| A | B | C | D | E | F | G | H | I | J | |
|---|---|---|---|---|---|---|---|---|---|---|
1 | Name of the presenter | Poster Title | Abstract | |||||||
2 | 1 | Avery Hill, Tulane University | Optimal Ion Placement in Nanoporous Zeolites | We perform minima hopping optimization to find where Cu atoms naturally reside inside the nano porous zeolite, SSZ-13. | ||||||
3 | 2 | Kenneth Kusima, University of Houston | Machine Learned Corrections to Transient Micro-Kinetic Models | Transient mean-field micro-kinetic (MF-MK) modeling is a powerful approach to study, simulate, and forecast the reaction kinetics of heterogeneously catalyzed reactions. Using elementary steps to represent molecular-level interactions among chemical species allows for the exploration of the reaction mechanism to provide detailed kinetic information that aids in the interpretation of experimental findings. Typical MK models rely on the mean field approximation that disregards lateral interactions between surface adsorbates. In many cases, such interactions have significant effects on overall reaction kinetics. Kinetic Monte Carlo (kMC) simulations produce coverage and reaction rate information that accounts for lateral surface adsorbate interactions. Nevertheless, kMC simulations are computationally expensive and require meticulous setup for accurate results. Using machine learning (ML), we have been able to develop an improved MK model that uses a machine learned correction factor to obtain kMC-like results from the MF-MK model simulations at different coverages. The resulting ML-MK model appears to match closely with transient kMC simulation results. Consequently, our enhanced ML-MK model can predict the effects of surface adsorbate interactions that cannot be captured in a MF-MK model, while also maintaining an easy setup and minimal computational effort. | ||||||
4 | 3 | Woodrow Wilson, Mississippi State University | Developing a Baysian Force Field for Exploring the Reaction Pathways of Furfural Conversion & Hydrogen Dynamics over β-Mo2C | Using conventional ab initio molecular dynamics (AIMD) to sample the free energy surface of reactions over transition metal surfaces is computationally prohibitive due to the numerical complexity of solving the Kohn-Sham equations for conductive materials. In this case study, we begin developing a Bayesian force field based on Gaussian process regression trained on short AIMD trajectories for furfural conversion over the (101) facet of Mo2C in the surface adsorbed and molecular hydrogen. While the current iteration of the force field is unable to model furfural over Mo2C, the model can capture hydrogen dynamics and adsorption of hydrogen over the catalyst surface. With this force field, we observe that hydrogen adsorbs strongly to hollow sites on the catalyst coordinated by 3 molybdenum atoms, and diffusion of hydrogen on the surface only occurs at higher temperatures and does not occur over the carbon atoms. Challenges with training the force field and possible approaches to sample furfural conversion are also discussed. | ||||||
5 | 4 | Liwei Chang, University of Florida | What AlphaFold know about protein folding and binding | Machine learning approaches have recently revolutionized structural biology by predicting 3D structures of proteins at atomistic level from their sequence with high confidence and accuracy. The community is pushing the limits of interpretability and application of these algorithms beyond their original objective. Here, we present our studies on what deep neural network models for protein structure prediction such as AlphaFold can shed light on two fundamental biological processes: (1) protein folding and (2) protein-peptide binding. We first demonstrate our studies on deciphering the folding mechanism of protein G, L and their mutants, two pair of proteins with low sequence similarity that fold into the same topology. By using large-scale molecular dynamics simulations with advanced enhanced sampling methods and Markov state modeling analysis, we observed how the sequence difference impacts folding pathways and rate in unprecedented detail. Then, we show a novel fragment decomposition approach using AlphaFold to identify preferences for secondary structure element combinations that follow the order of events observed in the folding pathways. Standing upon the highly sensitive mapping of AlphaFold between the sequence and structure, we discovered how to directly use the model to identify high affinity peptide binders using a competitive binding assay. For systems in which the individual structures of the peptides are well predicted, the assay captures the higher affinity binder in the bound state, and the other peptide in the unbound form. The speed and robustness of the method will make it readily applicable in screening libraries of peptide sequences to prioritize for detailed experimental characterization. | ||||||
6 | 5 | Feranmi Olowookere, University of Alabama | Effects of Chain Length on the Structure and Dynamics of Polyvinyl Chloride During Atomistic Molecular Dynamics Simulations | Molecular dynamics (MD) simulations have proven to be useful for predicting and interpreting the conformational and dynamic properties of various polymer-solvent systems. The number of repeat units used to represent a polymer chain in an MD study is intended to provide a balance between the computational demands and the reliability of the specific phenomena being studied. To date, this balance has not received sufficient attention. Here, we investigate how the chain length of an atomistic polymer model influences the structure and dynamics of the polymer in different solvents. Seven different polyvinyl chloride (PVC) models, ranging from 5 to 240 –(CH2CHCl)- repeat units, are studied using atomistic MD simulations in two polar organic solvents: tetrahydrofuran (THF) and dimethylformamide (DMF). After benchmarking our MD results against experimental density data, we calculate polymer end-to-end distances, radii of gyration, radial distribution functions, shape descriptors, end-to-end vector correlation functions, dihedral autocorrelation functions, surface areas, and surface electrostatic potentials. Our MD simulations demonstrate that most of these properties converge when approximately 100-120 repeat units are used to represent PVC, and this convergence behavior is observed in different solvents and at different temperatures. | ||||||
7 | 6 | Varun Gopal and Salman Bin Kashif, University of Minnesota | Overcoming near-sightedness in protein-surface molecular simulations with machine learning lenses | Molecular dynamics simulations are instrumental in understanding atomic-scale phenomena. However, the timescales accessible are smaller than those associated with conformational transitions of many proteins. Machine learning (ML) techniques can help overcome this timescale problem. We demonstrate an ML method for studying the adsorption of model molecules on two surfaces. This method can accelerate material design in fields ranging from water treatment to medical implants. | ||||||
8 | 7 | Praveen Muralikrishnan, University of Minnesota | Investigating the Effects of Arginine on Hydrophobic Interactions: Insights Towards Temperature-Stable Vaccine Design | The storage and distribution of vaccines currently rely on a temperature-controlled supply chain to preserve their effectiveness. However, this reliance poses significant challenges, making the development of thermostable vaccines crucial. Past approaches to enhancing vaccine stability include the addition of excipients. However, these methods are empirical, and the underlying excipient mechanisms in formulation stability remain largely unknown. This study aims to unravel the excipient mechanisms behind increased vaccine stability using molecular dynamics simulations. To understand excipient mechanisms, it is essential to comprehend their impact on hydrophobic interactions, a dominant force in protein folding. A coarse-grained hydrophobic polymer model was utilized to isolate the role of hydrophobic interactions, and the effect of arginine, which has stabilizing and destabilizing effects on proteins, is studied. Results from simulations indicate that arginine enhances the stability of the polymer’s folded state. Furthermore, the arginine-glutamate solution shows higher folded-state polymer stability than solutions containing pure arginine or pure glutamate. Future work will include expanding the study to more complex biopolymers, exploring other excipients, and developing a machine-learning model to predict excipient effects. The findings can potentially guide the design of thermostable vaccines and improve immunization access worldwide. | ||||||
9 | 8 | Montana Carlozo, University of Notre Dame | Bayesian Optimization Formulations for Nonlinear Model Calibration | Nonlinear model calibration is difficult with expensive computational models such as molecular simulations as each evaluation can take weeks and must be evaluated hundreds of times. This makes the cost of existing techniques exceedingly expensive. Gaussian process surrogate model based regression can be used with Bayesian optimization (GPBO) to rapidly regress predictive molecular modeling force field parameters. These force fields represent the intramolecular and intermolecular energies of a system and are useful in process and material design when calibrated effectively. We demonstrate the use of three formulations of GPBO and compare them to a traditional nonlinear regression approach. The first method uses a GP to model the logarithmic error between the simulation and experimental results. The second method uses a GP to emulate the molecular model using an approximation. The third method is the same as the second, except that a sparse grid is used instead of an approximation. Our findings indicate that the second and third methods capture the overall model behavior best for parameter calibration without sacrificing accuracy or computational speed. Future work aims to benchmark performance for the second and third methods to higher dimensions and apply them to real refrigerant force fields to generate data in cases where experiments are difficult. | ||||||
10 | 9 | Daniela Rivera Mirabal, University of California Santa Barbara | Controlling Polymer Structure with Precise Sequence Design | Optimal polymer design is challenging due to the vastness of polymeric chemical and structural possibilities. This work provides atomic-level insight to establish design rules to control polymer structure through precise sequencing. We use sequence-specific polypeptoids as a platform for developing design rules relating chemical sequence and conformation as they possess both the robustness of synthetic polymers and the tunability of polypeptides. Specifically, two model systems are studied to examine changes in the local and global structure of the polypeptoid chains. To explore these effects, we simulate polypeptoid chains in solution to understand how the number and location of the hydrophobic and chiral monomers lead to changes in their structural ensemble. The number and position of chiral centers alters local peptoid helical formation. Furthermore, global polymer structure is controlled by leveraging the role of hydrophobicity. These computational methods will provide a molecular insight into the driving forces for polymer conformation and guide the development of new materials with tunable properties. | ||||||
11 | 10 | Lexin Chen, University of Florida | Protein Retrieval via Integrative Molecular Ensemble (PRIME) through extended similarity indices | Molecular dynamics (MD) is a computer simulation method for analyzing systems' dynamics by integrating Newton’s law of motion. Although the advent of graphical processing units has made microsecond timescale MD simulations a routine, simulation post-processing analysis has not caught up to speed with the increasing size of simulation datasets. Clustering, which groups objects based on structural similarity, is pertinent for finding an ensemble of representative structures, which is key to finding the protein folding pathway. Traditionally, clustering algorithms, such as agglomerative clustering, scale quadratically; this is unfavorable with the increasing dataset size. Recently, we have proposed clustering algorithms and frame-selection methods based on extended continuous similarity indices, leading to linear-scaling workflows, which can predict the number of clusters, and create a diverse selection of representative structures. In this study, we developed a package—Protein Refinement in Molecular Ensembles (PRIME), that consist of tools to determine the native structure using the extended continuous similarity. PRIME was validated to several replica-exchanged systems, flexible protein and protein-peptide, to identify the ensemble representative. PRIME was able to perfectly map all the structural motifs in the studied systems and required unprecedented linear scaling. | ||||||
12 | 11 | John Michael Lane, Mississippi State University | Developing a Bayesian Force Field for Exploring the Reaction Pathways of Furfural Conversion & Hydrogen Dynamics over β-Mo2C | Using conventional ab initio molecular dynamics (AIMD) to sample the free energy surface of reactions over transition metal surfaces is computationally prohibitive due to the numerical complexity of solving the Kohn-Sham equations for conductive materials. In this case study, we begin developing a Bayesian force field based on Gaussian process regression trained on short AIMD trajectories for furfural conversion over the (101) facet of Mo2C in the surface adsorbed and molecular hydrogen. While the current iteration of the force field is unable to model furfural over Mo2C, the model can capture hydrogen dynamics and adsorption of hydrogen over the catalyst surface. With this force field, we observe that hydrogen adsorbs strongly to hollow sites on the catalyst coordinated by 3 molybdenum atoms, and diffusion of hydrogen on the surface only occurs at higher temperatures and does not occur over the carbon atoms. Challenges with training the force field and possible approaches to sample furfural conversion are also discussed. | ||||||
13 | 12 | Rugwed Lokhande, University of Florida | Hierarchical partition of Hilbert space based on excitation and seniority weightage | Full Configuration Interaction (FCI) is the special case of CI where we include all the possible Slater determinants (or configurations) in the variational procedure to obtain the electronic energies. There have been multiple established procedures to reduce these number of Slater determinants in the variational procedure based on excitation (e.g., CID, CISD, CISDT) and seniorities (e.g., DOCI). To achieve the same (and following a recent proposal by Loos et al.), we use a hierarchical parameter ‘h’ which has two weights α1and α2measuring the importance of excitation (e) and seniority (s) contributions to the wavefunctions according to: h= α1*e + α2*s Standard, excitation-level CI (CISD, CISDT, etc.) corresponds to α1=1 and α2=0. Seniority-based CI, on the other hand, corresponds to α1=0 and α2=1. In his work, Loos only considers α1=0.5 and α2=0.25. The question is then: can we find other ways to generate partitions of the Hilbert space? We vary α1 and α2values in the [-1, 1] interval in order to partition the Hilbert space. The key idea is to see how the interplay between excitation and seniority leads to descriptions of systems where dynamic or static correlation is dominant (and, hopefully, identifying a regime that provides a balanced description of both correlation types). We also explore this partitionconcept in the Coupled Cluster (CC)formalism in a more general way. By limiting the outsetofthe seniority sectors accessible via the excitation operators in the CC operator, leads to a new family of CC methods that we call seniority-restricted CC (sr-CC). For instance, the sr-CCSD(0) wavefunction includes all the S excitations, but only the seniority zero doubles.These new CI and CC hierarchies were implemented in our Fanpy package, and tested on model systems. | ||||||
14 | 13 | Hemant Nagar, Ohio University | Determining stable configurations of phosphorylated IRF3 protein dimers using molecular simulation | Protein dimerization is a mechanism through which proteins participate in numerous biological signal pathways. The objective of this work is to determine stable dimeric configurations of proteins via free energy analysis using advanced molecular simulation techniques. Specifically, our focus is the dimerization of interferon regulatory factor 3 (IRF3) protein, which is known to play an important role in the innate immune response. Innate immune system response is the first line of defense when a pathogen invades our body. Many times, viruses can inhibit IRF3 dimerization, or aberrant dimerization of IRF3 can lead to many diseases, including autoimmune disease, cancer, and diabetes. Therefore, identifying the stable dimeric configurations of IRF3 as a function of its phosphorylation state will allow for the development of therapeutics for the treatment of these diseases. Using metadynamics, we have studied the stable dimeric configurations for different phosphorylation states and have identified key interactions that stabilize these configurations. | ||||||
15 | 14 | Wei-Tse Hsu, University of Colorado Boulder | Enhancing configurational sampling, flexibility,and parallelizability of alchemical free energy methods | Over the past decade, alchemical free energy methods have been a popular choice for the computation of solvation free energies and binding free energies, given their ability to connect the end states of interest via nonphysical pathways free from unsurmountable free energy barriers. However, traditional methods like expanded ensemble (EXE) or Hamiltonian replica exchange (HREX) may not be efficient in accurate free energy calculations if the slowest degrees of freedom are largely orthogonal to the alchemical variable. In addition, there exist systems where traversing all alchemical intermediate states is challenging, even if alchemical biases (e.g., in EXE) or coordinate exchanges (e.g., in HREX) are applied. This issue is exacerbated when the state space is multidimensional, which can require seamless communications between hundreds of cores that current parallelization schemes do not fully support. In this study, we present our recent efforts in addressing these two issues separately. First, we propose alchemical metadynamics, which allows explicitly biasing configurational collective variables in alchemical free energy calculations using the metadynamics framework. With test systems with varying complexity, we demonstrate that alchemical metadynamics captures metastable states not necessarily accessible in traditional alchemical free energy methods, thus capable of accurate free energy calculations. For the second issue, we introduce the method of synchronous ensemble of expanded ensembles (EEXE), which periodically exchanges coordinates of different replicas of EXE simulations. We show that this method can compute free energies consistent with the estimates obtained in EXE and HREX, while offering much higher flexibility. Importantly, its planned successor, asynchronous EEXE, performs parallelized simulations asynchronously, which allows close communications between large numbers of adopted processors, and adaptive changes to the parameters of ensembles in response to data collected. These recent efforts can be easily integrated together in ways such as asynchronous ensemble of alchemical metadynamics, promising enhanced flexibility, parallelizability, and configurational sampling for a wider range of systems. | ||||||
16 | 15 | Tanay Debnath, University of Texas at Dallas | Computational Investigation of Unnatural Base Pair (UBP) Incorporated DNA in Solution and in Taq DNA Polymerase | We will present our investigation of the structural and dynamical aspects of UBP-incorporated DNA using AMBER and polarizable AMOEBA force field in MD simulations, both in solution and in DNA polymerase (KlenTaq) system. We have considered D5SICS-dNaM as UBP, one of the most efficiently replicated UBPs, which interacts with one another through hydrophobic interactions and forms an unconventional intercalated base pair. Both the UBs have been parameterized for AMBER and AMOEBA force fields. We have explored both pre-insertion and post-insertion of UBP into the 3’ terminus of a DNA substrate during replication process. RMSD and RMSF values obtained from our simulations suggest that UBP containing DNA is stable in TAC polymerase. We have also analyzed Energy Decomposition Analysis (EDA), correlation matrix, normal modes of the semi-synthetic DNA. It has been found from our analysis that the major stabilizing factor of UBP inside DNA is the dispersion energy which drives them to sliding through one another leading to formation of stacking geometry. Overall, our detailed analysis can predict the structural understanding of UBP incorporated DNA as well as insertion mechanism of UBs in the DNA template during replication process. | ||||||
17 | 16 | Shubham Chatterjee, University of Texas at Dallas | Selecting Reactant-Conformers from MD data via Clustering for QM/MM Studies | K-nearest clustering can be used in QM/MM (Quantum Mechanics/Molecular Mechanics) calculations to select the reactant. This approach clusters the configurations generated from a molecular dynamics simulation based on similarity. The centroid or representative configuration of each cluster can then be selected as the reactant, effectively reducing the complexity and computational cost for subsequent QM/MM calculations. This study used K-nearest clustering to select the reactant for a QM/MM study of enzymatic catalysis by horseradish peroxidase in water. The study also validates the use of K-nearest clustering with confusion matrix and F1-score. Finally, the selected reactant id used for a QM/MM study to approximate the structure of the transition state. | ||||||
18 | 17 | Felipe Perez, University of Oklahoma | Unsupervised Classification Reveals Bias in Simulations of PAHs | Experimental imaging of polycyclic aromatic hydrocarbons (PAHs) has resolved over a hundred molecular geometries. We use imaged structures and classify them in four categories using an unsupervised machine learning approach. Then we review computational studies in the literature and use our classification model to determine the groups they represent, thus we are able to identify the types of PAHs that are largely missing in simulation studies. | ||||||
19 | 18 | Payal Chaudhary, University of Nebraska-Lincoln | Carbon-Based Catalysts for 2-Electron WOR: DFT Analysis | Electrosynthesis of H2O2 is a sustainable and cost-effective alternative for H2O2 production to replace the energy-intensive anthraquinone process. Two-electron water oxidation reaction (2e-WOR) is a promising route that can be combined with the hydrogen evolution reaction. The 2e-WOR has several advantages over the oxygen reduction reaction (ORR) to produce H2O2 - the starting material is only H2O, while the reaction products are simply H2O2 and H2. Also, it is a homogeneous liquid-phase reaction having better mass transfer than the ORR. In this study, we employ density-functional-theory calculations to screen a range of single- and dual-atom metal-nitrogen-graphene structures as catalysts for the 2e-WOR. We perform a thermodynamic analysis to identify chemistries with high activity for H2O2 formation as well as high selectivity over competing 1e- and 4e-WORs. We found that structures based on NiNx–C moieties exhibit promising combinations of activity/selectivity. Some dual-atom structures potentially rival the reported best-performing metal oxides such as ZnO and CaSnO3. Our study demonstrates the potential for low-cost M-N-C systems to efficiently catalyze water oxidation to produce H2O2, which can be validated through experimental studies in the future. | ||||||
20 | 19 | Meghan Osata, UC Irvine | Benchmarking Free Energies with OpenFF sage2.0.0 and sage2.1.0 using Separated Topologies and Relative Hybrid Topology protocols in BACE1 | Binding free energy calculations are employed to estimate the strength of interaction between a compound and a target protein, facilitating the prioritization of compounds for further development in the drug pipeline. To simplify these calculations, we use force fields to estimate the forces and interactions between atoms and molecules. Using both Openforcefield Sage 2.0.0 and newly released Sage 2.1.0 small molecule force fields, and with Separated Topologies (SepTop) method we calculated relative free energies (∆∆G) of a BACE1 ligand set which included scaffold hopping. SepTop performed comparably using both Sage 2.0.0 and Sage 2.1.0 force fields in both calculated ∆∆G and estimated binding free energy (∆G). In this work, we benchmarked the new Sage 2.1.0 force field by evaluating ∆∆G using SepTop. We also found that SepTop is a flexible binding free energy calculation method giving reasonable results irrespective of the small molecule force field used. | ||||||
21 | 20 | Mgcini Keith Phuthi, Carnegie Mellon | Quantitatively accurate predictions of mechanical and surface properties of lithium metal at large length and time scales | The properties of lithium metal are key parameters in the design of lithium ion and lithium metal batteries. They are difficult to probe experimentally due to the high reactivity and low melting point of lithium as well as the microscopic scales at which lithium exists in batteries where it is found to have enhanced strength, with implications for dendrite suppression strategies. Computationally, there is a lack of empirical potentials that are consistently quantitatively accurate across all properties and ab-initio calculations are too costly. In this work, we train Machine Learning Interaction Potentials (MLIPs) on Density Functional Theory (DFT) data to state-of-the-art accuracy in reproducing experimental and ab-initio results across a wide range of simulations at large length and time scales. We accurately predict thermodynamic properties, phonon spectra, temperature dependence of elastic constants and various surface properties inaccessible using DFT. We establish that there exists a Bell-Evans-Polanyi relation correlating the self-adsorption energy and the minimum surface diffusion barrier for high Miller index facets. | ||||||
22 | ||||||||||
23 | ||||||||||
24 | ||||||||||
25 | ||||||||||
26 | ||||||||||
27 | ||||||||||
28 | ||||||||||
29 | ||||||||||
30 | ||||||||||
31 | ||||||||||
32 | ||||||||||
33 | ||||||||||
34 | ||||||||||
35 | ||||||||||
36 | ||||||||||
37 | ||||||||||
38 | ||||||||||
39 | ||||||||||
40 | ||||||||||
41 | ||||||||||
42 | ||||||||||
43 | ||||||||||
44 | ||||||||||
45 | ||||||||||
46 | ||||||||||
47 | ||||||||||
48 | ||||||||||
49 | ||||||||||
50 | ||||||||||
51 | ||||||||||
52 | ||||||||||
53 | ||||||||||
54 | ||||||||||
55 | ||||||||||
56 | ||||||||||
57 | ||||||||||
58 | ||||||||||
59 | ||||||||||
60 | ||||||||||
61 | ||||||||||
62 | ||||||||||
63 | ||||||||||
64 | ||||||||||
65 | ||||||||||
66 | ||||||||||
67 | ||||||||||
68 | ||||||||||
69 | ||||||||||
70 | ||||||||||
71 | ||||||||||
72 | ||||||||||
73 | ||||||||||
74 | ||||||||||
75 | ||||||||||
76 | ||||||||||
77 | ||||||||||
78 | ||||||||||
79 | ||||||||||
80 | ||||||||||
81 | ||||||||||
82 | ||||||||||
83 | ||||||||||
84 | ||||||||||
85 | ||||||||||
86 | ||||||||||
87 | ||||||||||
88 | ||||||||||
89 | ||||||||||
90 | ||||||||||
91 | ||||||||||
92 | ||||||||||
93 | ||||||||||
94 | ||||||||||
95 | ||||||||||
96 | ||||||||||
97 | ||||||||||
98 | ||||||||||
99 | ||||||||||
100 |