Abstract:Structural elucidation is an important and challenging stage in the discovery of new organic molecules. Single-crystal X-ray analysis provides the most unquestionable results, though in practice the availability of suitable crystals limits its broad use. On the other hand, NMR spectroscopy has become the leading and universal technique to accomplish the task. Despite continuous advances in the field, the misinterpretation of NMR data is commonplace, evidenced by the large number of erroneous structures being published in top journals. Quantum calculations of NMR chemical shifts and scalar coupling constants emerged as ideal complements to facilitate the elucidation process when experimental NMR data is inconclusive. Since seminal reports demonstrated that affordable DFT methods provide NMR predictions accurate enough to differentiate among closely related isomers, the discipline has experienced substantial growth. The impact has been felt in different areas, and nowadays the results of such calculations are routinely seen in high impact literature.This Account describes our investigations in the field of quantum NMR calculations, focusing on the development of tools for structural elucidation and practical applications. We pioneered the use of artificial intelligence methods in the development of novel strategies of structural validation. Our first generation of trained artificial neural networks (ANNs) showed excellent ability to identify mistakes at the atom connectivity level, whereas the use of multidimensional pattern recognition pushed the performance to the stereochemical limit. In a conceptually different approach, we developed DP4+, an updated version of the DP4 probability used to determine the most likely structure among two or more candidates when one set of experimental data is available. Increasing the level of theory in NMR calculations and including unscaled data in the formalism improved the performance of the method, further validated to settle the configuration of challenging motifs such as spiroepoxides or Mosher's derivatives. One of the limitations of DP4+ is related to the relatively large computational cost involved in obtaining DFT-optimized geometries, which led to the development of a fast variant including the valuable information provided by coupling constants (J-DP4 method).These tools were explored to suggest the most probable structure of controversial natural or unnatural products originally misassigned, with some predictions further validated by synthesis (as in the case of pseudorubriflordilactone B). The possibility of predicting the structure of a natural product without requiring authentic sample was investigated in collaboration with Prof. Pilli (UNICAMP, Brazil) in the computer-guided total synthesis and stereochemical revisions of several natural products. Despite these advances, there remain considerable challenges, such as the case of configurational assessment of polar systems featuring multiple intramolecular hydrogen bonding interactions because of the poor energy predictions provided by most DFT methods. In our latest work, we tackle this problem by averaging the results provided by randomly generated ensembles, paving the way for a new paradigm in quantum NMR-assisted structural elucidation.

Improved Prediction of Carbonless NMR Spectra by the Machine Learning of Theoretical and Fragment Descriptors for Environmental Mixture Analysis

Machine Learning Spectroscopy Based on Group Contribution and Molecule Contribution Methods

Multiplet analysis by strong-coupling-artifact-suppression 2D <i>J</i>-resolved NMR spectroscopy

Highly Accurate Prediction of NMR Chemical Shifts from Low-Level Quantum Mechanics Calculations Using Machine Learning

Improving Efficiency of Measuring Individual 1H Coupling Networks by Pure Shift 2D J-resolved NMR Spectroscopy.

High-resolution Two-Dimensional J-resolved NMR Spectroscopy for Biological Systems.

Fully Exploiting the Power of 2D NMR J-resolved Spectroscopy.

TransPeakNet: Solvent-Aware 2D NMR Prediction via Multi-Task Pre-Training and Unsupervised Learning

Machine Learning Assisted Interpretation of 2D Solid-State Nuclear Magnetic Resonance Spectra.

Elucidating Structures of Complex Organic Compounds Using a Machine Learning Model Based on the 13C NMR Chemical Shifts

General Two-Dimensional Absorption-Mode J-Resolved NMR Spectroscopy

Real-time prediction of 1H and 13C chemical shifts with DFT accuracy using a 3D graph neural network

The development of machine learning approaches in two-dimensional NMR data interpretation for metabolomics applications

Accurate and efficient structure elucidation from routine one-dimensional NMR spectra using multitask machine learning

Simultaneous Determination of Multiple Coupling Networks by High-Resolution 2D J-edited NMR Spectroscopy.

Machine Learning in Complex Organic Mixtures: Applying Domain Knowledge Allows for Meaningful Performance with Small Data Sets

Prediction of Protein 13cα NMR Chemical Shifts Using a Combination Scheme of Statistical Modeling and Quantum-Mechanical Analysis

NMR Calculations with Quantum Methods: Development of New Tools for Structural Elucidation and Beyond

An Orthogonal-Pattern Absorption-Mode 2D J -Resolved NMR Spectroscopy for Analyses on Complex Samples

Machine Learning in Complex Organic Mixtures: Applying Domain Knowledge Allows for Meaningful Performance with Small Datasets.