interacting Quantum Atoms (IQA) analysis. Here, the REG method is utilised to determine the mechanism of peptide hydrolysis in the aspartic active site of the enzyme HIV-1 Protease. Using the REG method along with the IQA approach we determine the mechanism of peptide hydrolysis without employing any arbitrary parameters and with remarkable ease (albeit at large computational cost: the system contains 133 atoms, which means that there are 17 689 individual IQA terms to be calculated). When REG and IQA work together it is possible to determine a reaction mechanism at atomistic resolution from data directly derived from quantum calculations, without arbitrary parameters. Moreover, the mechanism determined by this novel method gives concrete insight into how the active site residues catalyse peptide hydrolysis.

Fluorine Gauche Effect Explained by Electrostatic Polarization Instead of Hyperconjugation: An Interacting Quantum Atoms (IQA) and Relative Energy Gradient (REG) Study.

We present an interacting quantum atoms (IQA) study of the gauche effect by comparing 1,2-difluoroethane, 1,2-dichloroethane, and three conformers of 1,2,3,4,5,6-hexafluorocyclohexane. In the 1,2-difluoroethane, the gauche effect is observed in that the gauche conformation is more stable than the anti, whereas in 1,2-dichloroethane the opposite is true. The analysis performed here is exhaustive and unbiased thanks to using the recently introduced relative energy gradient (REG) method [ Thacker , J. C. R. ; Popelier , P. L. A. Theor. Chem. Acc . 2017 , 136 , 86 ], as implemented in the in-house program ANANKE. We challenge the common explanation that hyperconjugation is responsible for the gauche stability in 1,2-difluoroethane and instead present electrostatics as the cause of gauche stability. Our explanation of the gauche effect is also is seen in other molecules displaying local gauche conformations, such as the recently synthesized "all-cis" hexafluorocyclohexane and its conformers where all the fluorine atoms are in the equatorial positions. Using our extension of the traditional IQA methodology that allows for the partitioning of electrostatic terms into polarization and charge transfer, we propose that the cause of gauche stability is 1,3 C···F electrostatic polarization interactions. In other words, if a number of fluorine atoms are aligned, then the stability due to polarization of nearby carbon atoms is increased.

An interacting quantum atom study of model SN 2 reactions (X- ···CH3 X, X = F, Cl, Br, and I).

The quantum chemical topology method has been used to analyze the energetic profiles in the X- + CH3 X → XCH3 + X- SN 2 reactions, with X = F, Cl, Br, and I. The evolution of the electron density properties at the BCPs along the reaction coordinate has been analysed. The interacting quantum atoms (IQA) method has been used to evaluate the intra-atomic and interatomic energy variations along the reaction path. The different energetic terms have been examined by the relative energy gradient method and the ANANKE program, which enables automatic and unbiased IQA analysis. Four of the six most important IQA energy contributions were needed to reproduce the reaction barrier common to all reactions. The four reactions considered share many common characteristics but when X = F a number of particularities occur. © 2017 Wiley Periodicals, Inc.

The ANANKE relative energy gradient (REG) method to automate IQA analysis over configurational change.

Much chemical insight ultimately comes down to finding out which fragment of a total system behaves like the total system, in terms of an energy profile. A simple example is that of the water dimer, where this system is regarded as held together by a hydrogen bond. The hydrogen bond consists of two atoms (H···O), which energetically behave similarly to the total system (H2O)2. However, from a quantum mechanical point of view, each atom in the total system interacts with any other atom. Thus, the view that the hydrogen bond by itself governs the energetic stability of the water dimer needs rigorous justification. In this work, we propose a method that provides such a justification, in general, but only illustrated on the water dimer here. This method is based on the topological energy partitioning method called interacting quantum atoms (IQA). The method is implemented in the program ANANKE, which calculates correlations between the energy profile of the total system and those of subsystems (or fragments). ANANKE acts on the IQA energy contributions obtained for a sequence of full-system geometries controlled by a coordinate of interest (e.g. the O···H distance in the water dimer). Although applied only for the water dimer in this work, the method is general and able to explain the gauche effect, the torsional barrier in biphenyl, the arrow-pushing scheme of an enzymatic reaction (peptide hydrolysis in the HIV-1 Protease active site), and halogen-alkane nucleophilic substitution (SN2) reactions. Those applications will appear elsewhere as separate and elaborated case studies; here we focus on the details of the ANANKE method and its justification, using the water dimer as a concrete case.