In the theoretical study of molecular systems, a compromise between speed and accuracy is required to study the energetics of chemical systems. Quantum mechanical (QM) methods allow accurate energies and forces to be calculated but require massive computational effort. Classical force fields are fast but only accurate near equilibrium and are generally unusable in reactivity studies due to restrictive functional forms. A possible solution to these problems is the development of empirical potentials built through machine learning methods. Artificial neural networks have been used to develop neural network potentials (NNP), which are fit to QM reference energies. Through the development of a new methodology, known as ANAKIN-ME (ANI), we provide the tools to build a new class of NNP, which is fully transferable and chemically accurate. With the ANI method, we develop the ANI-1.2 potential for organic molecules containing H, C, N, and O. Through extensive benchmark and case studies, the ANI-1.2 potential provides evidence that the ANI method produces chemically accurate and size extensible potentials. The ANI method brings a new, highly efficient, and accurate method for the development of NNPs into the realm of reality, and opens the door for a new generation of “out-of-the-box” general purpose potentials.

The activities of many proteins, including GPCRs, NTFs and Igs, are regulated by small changes in structures that are comparable to thermal fluctuations. Consequently, their regulatory mechanisms cannot be modeled in terms of how their energy-minimum structures differ between states. Understanding their regulatory mechanisms requires assessment of relationships between high-dimensional conformational ensembles of different states. To realize this, we present development of a new class of approaches based on support vector machines (SVMs). However, we do not use SVMs in the traditional manner to predict group identities of unseen instances. Instead, we use the mathematical framework of SVMs to identify instances and spatial regions that overlap between pre-classified groups. We also show how this development enables us to statistically analyze relationships between multiple conformational ensembles and compute correlations in inter-site ensemble shifts, thereby providing direct insight into how regulatory signals are spread through a combination of changes in structure and dynamics.

Identification of bioactive compounds with desired therapeutic effects is still a laborious, expensive process.  Moreover, once a compound has been identified,  which acts against a particular biomolecular target, the question still remains as to whether all relevant protein targets for the compound have been identified. Thus, it is desirable to understand the breadth of biomolecular targets with which a particular compound interacts in order to establish the compound’s selectivity or lack thereof (polypharmacology). Such an endeavor would also be of utility for re-purposing known drugs and elucidation of the possible metabolic fate of a compound of interest.  Previously, we developed Virtual Target Screening (VTS) methodology that allows one to computationally screen a compound of interest against a collection of  protein targets. By scoring each compound‑protein interaction, we can compare against averaged scores of a collection of drug-like molecules docked to the same proteins to determine if a particular protein would be a potential target of the compound of interest.  We have validated our VTS system using kinase inhibitors and natural products.  In this talk, we discuss our work in progress.

Cytochrome P450cam from P. putida (P450cam) and chloroperoxidase from C. fumago (CPO) are highly versatile enzymes capable of catalyzing a broad spectrum of oxidation reactions, including the epoxidation and hydroxylation of organic substrates.  These reactions can be highly enantiospecific, making engineered CPO variants potentially useful biocatalysts.  We have previously reported a body of simulation work on the structure-function relationships of CPO.  Here we report work on CPO and P450cam that traces the difference between the epoxidation and hydroxylation activities to the environment of the heme’s proximal thiolate ligand.  We show, on the basis of QM and QM/MM models, that this environment can favor epoxidation over hydroxylation by up to 5 kcal/mol, even though the reaction takes place on the opposite heme face (distal binding pocket), resolving a long-standing discrepancy between experiment and simulation.  We also provide evidence that the proximal environment can, via electronic effects, be a significant factor in controlling the enantiospecificity of CPO-catalyzed epoxidation reactions.

The sulfur nucleotide PAPS (3’-phosphoadenosine 5’-phosphosulfate) is the universal sulfuryl donor of the cell. In mammals 3’-phosphoadenosine 5’-phosphosulfate Synthase (PAPSS), using ATP, converts biochemically inert inorganic sulfate to metabolically active PAPS. PAPS synthase is a bi-functional enzyme and catalyzes the formation of PAPS in two sequential steps. Prior experimental, site selected mutagenesis of human PAPSS HNGH motif (amino acid residues 425-428) revealed drastic changes in enzymatic activity. With this prior knowledge, retrospectively we first set out to calculate in silico, the binding energies of the wild type and the mutants using molecular modeling and dynamics. The binding of ATP in wild type was exothermic that required -40 kJ/mol energy. In mutant H425A, H428A it required 524 kJ/mol of energy for ATP binding, and change into alanine made it very endothermic and inactive as a catalyst. G427A mutant was endothermic, requiring binding energy of 218 kJ/mol. Interestingly, N426K was very exothermic which required even less binding energy (-70kJ/mol), matching with increased activity. With autodocking program, the pre-reaction position of inorganic sulfate was predicted. Two arginine (R522 and R421) of PAPSS1 were well poised with sulfate for nucleophilic attack with ATP.

Metal-organic frameworks are highly porous crystalline materials well-suited for computer simulation (primarily because of periodicity of structure). Grand-canonical (constant μ,V,T) Monte Carlo (a method mostly known for random perturbations which are accepted or rejected by a Boltzmann probability) simulations of hydrogen (77 and 87K), carbon dioxide, methane, acetylene, ethylene, and ethane gas (298K) sorption in rht-MOF-9 were performed using Massively Parallel Monte Carlo (MPMC), a statistical-mechanical molecular simulation code developed by our lab. Gas uptake (storage) isotherms and Qst (heat of adsorption) were calculated, and efforts were made to discover the primary binding-sites of the gases via radial distribution calculations and simulated annealing. rht-MOF-9 is a copper based MOF with 3 distinct cages. Theoretical results were compared to experimental data.

Modeling and understanding quantum heat engines and photochemical energy harvesting processes requires an extension of traditional thermodynamical models.  They are challenging both because net flows of energy are only possible out of equilibrium and also because quantum systems do not stay diagonal under these conditions.  This talk derives a consistent thermodynamics for quantum nonequilibrium processes when repeated measurement of some part of the system is possible.  Its limit under a slow rate of measurement results in well-understood weak-coupling expressions for energy and entropy.  Using the example of a molecular dipole emitting light into an empty room, we show that under fast measurement, the emission process displays damped oscillatory behavior.  Moreover, full measurement of the emitted light leads to a non-Boltzmann steady-state for the dipole.

Plasmonic crystals are periodic arrays of metallic nanostructures that support surface plasmons, or collective oscillations of conducting electrons at the interface between a metal and a dielectric. Surface plasmons can exhibit unique optical properties, such as enhanced absorption, scattering and field confinement of light at the resonance frequency. This so called surface plasmon resonance (SPR) is sensitive to the size, shape, and dielectric environment of a plasmonic crystal and is therefore tunable for particular applications. Such properties make plasmonic crystals attractive for spectroscopy based sensing applications. Here we present a survey of work that involves two types of plasmonic crystals composed of metal coated, square arrays of either nanowell or nanopost structures formed via soft nanoimprinting as Surface Enhanced Raman Scattering (SERS) substrates and refractive index sensors. Such crystals can exhibit SERS enhancement factors of 10⁵ to 10⁶ over large areas and with sufficiently high levels of uniformity for precise two-dimensional Raman mapping of surface bound monolayers. Three-dimensional finite-difference time-domain (3D FDTD) simulations qualitatively capture the key features of these systems and suggests a route to the fabrication of optimized, highly efficient SERS substrates or refractive index sensors in silco. Collectively, the ease of fabrication together with the high sensitivities and spatial resolution that can be achieved suggests an attractive route to the design and optimization SERS substrates and sensors for portable chemical warfare agent detection, environmental monitors, noninvasive imaging of biomolecules, and other applications.