Automated Synthesis, Data Science and NMR in Natural Product Analysis and Synthesis
Lead Research Organisation:
University of Bristol
Department Name: Chemistry
Abstract
Full 3D structural elucidation of natural products is essential, particularly in cases in which their bioactivity is of interest. In such instances, "small" stereochemical features can have large impacts on binding and/or efficacy highlighting the importance of structural elucidation. Determining the 3D structure of natural products is however not a trivial task typically requiring a combination of complex NMR (e.g. NOESY, ROESY, etc.), derivatisation, and X-Ray crystallography.
Comparison with empirically predicted NMR is one common strategy during structural elucidation however these predictions are often highly inaccurate and typically cannot account for stereochemistry within molecules. Predictions made from quantum mechanical calculations (e.g. DFT) are far more accurate however require significant computational time, taking hours to days to predict for a single structure. Machine learning offers an appealing alternative, producing predictions in a matter of seconds. Within the Butts group this has been achieved through development of the graph transformer neural network IMPRESSION, which predicts NMR data (chemical shifts and coupling constants) from a given molecular graph.
More recently the group's attention has also turned to the inverse of this problem, predicting molecular structure utilising NMR data as the input. A model of this sort could revolutionise the way chemists approach structural elucidation and ultimately could serve to accelerate this process greatly. Through this project it is hoped that development of inverse IMPRESSION might be continued and ultimately utilising synthesis and NMR we might be able to validate the effectiveness of the model and general strategy in solving structural elucidation problems.
Comparison with empirically predicted NMR is one common strategy during structural elucidation however these predictions are often highly inaccurate and typically cannot account for stereochemistry within molecules. Predictions made from quantum mechanical calculations (e.g. DFT) are far more accurate however require significant computational time, taking hours to days to predict for a single structure. Machine learning offers an appealing alternative, producing predictions in a matter of seconds. Within the Butts group this has been achieved through development of the graph transformer neural network IMPRESSION, which predicts NMR data (chemical shifts and coupling constants) from a given molecular graph.
More recently the group's attention has also turned to the inverse of this problem, predicting molecular structure utilising NMR data as the input. A model of this sort could revolutionise the way chemists approach structural elucidation and ultimately could serve to accelerate this process greatly. Through this project it is hoped that development of inverse IMPRESSION might be continued and ultimately utilising synthesis and NMR we might be able to validate the effectiveness of the model and general strategy in solving structural elucidation problems.
Organisations
People |
ORCID iD |