QM assisted ML for 19F NMR chemical shift prediction

BACKGROUND

Ligand-observed 19F NMR detection is an efficient method for screening libraries of fluorinated molecules in fragment-based drug design campaigns. Screening fluorinated molecules in large mixtures makes 19F NMR a high-throughput method. Typically, these mixtures are generated from pools of well-characterized fragments. By predicting 19F NMR chemical shift, mixtures could be generated for arbitrary fluorinated molecules facilitating for example focused screens.

METHODS

In a previous publication, we introduced a method to predict 19F NMR chemical shift using rooted fluorine fingerprints and machine learning (ML) methods. Having observed that the quality of the prediction depends on similarity to the training set, we here propose to assist the prediction with quantum mechanics (QM) based methods in cases where compounds are not well covered by a training set.

RESULTS

Beyond similarity, the performance of ML methods could be associated with individual features in compounds. A combination of both could be used as a procedure to split input data sets into those that could be predicted by ML and those that required QM processing. We could show on a proprietary fluorinated fragment library, known as LEF (Local Environment of Fluorine), and a public Enamine data set of 19F NMR chemical shifts that ML and QM methods could synergize to outperform either method individually. Models built on Enamine data, as well as model building and QM workflow tools, can be found at https://github.com/PatrickPenner/lefshift and https://github.com/PatrickPenner/lefqm .

17 O nuclear magnetic resonance: Recent advances and applications

The present review is focused on the most recent achievements in the application of liquid phase 17 O nuclear magnetic resonance (NMR) to inorganic, organic, and biochemical molecules focusing on their structure, conformations, and (bio)chemical behavior. The review is composed of four basic parts, namely, (1) simple molecules; (2) water and hydrogen bonding; (3) metal oxides, clusters, and complexes; and (4) biological molecules. Experimental 17 O NMR chemical shifts are thoroughly tabulated. They span a range of as much as almost 650 ppm (from -35.6 to +610.0 ppm) for inorganic and organic molecules, whereas this range is much wider for biological species being of about 1350 ppm (from -12 to +1332 ppm), and in the case of hemoproteins and heme-model compounds, isotropic chemical shifts of up to 2500 ppm were observed. The general prospects and caveats in the modern development of the liquid phase 17 O NMR in chemistry and biochemistry are critically discussed and briefly outlined in view of their future applications.

Computational 19 F NMR of trifluoromethyl derivatives of alkenes, pyrimidines, and indenes

The ¹⁹ F NMR chemical shifts of 13 trifluoromethyl derivatives of alkenes, pyrimidines, and indenes were calculated at the DFT level using the BhandHLYP, BHandH, PBE, PBE0, O3LYP, B3LYP, KT2, and KT3 functionals in combination with the pcS-2 basis set. Best result was documented for the BHandHLYP functional: The mean absolute error (MAE) of 0.66 ppm for the scaled values was achieved for the range of about 20 ppm. Solvent, vibrational, and relativistic corrections were found to be rather small, especially when taken in combination, generally demonstrating a slight decrease in the difference between calculated and experimental fluorine chemical shifts. As a measure of the practical importance of these compounds, one should recall that the growing number of life science products that contain trifluoromethyl groups provides a continuing driving force for the development of an effective methodology that enables both regio- and stereoselective introduction of trifluoromethyl groups into both aliphatic and aromatic systems.

Tritium NMR: A compilation of data and a practical guide

The present review is focused on experimental methods and structural applications of tritium NMR. It consists of five parts covering accordingly, introduction, brief overview, early (based on the papers appearing before 2000), more recent (based on the papers appeared in the interim of 2000 to 2015), and recent (based on the papers that appeared after 2015) reports. A special interest in this review is focused on practical aspects of tritium NMR spectroscopy, which is thoroughly illustrated by its numerous applications in chemistry and biochemistry.

Recent advances in the liquid-phase 6,7 Li nuclear magnetic resonance

The present review is focused on experimental methods and structural applications, including computational aspects, of classical lithium liquid-phase nuclear magnetic resonance (NMR). It consists of four parts covering accordingly a brief overview, early experimental reports (papers of up to about 2015) and more recent (papers appearing in the interim of about 2015 until 2022) results, together with very few but highly prospective computational results.

Searching for the Best Values of NMR Shielding and Spin-Spin Coupling Parameters: CH4-nFn Series of Molecules as the Example

Attempts at the theoretical interpretation of NMR spectra have a very long and fascinating history. Present quantum chemical calculations of shielding and indirect spin-spin couplings permit modeling NMR spectra when small, isolated molecules are studied. Similar data are also available from NMR experiments if investigations are performed in the gas phase. An interesting set of molecules is formed when a methane molecule is sequentially substituted by fluorine atoms-CH_4-nF_n, where n = 0, 1, 2, 3, or 4. The small molecules contain up to three magnetic nuclei, each with a one-half spin number. The spectral parameters of CH_4-nF_n can be easily observed in the gas phase and calculated with high accuracy using the most advanced ab initio methods of quantum chemistry. However, the presence of fluorine atoms makes the calculations of shielding and spin-spin coupling constants extremely demanding. Appropriate experimental ¹⁹F NMR parameters are good but also require some further improvements. Therefore, there is a real need for the comparison of existing NMR measurements with available state-of-the-art theoretical results for a better understanding of actual limits in the determination of the best shielding and spin-spin coupling values, and CH_4-nF_n molecules are used here as the exceptionally important case.

Fluorine spin-spin coupling constants of pentafluorobenzene revisited at the ab initio correlated levels

All possible spin-spin coupling constants, ¹⁹ F-¹⁹ F, ¹⁹ F-¹³ C, and ¹⁹ F-¹ H, of pentafluorobenzene were calculated at five different levels of theory, HF, DFT, SOPPA (CCSD), CCSD, and the SOPPA (CCSD)-based composite scheme with taking into account solvent, vibrational, relativistic, and correlation corrections. Most corrections were next to negligible for the long-range couplings but quite essential for the one-bond carbon-fluorine coupling constants. Hartree-Fock calculations were found to be entirely unreliable, while DFT results were comparable in accuracy with the data obtained using the wave function-based methods.

Stereochemical dependence of substituent γ-effects in the 19 F NMR shielding constants

The substituent α-, β-, and γ-effects of the elements of the second and third periods on ¹⁹ F NMR chemical shifts are evaluated including the establishment of stereochemical dependence of γ-effect, the latter particularly important in stereochemical studies of fluorine-containing compounds. Benchmark calculations performed for a series of 32 simple inorganic fluorine-containing molecules demonstrated a markedly good correlation between calculated and experimental fluorine chemical shifts characterized by a mean absolute error of 22.5 ppm in the range of about 900 ppm, which corresponds to a 2.5% error in the percentage terms.

Computational 1 H and 13 C NMR in structural and stereochemical studies

Present review outlines the advances and perspectives of computational ¹ H and ¹³ C NMR applied to the stereochemical studies of inorganic, organic, and bioorganic compounds, involving in particular natural products, carbohydrates, and carbonium ions. The first part of the review briefly outlines theoretical background of the modern computational methods applied to the calculation of chemical shifts and spin-spin coupling constants at the DFT and the non-empirical levels. The second part of the review deals with the achievements of the computational ¹ H and ¹³ C NMR in the stereochemical investigation of a variety of inorganic, organic, and bioorganic compounds, providing in an abridged form the material partly discussed by the author in a series of parent reviews. Major attention is focused herewith on the publications of the recent years, which were not reviewed elsewhere.

Quantum Chemical Approaches to the Calculation of NMR Parameters: From Fundamentals to Recent Advances

Quantum chemical methods for the calculation of indirect NMR spin–spin coupling constants and chemical shifts are always in progress. They never stay the same due to permanently developing computational facilities, which open new perspectives and create new challenges every now and then. This review starts from the fundamentals of the nonrelativistic and relativistic theory of nuclear magnetic resonance parameters, and gradually moves towards the discussion of the most popular common and newly developed methodologies for quantum chemical modeling of NMR spectra.

Stereoelectronic interactions: A booster for 4 JHF transmission

Long-range proton-fluorine coupling constants (ⁿ J_HF ) are helpful for the structure elucidation of fluorinated molecules. However, their magnitude and sign can change with the relative position of coupled nuclei and the presence of substituents. Here, trans-4-tert-butyl-2-fluorocyclohexanone was used as a model compound for the study of the transmission of ⁴ J_HF . In this compound, the ⁴ J_H6axF was measured to be +5.1 Hz, which is five times larger than the remaining ⁴ J_HF in the same molecule (⁴ J_H4F  = +1.0 Hz and ⁴ J_H6eqF  = +1.0 Hz). Through a combination of experimental data, natural bond orbital (NBO) and natural J-coupling (NJC) analyses, we observed that stereoelectronic interactions involving the π system of the carbonyl group are involved in the transmission pathway for the ⁴ J_H6axF . Interactions containing the π system as an electron acceptor (e.g., σ_C6H6ax  → π^*_C═O and σ_CF  → π^*_C═O ) increase the value of the ⁴ J_H6axF , while the interaction of the π system as an electron donor (e.g., π_C═O  → σ^*_CF ) decreases it. Additionally, the carbonyl group was shown not to be part of the transmission pathway of the diequatorial ⁴ J_H6eqF coupling in cis-4-tert-butyl-2-fluorocyclohexanone, revealing that there is a crucial symmetry requirement that must be fulfilled for the π system to influence the value of the ⁴ J_HF in these systems.

Efficient Screening of Target-Specific Selected Compounds in Mixtures by 19F NMR Binding Assay with Predicted 19F NMR Chemical Shifts

Ligand-based 19 F NMR screening is a highly effective and well-established hit-finding approach. The high sensitivity to protein binding makes it particularly suitable for fragment screening. Different criteria can be considered for generating fluorinated fragment libraries. One common strategy is to assemble a large, diverse, well-designed and characterized fragment library which is screened in mixtures, generated based on experimental 19 F NMR chemical shifts. Here, we introduce a complementary knowledge-based 19 F NMR screening approach, named 19 Focused screening, enabling the efficient screening of putative active molecules selected by computational hit finding methodologies, in mixtures assembled and on-the-fly deconvoluted based on predicted 19 F NMR chemical shifts. In this study, we developed a novel approach, named LEFshift , for 19 F NMR chemical shift prediction using rooted topological fluorine torsion fingerprints in combination with a random forest machine learning method. A demonstration of this approach to a real test case is reported.

An inexpensive density functional theory-based protocol to predict accurate 19 F-NMR chemical shifts

Thanks to its advantages, ¹⁹ F-NMR is an increasingly popular technique for the structural characterization of F-containing molecules, among which polymers, materials, fluorophores, pharmaceuticals, and so forth. However, the computational calculation of the ¹⁹ F-NMR chemical shifts, both for prediction and interpretation of experimental spectra, remains a challenge. In this work a density functional theory (DFT) based protocol for the calculation of the chemical shifts is established within the framework of the gauge-independent atomic orbital method, upon verifying the performance of Hartree-Fock and 60 DFT functionals coupled with seven different basis sets. The benchmark is conducted using two sets of molecules, namely one used for testing methods and another used for probing; the former set consists of 134 molecules, the latter 50, yet both of them with F in different chemical environments. Following Bally-Rablen-Tantillo strategy, the scaling parameters and other statistical quantities were computed for each method upon least squares linear regression between experimental and computed chemical shifts. The designed computational workflow is computationally inexpensive and represents a significant improvement with respect to the current state of the art.

Computational NMR of charged systems

This review covers NMR computational aspects of charged systems-carbocations, heterocations, and heteroanions, which were extensively studied in a number of laboratories worldwide, first of all, at the Loker Hydrocarbon Research Institute in California directed for several decades by a distinguished scientist, the Nobel laureate George Andrew Olah. The first part of the review briefly outlines computational background of the modern theoretical methods applied to the calculation of chemical shifts and spin-spin coupling constants at the DFT and the non-empirical levels. The second part of the review deals with the historical results, advances, and perspectives of the computational NMR of classical carbocations like methyl cation, CH₃⁺ , and protonated methane, CH₅⁺ , together with their numerous homologs and derivatives. The third and the forth parts of this survey are focused on the NMR computational aspects of accordingly, heterocations and heteroanions, the organic and inorganic ions with a charge localized mainly on heteroatoms like boron, oxygen, nitrogen, and heavier elements.

On the Unexpected Accuracy of the M06L Functional in the Calculation of 1JFC Spin-Spin Coupling Constants

One-bond spin-spin coupling constants (SSCCs) between F and C are computed with density functional theory (DFT). Surprisingly, M06L stands out for its striking accuracy, outperforming any other investigated functional, including PBE0, otherwise considered one of the most reliable for couplings involving F. Although the computation of nuclear magnetic resonance (NMR) parameters involving F is known to be a challenging task, even with a rather small basis set as pcJ-1, M06L provides results with a MAD = 11.7 Hz, whereas the average deviation gets as much as 5 times larger for PBE0 (MAD = 60.0 Hz). In the context of SSCCs on the order of 300 Hz, this is particularly remarkable. We find that the accuracy of M06L/pcJ-1 in predicting ¹J_FC constants does not stem from a well-suited exchange or correlation part of the functional. Instead, it is believed to arise from a fortuitous cancellation of errors, as revealed by investigating the convergence of the basis set. Our findings also indicate that ¹J_FC constants are highly dependent on the amount of exact exchange included in the expression of the functional, with large fractions being critically important to achieving satisfactory results. Studying the effects of the geometry on ¹J_FC, we find that optimizing the geometry at the level of theory used to calculate SSCCs generally improves the quality of the results, although the combination of a M06-2X/aug-cc-pVTZ geometry with M06L/pcJ-1 ¹J_FC constants best reproduces the experimental data for organofluorine compounds (with the exception of fluoroalkenes).

Computational NMR of Carbohydrates: Theoretical Background, Applications, and Perspectives

This review is written amid a marked progress in the calculation of NMR parameters of carbohydrates substantiated by a vast amount of experimental data coming from several laboratories worldwide. By no means are we trying to cover in the present compilation a huge amount of all available data. The main idea of the present review was only to outline general trends and perspectives in this dynamically developing area on the background of a marked progress in theoretical and computational NMR. Presented material is arranged in three basic sections: (1)-a brief theoretical introduction; (2)-applications and perspectives in computational NMR of monosaccharides; and (3)-calculation of NMR chemical shifts and spin-spin coupling constants of di- and polysaccharides.

Recent advances in computational liquid-phase 77Se NMR

This review aims to highlight significant progress in the calculation of 77Se NMR chemical shifts and spin – spin coupling constants involving selenium substantiated with a vast amount of experimental data. The material is arranged in two basic sections: the first one dealing with the calculation of 77Se NMR chemical shifts and the second one dealing with the computation of spin – spin coupling constants involving 77Se nucleus, namely 77Se–1H, 77Se–13C and 77Se–77Se together with some more exotic types of couplings, 77Se – 15N, 77Se–19F, 77Se–29Si and 77Se–31P. A special attention is focused on the stereoelectronic effects involving selenium atom and their manifestation in the 77Se NMR spectra of organoselenium compounds studied with the aid of the modern calculation of 77Se NMR parametres in combination with experimental results.

The bibliography includes 114 references.