Session ThOC. There are 4 abstracts in this session.



Session: SPEED (BEYOND FOURIER) AND SENSITIVITY, time: 8:30 - 8:45 am

Using Deep Neural Networks to Reconstruct and Analyse Non-Uniformly Sampled NMR Spectra


D. Flemming Hansen
ISMB, Univ. College London, London, United Kingdom

Non-uniform and sparse sampling of multi-dimensional NMR spectra has over the last decade become an important tool to allow for fast acquisition of multi-dimensional NMR spectra with high resolution. Also over the last decade, deep neural networks and artificial intelligence have seen new applications in an enormous range of sciences, including analysis of MRI spectra. A proof-of-principle and deep neural networks trained to reconstruct sparsely sampled 2D and 3D NMR spectra will be presented. For reconstruction of two-dimensional NMR spectra, deep neural network performs as well, if not better than, the currently used techniques. It is anticipated that deep learning provides a very valuable tool for the analysis of sparsely sampled NMR spectra in the near future to come.


Session: SPEED (BEYOND FOURIER) AND SENSITIVITY, time: 9:20 - 9:35 am

Structure elucidation from 1D spectra : how far can we go?


Eric Jonas
University of Chicago, Chicago, IL

Contemporary structure-elucidation systems are powerful but depend on multidimensional NMR data. Understanding both low-concentration analytes and mixtures would be enhanced if 1D structure elucidation were viable. But how unique are the information in 1D spectra, and can we determine molecular structure from 1D data alone? Here we describe elucidation techniques only using 1D spectra. We use machine learning to perform “imitation learning", where we learn to imitate an oracle which always builds the correct structure. We can reliably predict the structure of over 50% of candidate molecules confidently with 90% accuracy using 13C 1D spectra.  We describe ongoing work on extending our methods to incorporate rapid 1H spectra, geometry, and extending the validation process to all data in PubChem.


Session: SPEED (BEYOND FOURIER) AND SENSITIVITY, time: 9:35 - 9:50 am

NMRFx:  New Advances in an Integrated Cross-Platform NMR Software Application


Martha Beckwith; Teddy Cohen; Ellen Koag; Audrigue Jean-Louis; Bruce Johnson
CUNY Advanced Science Research Ctr., New York, NY

NMRFx is an integrated suite of three cross-platform programs for the analysis of NMR data.  NMRFx Processor is a processing program for full signal processing of 1D and multi-dimensional NMR data.  NMRFx Structure is a molecular structure program that can generate three-dimensional structures based on input constraints and can predict NMR chemical shifts.  NMRFx Analyst is a new NMR visualization and analysis program and has access to both the processing and structure analysis features of the other two components.  In this presentation we describe recent developments in NMRFx that include new processing tools for NUS datasets, improved chemical shift prediction tools, and new tools for following titrations of ligands and pressure induced changes.


Session: SPEED (BEYOND FOURIER) AND SENSITIVITY, time: 9:50 - 10:05 am

SCRUFI: A Matlab toolbox for the analysis of INADEQUATE spectra


David Bennett; Iain Day
University of Sussex, Brighton, United Kingdom

Determining the heavy atom connectivity is the one of the key steps for structure determination. In principle, despite its inherently poor sensitivity, the INADEQUATE experiment gives these connectivities directly. 

The automation of spectral analysis allows correlations to be identified. Parameterisation of the peak shape allows an automated search over all possible expected correlations. Correlations can then be classified by comparison with the  spectral noise floor. Ultimately a connectivity diagram can be produced directly from the spectrum.

Here we present SCRUFI (Skeletal Connectivity Revealed Using Fitted Inadequate), a Matlab toolbox and utility designed to automate the analysis of INADEQUATE spectra. We achieve this by a modern implementation of the approach of Dunkel.