Session PD. There are 23 abstracts in this session.

Session: Computational NMR, poster number: 163
Towards Unbiased and More Efficient NMR Based Structure Elucidation: a Powerful Combination of CASE Algorithms and DFT Calculations
Alexei Buevich1; Mikhail Elyashberg2
1Merck and Co., Kenilworth, NJ; 2Advanced Chemistry Development (ACD/Labs), Moscow, Russia

Computer‐assisted structure elucidation (CASE) is composed of two steps: (a) generation of all possible structural isomers for a given molecular formula and 1D and 2D NMR data (COSY, HSQC, HMBC) and (b) selection of the most probable isomer based on empirical chemical shift predictions. This method has been very successful in solving structural problems of small organic molecules and natural products. However, CASE applications are generally limited to structural isomer problems and can sometimes be inconclusive due to insufficient accuracy of empirical shift predictions. Here, we report a synergistic combination of CASE algorithms and DFT calculations, which can be used in situations when CASE programs fail. This approach allows not only determining the correct structure but its relative stereochemistry too.


Session: Computational NMR, poster number: 164
Investigating Y2(Ti,Sn)2O7 ceramics using Site Occupancy Disorder and NMR spectroscopy
Robert Moran1; Arantxa Fernandes1; David McKay1; Paulynne C. Tornstrom2; Ricardo Grau-Crespo2; Sharon E. Ashbrook1
1University of St Andrews, St Andrews, United Kingdom; 2University of Reading, Reading, United Kingdom
Building on previous studies of the Y2(Ti,Sn)2O7 pyrochlores, we use first-principles calculations to investigate local structure and the effect of B-site cation mixing on calculated solid-state NMR parameters. Using different approaches to generate possible structural models for the disordered Y2(Ti,Sn)2O7 ceramic, ranging from simple models where individual atoms, or a combination of atoms were substituted, to more complex approaches such as the Site Occupancy Disorder (SOD) technique, which allows every symmetry-unique arrangement of atoms for a given composition to be determined, as well as the corresponding configurational degeneracy for each of these structures. Comparing to experimental NMR spectra allows us to deduce which of the SOD-generated structures contribute, providing more insight into cation disorder in Y2(Ti,Sn)2O7 pyrochlores.

Session: Computational NMR, poster number: 165
A new distance-geometry based optimization in parts for improved NMR structure solution
Niladri Ranjan Das1; Kunal Narayan Chaudhury2; Debnath Pal3
1PhD student, Indian Institute of Science, Bangalore, India; 2Assistant Professor, Indian Institute of Science, Bangalore, India; 3Professor, Indian Institute of Science, Bangalore, India
The main challenge in protein conformation using NMR is the paucity of distance-geometry information. We address this using an approach that leverages the available experimentally-derived distance constraints and the canonical covalent-bond geometry information. We individually model small structural parts with adequate experimental constraints, which are thereafter combined in one-shot using modern optimization techniques. This yields incomplete structure further post-processed for modeling, respecting the distance bounds. The result is an ensemble of conformations varying only at the post-processed regions. Our method is scalable, robust to data deletion, and does not require multiple starting structures. Importantly, we are able to bypass the pitfalls of potential energy based optimization. Our benchmarked results on proteins with 84-299 residues show satisfactory structure-solution.

Session: Computational NMR, poster number: 166
AUTOMAP Image Reconstruction — Improved Low Signal-to-Noise MR Data at ultra-low field
Neha Koonjoo1, 2; Bo Zhu1, 2; Matthew S. Rosen1, 2
1A.A. Martinos Biomedical Imaging Center/ MGH / HMS, Boston, MA; 2Department of Physics, Harvard University, Cambridge, MA
Due to very low Boltzmann polarization, MR images acquired at ultra-low field (ULF), MR images are mostly corrupted with noise, thus resulting in low signal-to-noise. In the aim of improving image quality at ULF, we apply the deep neural network image reconstruction technique, AUTOMAP, to low SNR datasets acquired at 6.5 mT. The performance of AUTOMAP (Automated Transform by Manifold Approximation) versus the conventional Inverse Fast Fourier Transform (IFFT) on this data was evaluated. The results for AUTOMAP reconstruction show a significant noise reduction, leading to more than 30% gain in signal to noise ratio as compared to standard IFFT.

Session: Computational NMR, poster number: 167
Deep learning MR reconstruction with Automated Transform by Manifold Approximation (AUTOMAP) in real-world acquisitions with imperfect training
Bo Zhu; Berkin Bilgic; Congyu Liao; Bruce Rosen; Matthew Rosen
MGH / Martinos Center for Biomedical Imaging, Cambridge, MA
Automated Transform by Manifold Approximation (AUTOMAP) is a generalized MR image reconstruction framework based on supervised manifold learning and universal function approximation implemented with a deep neural network architecture. Here we investigate the effect of significant sampling trajectory error in spiral acquisitions, where mismatch between training and runtime scanner trajectories may result in unpredictable reconstruction artifacts. We demonstrate through Monte Carlo analysis that the error in AUTOMAP reconstruction increases smoothly as a function of trajectory error, demonstrating reasonable robustness to trajectory deviation. We find these simulation results are consistent with reconstruction performance on real scanner data acquired from human subjects.

Session: Computational NMR, poster number: 168
Learning multichannel coil combination with Automated Transform by Manifold Approximation (AUTOMAP) using complex-valued neural networks
Bo Zhu; Stephen Cauley; Bruce Rosen; Matthew Rosen
MGH / Martinos Center for Biomedical Imaging, Cambridge, MA
End-to-end learning of the image reconstruction domain transform with AUTOMAP (Automated Transform by Manifold Approximation) has been demonstrated on a variety of spatial encoding strategies previously limited to single-channel data. We extend this framework to learning reconstruction of highly undersampled multichannel k-space data solely from pairs of multichannel k-space and image training data without employing conventional parallel imaging formulations such as SENSE or GRAPPA, and show improved RMSE and artifact reduction with the trained AUTOMAP reconstruction network.

Session: Computational NMR, poster number: 169
AUTOmated pulse SEQuence generation (AUTOSEQ) using Bayesian reinforcement learning in an MRI physics simulation environment
Bo Zhu1; Jeremiah Liu2; Neha Koonjoo1; Bruce Rosen1; Matthew Rosen1
1MGH / Martinos Center for Biomedical Imaging, Cambridge, MA; 2Dept. of Biostatistics, Harvard University, Boston, MA
Although the macroscopic equations of motion for nuclear magnetic resonance have been described and modeled for decades by the Bloch equations, limited human intuition of their nonlinear dynamics is an obstacle to fully exploiting the vast parameter space of MR pulse sequences. Here we recast the general problem of pulse sequence development as a game of perfect information, and propose an approach to optimize game play with a Bayesian derivative of reinforcement learning within a MRI physics simulation environment. We demonstrate an AI agent learning a canonical pulse sequence (gradient echo) and generating non-intuitive pulse sequences approximating Fourier spatial encoding.

Session: Computational NMR, poster number: 170
Automated Discrimination and Verification of Cooking Oils with Portable Low-Field NMR
Mason Greer; Cheng Chen; Soumyajit Mandal
Case Western Reserve University, Cleveland, OH
Food and drug adulteration and contamination is a large concern today.  Because of this, there exists a need for an automated, low-cost tool to detect such adulteration or contamination. One product that is of concern is cooking oils.  We describe a system (both hardware and software) for the automatic detection of adulterated material in food and drug products.  We test the feasibility of such a system with two custom low-field NMR sensors and support vector machine for classification.  With a hand-held device, we classify canola, corn, olive, and vegetable oil with an 80% accuracy, with most of the misclassifications coming from vegetable and corn oils. With a 0.5T desktop device, we classify the same oils with 99.5% accuracy.

Session: Computational NMR, poster number: 171
Enhanced In Vivo Human Brain MR Imaging using Automated Transform by Manifold Approximation (AUTOMAP) in varying SNR regimes
Neha Koonjoo1, 2; Bo Zhu1, 2; Matthew Christensen1; John E. Kirsch1; Matthew Rosen1, 2
1A.A. Martinos Biomedical Imaging Center/ MGH / HMS, Boston, MA; 2Department of Physics, Harvard University, Cambridge, MA
Scan times in clinical MR imaging are often long and costly compared with other imaging modalities Long acquisition times are needed to attain highly resolved images with good signal-to-noise ratio (SNR). In the aim of improving image quality and reconstruction efficiency, we apply the deep neural network image reconstruction technique, AUTOMAP, to varying SNR datasets acquired on a human subject at 1.5 T. The performance of AUTOMAP (Automated Transform by Manifold Approximation) versus the conventional Inverse Fast Fourier Transform (IFFT) on this in vivo brain data was evaluated. The results for AUTOMAP reconstruction show a significant noise reduction, leading to more than 35% gain in SNR as compared to standard IFFT.

Session: Computational NMR, poster number: 172
An NMR-Guided Screening Method for Selective Fragment Docking and Synthesis of a Warhead Inhibitor
Daniel Morris
The University of Akron, Akron, Ohio
Selective hits for the glutaredoxin orthologs from Brucella melitensis and Pseudomonas aeruginosa were determined using STD NMR and verified by trNOE and 15N HSQC titration. The most promising hits were docked into the target proteins using a chemical shift perturbation scoring function, SHIFTS, to compare simulated poses to experimental data. After elucidating possible poses, the hits were further optimized into lead compounds by extension with an electrophilic acrylamide warhead. We believe that focusing on selectivity in this early stage of drug discovery will limit cross-reactivity that might occur with the human ortholog as the lead compound is optimized.

Session: Computational NMR, poster number: 173
Evaluation of 13C Spectra of Polystyrene Sulfonate Using Experimental and Computational NMR Methods
Akshar P. Gupta; Andrew B. Jackson
Solvay Technology Solutions, Stamford, CT
Density-functional theory (DFT) calculations of magnetic shielding in NMR provide significant contributions in understanding experimental chemical shifts. 13C resonance assignments and spectral interpretation of polystyrene sulfonates (PSS) are challenging due to overlap in the aromatic region. This study focuses on the 13C characterization of PSS with varying sulfonation fraction. DFT measurements and empirical predictions were performed on monomeric models by evaluating the 13C chemical shifts of structures at the ωB97X-D/6-31G* level of theory. Structures further optimized to include a sodium counterion and explicit water molecule(s) revealed key hydrogen bonding interactions. Quantum mechanical shifts were found to be consistent with experimental results. DFT calculations coupled with experimental 2D NMR measurements identified the unique para-C-SO3Na 13C resonance used in estimating sulfonation fraction.

Session: Computational NMR, poster number: 174
Accurate Measurement and Calculation of Chemical Shift Tensors in the Carbohydrate Recognition Domain of Galectin-3C by MAS NMR and QM/MM
Jodi Kraus
University of Delaware, Newark, DE
NMR chemical shift tensors (CST) sensitive probes of atomic level structure and environment in biomolecules. Experimental and calculated CSTs have promising applications for protein structure determination and refinement. We present here experimental isotropic chemical shifts and CSA’s from the carbohydrate recognition domain of Galectin-3C, obtained from R-symmetry based recoupling experiments. We also present a systematic investigation of calculated 13Ca and 15NH CSA tensors from Galectin-3C using hybrid quantum mechanics/molecular mechanics (QM/MM). Excellent agreement between experimental and computed shifts is obtained for 13Ca while larger scatter is observed for 15NH chemical shifts, which are influenced to greater extent by electrostatic interactions, hydrogen bonding, and solvation. Nevertheless, the agreement between 15NH experiment and calculation has been improved relative to previous benchmarking studies.

Session: Computational NMR, poster number: 175

Prediction of Disulfide Bond Connectivity and Configuration Based on Chemical Shifts


David Armstrong1; Quentin Kaas2; K.Johan Rosengren1
1Faculty of Medicine, The University of Queensland, Brisbane, Australia; 2Institute for Molecular Biosciences, Brisbane, Australia

Disulfide rich peptides are a ubiquitous class of peptides that have proven to be promising drug candidates in a number of fields due to their exquisite selectivity and potency. The cystine residues that cross link the backbone are essential for both the activity and structure of the peptide. 2D NMR spectroscopy is ideally suited for structure determination of disulfide rich peptides because of their small size and constrained nature. A major limitation using 2D NMR spectroscopy is that for native peptides neither the connectivity or cystine configuration can be directly resolved, leading to incorrect structure and activity calculations. We have developed novel machine learning methods that can predict both cystine configuration and connectivity based on chemical shift and structural inputs.


Session: Computational NMR, poster number: 176

Improvement of Protein Atomic Contacts in Xplor-NIH


Guillermo A. Bermejo; G. Marius Clore; Charles D. Schwieters
National Institutes of Health, Bethesda, MD

Xplor-NIH is a popular software package for biomolecular structure determination by NMR and other data sources. Recent work has aimed at improving protein conformation in terms of backbone Ramachandran and side chain rotamer criteria. Here, we focus on the quality of steric contacts in proteins. We present an improved force field, with nonbonded interactions based on the atomic radii of the MolProbity validation program. Use of the customary atomic repulsion energy term required the optimization of an overall radius scale factor, resulting in structures with significantly fewer atomic clashes and similar accuracy (as judged by residual dipolar coupling cross-validation), relative to those produced by the previous Xplor-NIH force field.


Session: Computational NMR, poster number: 177
NMR-STAR: Comprehensive data model for representing, archiving and exchanging NMR data of various kinds
Kumaran Baskaran; Eldon L Ulrich; Pedro R Romero; Dimitri Maziuk; Jon Wedell; Hongyang Yao; John L Markley
University of Wisconsin Madison, Madison, WI
The NMR-STAR ontology is designed to support the archiving and exchange of quantitative data and associated metadata generated from nuclear magnetic resonance (NMR) spectroscopic studies of biological systems. The NMR-STAR architecture is based on the concept of an object-relational data model implemented with the Self-defining Text Archival and Retrieval (STAR) specifications published by Hall and coworkers. The BMRB ADIT-NMR deposition system, the wwPDB’s OneDep deposition system, the tools used for the annotation and validation of BMRB entries, the export format for BMRB data, the construction of the working relational database at BMRB, and the BMRB query systems are all supported by the NMR-STAR ontology. 

Session: Computational NMR, poster number: 178
Equivalence of Floquet-Magnus and Fer Expansions and the Investigation of Spin dynamics in the Tree-Level System
Eugene Mananga
The City University of New York, New York City, NY
We investigated the orders to which the Floquet−Magnus expansion (FME) and Fer expansion (FE) are equivalent or different for the three-level system. Specifically, we performed the third-order calculations of both approaches based on elegant integrations formalism. As the propagator from the FME takes the form of the evolution operator, which removes the constraint of a stroboscopic observation, we appreciated the effects of time-evolution under Hamiltonians with different orders separately. Our work unifies and generalizes existing results of Floquet−Magnus and Fer approaches and delivers illustrations of novel springs that boost previous applications that are based on the classical information. This promising work is expected to play an important role for recording high resolution spectra, probing structural and dynamic information in biomolecules.

Session: Computational NMR, poster number: 179
Calculation, modification, and deposition of atomic force field parameters in molecular modeling of small molecules
Hesam Dashti1; Jonathan Wedell2; Gabriel Cornilescu3; Charles Schwieters4; William Westler5; Eldon L. Ulrich1; John L. Markley1; Hamid Eghbalnia6
1University of Wisconsin-Madison, Madison, WI; 2BioMagResBank, Monona, WI; 3University of Wisconsin - Madison, Madison, WI; 4National Institutes of Health, Bethesda, MD; 5Univ. of Wisconsin-Madison, Madison, WI; 6University of Wisconsin, Madison, WI
Computational molecular modeling has been a gateway to structural investigations of small molecules and studies of molecular interactions. We introduce a pipeline for reproducible generation of force field parameters necessary for molecular modeling of small molecules. In addition to producing these parameters, the pipeline provides tools for seamless modifications and adjustment of the parameters to be used in the well-known molecular modeling software package Xplor-NIH. Because of the run time expense for producing the force field parameters, and changes that arise from improvements to force fields, it is critical to provide a public database for deposition and archiving the molecular modeling parameter files. We have designed a data structure and introduced the first molecular modeling database of small molecule.

Session: Computational NMR, poster number: 180
Experiment Visualization - SpinDrops 2.0

Michael Tesch; Niklas Glaser; Bálint Koczor; Steffen Glaser
TUM, Munich, Germany
SpinDrops is a tool for interactively visualizing coupled spin
system evolution using the DROPS representation. It has uses in
teaching NMR, pulse sequence development and comprehension. Version
2.0 adds illustrative and exploratory tools: new propagator views,
non-ideal pulses, effective Hamiltonians, additional basis sets,
more powerful pulse sequence definitions, and illustrative views of
phase cycling and shaped pulses. The program is provided freely to
the community, and efforts are underway to grow an ecosystem
around related teaching and pulse development tools. SpinDrops 2.0
is now available to users of nearly any modern computer, including
those running macOS, Android, iOS, Windows, and directly in web
browsers supporting WebGL at https://spindrops.org.


Session: Computational NMR, poster number: 181
A Hybrid Approach for Protein Structure Determination Combining Sparse NMR Data with Evolutionary Coupling Sequence Data
Yuanpeng Janet Huang1; Kelly Brock2; Yojiro Ishida1; Gaohua Liu3; Yuefeng Tang1; G.V.T. Swapna1; Masayori Inouye1; Chris Sander2; Debora Marks2; Gaetano T. Montelione1
1Rutgers, The State University of New Jersey, Piscataway, NJ; 2Harvard Medical School, Boston , MA; 3Nexomics Biosciences, Inc, Bordentown, NJ
The massive increase in evolutionary sequence information coupled with maximum likelihood covariance analysis now provides a rich source of structural constraints for molecular modeling. Exploiting this synergy, we have developed an automated approach that uses sequence co-variation data (ECs), obtained from multiple sequence alignments of protein families, together with sparse NMR data to determine 3D protein structures. We demonstrate this hybrid “EC-NMR” method by determining accurate structures of twelve soluble proteins, and one integral membrane protein, ranging in size from 64 to 370 residues. The requirement that ECs are consistent with NMR data recorded on a specific member of a protein family, under specific conditions, also allows identification of ECs that reflect alternative allosteric or excited states of the protein.

Session: Computational NMR, poster number: 182
NMRbox: National Center for Biomolecular NMR Data Processing and Analysis
Mark W. Maciejewski1; Kumaran Baskaran2; Mosrur Chowdhury1; Hesam Dashti2; Frank Delaglio3, 4; Hamid Eghbalnia2; Oksana Gorbatyuk1; Michael R. Gryk1, 5; Dmitry Maziuk2; Ion Moraru1; Pedro Romero2; Adam Schuyler1; Eldon L. Ulrich2; Gerard Weatherby1; Jonathan Wedell2; Michael Wilson1; Terry Wright1; Jeffrey C. Hoch1
1UConn Health, Farmington, CT; 2University of Wisconsin-Madison, Madison, WI; 3University of Maryland, Rockville, MD; 4National Institute of Standards and Technology, Rockville, MD; 5University of Illinois, Urbana-Champaign, IL
NMRbox is a free resource for biomolecular-NMR computation. NMRbox provides a virtual machine (downloadable and cloud-based Platform-as-a-Service) provisioned with over a hundred packages. The NMRbox.org website hosts a software registry to simplify software discovery and provide access to documentation, publications, and usage statistics. The NMRbox PaaS provides high-performance compute and secure data storage services. In addition to simplifying the complex software environment needed for biomolecular NMR, NMRbox fosters the development of meta-software packages that rely on the cooperation of multiple packages. We describe resources recently deployed in NMRbox, including free access to some commercial NMR software, GPU acceleration, and additional computational and storage capacity. NMRbox is a Biomedical Technology Research Resource supported by NIH/NIGMS. Visit the NMRbox booth during ENC.

Session: Computational NMR, poster number: 183
Quality Assessment of Biomolecular NMR Structures
Roberto Tejero1; Yuanpeng Janet Huang2; Antonio Rosato3; Gaetano T. Montelione2
1Universidad de Valencia, Valencia, Spain; 2Rutgers, The State University of New Jersey, Piscataway, NJ; 3CERM - University of Florence, Florance, Italy
Biomolecular NMR structures are now routinely used in biology, chemistry, and bioinformatics. Methods and metrics for assessing the accuracy and precision of NMR structures are beginning to be standardized across the biological NMR community. These include both knowledge-based assessment metrics, parameterized from the database of biomolecular structures, and model-versus-data assessment metrics. On line servers are available that provide comprehensive protein structure quality assessment reports, and efforts are in progress by the world-wide Protein Data Bank (wwPDB) to develop a biomolecular NMR structure quality assessment pipeline as part of the structure deposition process. These quality assessment metrics and standards will aid in determining more accurate structures, and increase the value and utility of these structures for the broader scientific community.

Session: Computational NMR, poster number: 184
Predicting Structure in a Natural Product from One-Bond 13C-13C Scalar Couplings
Shannon Sestile; Jim Harper
University of Central Florida, Orlando, FL
This study explores feasibility of conformationally characterizing an unknown natural product using only experimental/theoretical one-bond 13C-13C scalar couplings (1JCC). Analysis focuses on the newly-discovered depsidone natural product, perisalazinic acid. Seventy-five conformationally-varied structures were evaluated and 1JCC values were computed (B3LYP/EPR-III). A best-fit structure was selected by comparison to 14 experimental values from INADEQUATE to find agreement. Perisalazinic acid has seven conformationally-variable moieties. Five were structurally determined in a single orientation. A sixth position was reduced to three conformations from 18 possibilities. The final position was reduced to two candidates. A weighted average of each conformation at both positions is presumed. This work illustrates the first realization of a complete three-dimensional structural characterization of an unknown using only theoretical/experimental 1JCC values.

Session: Computational NMR, poster number: 185
Reconstruction of Randomly Sampled Quadrature Components using SMILE
Jinfa Ying2; Frank Delaglio1; Dennis Torchia2; Ad Bax2
1NIST IBBR, Rockville, MD; 2NIH, Bethesda, MD
Non-Uniform Sampling (NUS) is increasingly familiar in biomolecular NMR, with most approaches sampling all real and imaginary components for a given time increment. Maciejewski and coworkers suggested subsampling of individual real or imaginary components within an increment (Random Phase Detection, RPD), potentially allowing a 50% reduction in measurement time per indirect dimension. Bostock, Holand, and Nietlispach demonstrated an extension of this concept suitable for gradient enhanced data (Random Quadrature Detection, RQD) where the measured components must be recombined to generate real and imaginary data, a complication when one or the other component is missing. We demonstrate a fast and effective extension of the SMILE approach for reconstruction of RQD data, and outline software tools for implementing RQD pulse sequences.