DU8ML: Machine Learning-Augmented Density Functional Theory Nuclear Magnetic Resonance Computations for High-Throughput In Silico Solution Structure Validation and Revision of Complex Alkaloids

Abstract

Machine learning (ML) profoundly improves the accuracy of the fast DU8+ hybrid density functional theory/parametric computations of nuclear magnetic resonance spectra, allowing for high throughput in silico validation and revision of complex alkaloids and other natural products. Of nearly 170 alkaloids surveyed, 35 structures are revised with the next-generation ML-augmented DU8 method, termed DU8ML.

Collective total synthesis of C4-oxygenated securinine-type alkaloids via stereocontrolled diversifications on the piperidine core

“…Spectroscopic data of the synthetic sample were in complete accordance with the isolation report 9 . Since the relative configuration of the epoxide moiety was unambiguously confirmed by a single-crystal X-ray diffraction analysis of the downstream derivative of 8b ( vide infra ), we experimentally confirmed the computationally proposed structure revision of securingine C ( 8b ) by the Kutateladze group 14 .

Fig.

…”

“…Importantly, single-crystal X-ray diffraction analysis of securingine D ( 9b , CCDC 2170237) unequivocally established the absolute and relative stereochemistry of the synthetic securingine D ( 9b ). Hence, the machine learning-based structural revision of securingine D ( 9b ) 14 was confirmed via its total synthesis.…”

“…Recently, we proposed an alternative structure of securingine A ( 7 ) based on its calculated ground-state structure and DP4 + probability analysis 13 . Kutateladze and coworkers suggested structure revisions of securingines C ( 8 ) and D ( 9 ) based on their machine learning (ML)-based hybrid density functional theory/parametric computations of nuclear magnetic resonance spectra 14 . However, these computationally proposed structures have yet to be experimentally confirmed via total synthesis.…”

Peculiar Reaction Products and Mechanisms Revisited with Machine Learning-Augmented Computational NMR

“…3 While such calculations are more commonly used to support structure elucidation of natural products, they are also used in assigning structures of synthetic products. 4 We have developed a machine learning-augmented DFT method for computational NMR, DU8ML, 5 which proved fast and accurate in revisions of many important natural products. Our approach is to "label" the substructure fragments, responsible for major deviations of the DFT-calculated values, with the appropriate SMARTS strings 6 and train the machine on a large set of reliable experimental nuclear spin coupling constants and chemical shifts to recognize the discrepancies and correct them.…”

Brief overview of recently reported misassigned natural products and their in silico revisions enabled by DU8ML, a machine learning-augmented DFT computational NMR method

“…For this work we rely on our recently developed machine learning-augmented DFT computational NMR method, DU8ML. 2 Hopefully, such overviews will become a recurring feature in Nat. Prod.…”