Session MOD. There are 5 abstracts in this session.

Session: Computation: Beyond Fourier (aka NUSCON @ ENC), time: 4:00-4:25
Non-uniform sampling sampling methods in structural studies of peptides and proteins
Mehdi Mobli; Tomas Miljenovic; Xinying Jia
Centre for Advanced Imaging, St Lucia, Australia
Advances in the development of non-Fourier spectral reconstruction methods have enabled design of new non-linear sampling methods that can drastically reduce experimental time, increasing the range of multidimensional experiments that can be acquired in feasible timeframe. Selection of appropriate sampling and reconstruction parameters remain a challenge for the non-expert. NUSCON is a competition designed to engage the community and work towards common benchmarks. Here I will in addition to sharing our results from NUSCON also present work from our group on development of NUS methods and their applications in structural characterisation of peptides and proteins. The long-term goal of our research is to automate protein structure determination by NMR spectroscopy.

Session: Computation: Beyond Fourier (aka NUSCON @ ENC), time: 4:25-4:50
XLSY: Reconstructing Very Large NMR Spectra
Vladislav Orekhov; Yulia Pustovalova; Maxim Mayzel
Gothenburg University, Goethenburg, Sweden
NMR studies of intrinsically disordered proteins and other complex biomolecular systems require spectra with the highest resolution and dimensionality. We present Extra-Large Spectroscopy (XLSY) an efficient approach for acquisition, processing, and handling of very large spectra. XLSY uses a combination of the radial and non-uniform sampling as well as a novel signal processing algorithm for spectrum reconstruction and statistical validation. The methodology is demonstrated by the first-time high-quality reconstructions of a full 7D HNCOCACONH, 5D HACACONH, and 5D HN(CA)CONH spectra of an intrinsically disordered protein α-synuclein. The 7D spectrum allowed us to resolve peaks from the EEG sequence repeats that fully overlap in the 5D spectra and cannot be resolved in any radially sampled 2D projections.

Session: Computation: Beyond Fourier (aka NUSCON @ ENC), time: 4:50-5:05
Monte Carlo Simulations of NMR Data Acquisition and Processing: Implications for Non-Uniform Sampling
Manpreet Kaler; Len Mueller
University of California, Riverside, Riverside, CA
Monte Carlo simulations offer unbiased quantification of spectral information in NMR signals, both as acquired in the time domain or as processed and displayed in the frequency domain. The advantage of the Monte Carlo approach is that no assumptions need be made regarding the nature of correlated noise that is typically introduced during processing. Here we apply this technique to various NMR acquisition and processing schemes, including non-uniform sampling in indirect dimensions. Notably, the purported advantages of NUS in terms of sensitivity and resolution gains are to a large extent borne out in these simulations, particularly in the case of multiple, closely spaced resonances.

Session: Computation: Beyond Fourier (aka NUSCON @ ENC), time: 5:05-5:20
Toward machines that can learn to interpret NMR spectra
Hamid R. Eghbalnia; Hesam Dashti; John L. Markley
University of Wisconsin-Madison, Madison, WI
Analysis of NMR experiments begins with obtaining a spectral representation for the time series data. This representation is subsequently interpreted by interrogating peak positions, heights, shapes, and other parameters. Successful applications of machine learning to other areas of science suggests the possibility that NMR analysis may also benefit from its use. However, machine learning is typically used to “learn” a function, while spectral representation is obtained through applying an operator. For regularly sampled data, the Fourier transform is the commonly used operator, while non-uniformly sampled data (NUS) is transformed through an algorithmically implemented operator. Here, we provide preliminary computational evidence that machines can learn the transform operator – potentially enabling automated NMR data interpretation of regular or NUS time series.

Session: Computation: Beyond Fourier (aka NUSCON @ ENC), time: 5:20-5:35
Back to Bayes'ics: NUS reconstruction as statistical inference
Bradley Worley; Michael Nilges; Therese Malliavin
Institut Pasteur, Paris, France
The task of recovering complete NMR spectra from non-uniformly sampled (NUS) data has always fallen to optimization methods, the most well-known of which are MaxEnt and IST. Both methods offer powerful optimality guarantees, but only yield point estimates with little consideration given to the statistical nature of the problem. We describe new methods of compressed sensing (CS) recovery that directly treat spectral estimation as a Bayesian inference problem, which is then approximately solved by optimization. One advantage of these methods is their ability to return posterior predictive uncertainty estimates, enabling direct discrimination between true signals and artefacts in recovered spectra. We also compare our methods against the state of the art on measured NMR data.