Abstract
Cheminformatics-based machine learning (ML) has been employed to determine optimal reaction conditions, including catalyst structures, in the field of synthetic chemistry. However, such ML-focused strategies have remained largely unexplored in the context of catalytic molecular transformations using Lewis-acidic main-group elements, probably due to the absence of a candidate library and effective guidelines (parameters) for the prediction of the activity of main-group elements. Here, the construction of a triarylborane library and its application to an ML-assisted approach for the catalytic reductive alkylation of aniline-derived amino acids and C-terminal-protected peptides with aldehydes and H2 is reported. A combined theoretical and experimental approach identified the optimal borane, i.e., B(2,3,5,6-Cl4-C6H)(2,6-F2-3,5-(CF3)2-C6H)2, which exhibits remarkable functional-group compatibility toward aniline derivatives in the presence of 4-methyltetrahydropyran. The present catalytic system generates H2O as the sole byproduct.
Similar content being viewed by others
Introduction
Catalysis is a fact of our daily lives. A wide variety of important commercial chemical substances are currently produced on both the fine and bulk scales in the presence of molecular catalysts that have been optimized based on specific factors such as efficiency, toxicity, cost, or a combination thereof. Recent advancements in cheminformatics-based machine learning (ML) offer chemists a way to bypass traditional Edisonian empiricism and develop more efficient approaches to optimizing catalysts1,2,3,4. Several groups have reported successful demonstrations of ML-driven optimizations of homogeneous catalysts such as phosphoric acids and Lewis-basic ligands for metal-based catalysts involving phosphines, N-heterocyclic carbenes, and nitrogen-based ligands5,6,7,8,9,10,11,12,13,14,15,16,17,18.
Recent progress in frustrated Lewis pairs (FLPs)19,20 has expanded the practical and sustainable application of main-group catalysis, e.g., enabling the hydrogenation of unsaturated molecules without toxic/precious metals21,22,23,24,25,26. In this context, the main-group-catalyzed reductive alkylation of amines with carbonyl compounds and H2 via the generation of FLP species has been widely accepted as a waste-minimizing process that generates valuable N-alkylated amines, whereby H2O is the only by-product23,27,28,29,30,31. Our group30 and that of Soós28 have independently shown that triarylboranes effectively catalyze the reductive alkylation of a variety of amines with aldehydes in the presence of H2. Moreover, we have demonstrated that an FLP that consists of Soós’ borane, i.e., B(2,6-Cl2-C6H3)(2,3,5,6-F4-C6H)2 (B1a)27, and tetrahydrofuran (THF) exhibits good functional-group tolerance for aniline derivatives including halogens, hydroxyl, and amide groups (Fig. 1A)30. Based on mechanistic studies, we proposed dual catalysis of B1a in the formation of imine intermediates and the hydrogenation of the imines via the generation of the FLP species (Fig. 1B). However, the direct reductive alkylation of amino acids with H2 has remained challenging, and such reactions have proceeded in only low-to-moderate yields even under forcing conditions. In terms of toxicity, the solvent THF, which also acts here as a Lewis base to generate FLPs with boranes (Fig. 1B), should be replaced with a less hazardous chemical32,33. Given the central role of the reductive alkylation of amines using carbonyl compounds in the synthesis of, e.g., pharmaceuticals, bio-active molecules, and agrochemicals34,35,36, the development of a straightforward and greener protocol for derivatizing amino acids and peptides would be worthwhile.
To this end, we envisioned an ML-assisted approach to identify a suitable triarylborane that is able to efficiently catalyze the reductive alkylation of amino acids and peptides with H2 through the construction of an in-silico library that includes a variety of triarylboranes. The synthesis of triarylboranes with unknown substitution patterns is typically a laborious and time-consuming process that often requires several weeks or even months for optimization. However, once the optimal procedures have been obtained, the optimized conditions can often be extrapolated, which is much faster. Therefore, using an ML-assisted approach to streamlining the selection of triarylborane candidates has the potential to significantly accelerate the optimization process and therefore the entire research process. Moreover, through the construction of this in-silico library, we aimed to contribute to the structural diversification of triarylboranes beyond the archetypical B(C6F5)337,38,39,40, which should expand the utility of this compound class in catalysis, materials science, and other areas. It should be noted that Dyson and Corminboeuf et al. have recently demonstrated the hydrogenation of CO2 to yield a formate salt [DBU‒H][H‒COO], which was catalyzed by an FLP consisting of tris(p-bromo)tridurylborane and 1,8-diazabicyclo[5.4.0]undec-7-ene (DBU)41. They identified this combination of borane and DBU using a cheminformatics-assisted approach that profiled the theoretically predicted catalyst activity based on the intrinsic acidity and basicity of the Lewis components.
Herein, we report the construction of an in-silico library with 54 triarylboranes, which was used for the ML-assisted identification of the optimal borane for the catalytic reductive alkylation of aniline-derived amino acids and peptides with aldehydes and H2 (Fig. 1A). We also explored the functional-group compatibility of the present system using the functional group evaluation (FGE) kit recently proposed by Morimoto and Oshima et al.42, which is based on the concept of robustness screening that has been proposed by Glorius et al.43,44.
Results and discussion
Theoretical and experimental variables collection
We started our investigation with the construction of an in-silico library of triarylboranes using the strategy described below. Generally, we explored triarylboranes that seemed to be synthetically accessible using common procedures37,40. Optimization of the gas-phase structures of the 54 boranes shown in Fig. 2A was accomplished using DFT calculations at the ωB97X-D/6-311+G(d,p)//ωB97X-D/6-31G(d,p) level. The 53 explored heteroleptic boranes Bxy (x = 1–6, y = a–w) include two 2,6-F2-3,5-R2-C6H groups, and their core structures are classified as B1–B6 depending on the R groups, i.e., Bx(R) = B1(F), B2(Cl), B3(Br), B4(CF3), and B5(H), whereas B6 includes 2,6-F2-3-Cl-C6H2 groups. With these core structures, we combined 23 aryl groups (a–w), and completed the construction of the borane library with the addition of B(C6F5)3. It should be noted that boranes Bxa (x = 1–3, 5, 6), B1v, and B1w were known before the construction of this library27,30,45,46,47, and their reactivity in the hydrogenation of unsaturated molecules inspired us when designing the library molecules. In fact, B1a has previously been employed for the catalytic reductive alkylation of aniline derivatives to generate active FLP species with THF30, and thus, in-silico derivatization of B1a was carried out via substitution of the meta and/or para H atoms in the 2,6-Cl2-C6H3 group with Cl, Br, CF3, OMe, OCF3 or C6F5 groups. We further conducted extensive in-silico derivatization of B2a and B3a, given that these boranes exhibit far superior catalytic activity than B1a, B6a, and B(C6F5)3 in the hydrogenation of N-heteroaromatics using a gaseous mixture of H2/CO/CO2/CH445. In these cases, we envisioned that the modulation of the intrinsic Lewis acidity of the triarylboranes, i.e., the energy levels of the LUMO, which includes the p orbital on the boron center48, as well as the remote back-strain49 that would influence the stability of the four-coordinated tetrahedral Lewis base–borane adducts. The introduction of 2,6-Br2-C6H3 (j) and its derivatives (k–r) into the B2 core was also explored, as we expected an increase in the front strain that influences the accessibility of the Lewis bases to the boron centers37.
We subsequently obtained the theoretical parameters. We thought that the use of structural parameters obtained from the gas-phase optimization of Bxy would not play a critical role in predicting the reactivity of the triarylboranes under the chosen conditions, given that no substantial differences were observed among them (Supplementary Table 4). Thus, we obtained the following energetic parameters, calculated at the ωB97X-D/6-311+G(d,p)//ωB97X-D/6-31G(d,p) level: (i) the energy levels of the LUMOs [eV], which include the p orbitals on the boron atoms, (ii) the energy barriers (ΔGH‡) [kcal mol‒1] for the heterolytic cleavage of H2 with the combination of Bxy and THF, and (iii) the relative Gibbs energy values (ΔGw°) [kcal mol‒1] for the formation of the H2O‒Bxy adducts with respect to [H2O + Bxy]. These theoretical values are shown in Fig. 3A for selected boranes. For parameter (ii), we have previously proposed that the heterolytic cleavage of H2 by FLPs should be involved in the rate-determining event of the B1a-catalyzed reductive alkylation of amines (Path I in Fig. 1B)30. For parameter (iii), H2O could be a potential quencher of the triarylborane catalysts via the formation of adducts (Path II in Fig. 1B) followed by proto-deboronation50. In this context, we envisioned that boranes Bxy that exhibit larger ΔGw° and smaller ΔGH‡ values should show superior performance as Lewis acids for the generation of more active FLPs with ethereal components. It should also be mentioned here that we theoretically optimized a structure that included a sole imaginary frequency related to the H−H bond cleavage; however, to reduce the calculation costs, we performed an IRC calculation only for selected cases and confirmed their validity as a possible transition state structure.
Next, we turned our attention to collecting experimental data for the reported triarylboranes (Bxa, x = 1–3, 5, 6; B1v; B1w; B(C6F5)3) and newly synthesized boranes (B1f; B2y, y = b, c, e; B3y, y = b, c, s); the latter compounds were used to analyze the influence of the derivatization of the 2,6-Cl2-C6H3 structure. We obtained the turnover frequency (TOF in h‒1; Fig. 3A) as an experimental parameter calculated based on the yield of 3aa from the reductive alkylation of amino acid 1a with benzaldehyde (2a) and H2 in the presence of 5 mol% Bxy under the shown conditions as a model reaction (Fig. 2B). The design of this model reaction is based on the potential of B1a to produce 3aa in approximately 50% yield after 6 h (i.e., TOF = 1.67 h‒1), as this consideration allows for a clearer evaluation of the positive and negative effects of structural derivatization during the optimization of Bxy.
Reaction conditions optimization
With the theoretical and experimental parameters in hand, Gaussian process regression (GPR), which is a radial basis function kernel-based statistical-learning algorithm, using GPy (a programming library for GPR)51, was applied to construct a model to predict the TOF values for the production of 3aa (it should be noted here that unless stated otherwise, mean values are presented for the theoretically predicted TOF values). GPR using GPy constructs a regression model using a limited number of observed data through ML and searches for a subsequent adequate parameter value of Bxy using the surrogate model. Note that GPR analysis has been widely explored in depth in information science and is characterized by the availability of a wide range of acquisition functions. Hence, its performance and flexibility to address various types of problems can be expected to efficiently promote the exploration of wide-ranging experimental conditions52. We evaluated the accuracy of each GPR model based on the coefficient of determination (Q2) of the leave-one-out (LOO) cross-validation using either training or validation data. It is noteworthy that higher Q2 values approaching 1 indicate a more robust explanation for the data. We carried out the initial GPR analysis using the experimental TOF values and pairs of two of the three theoretical parameters (LUMO energy level, ΔGw°, and ΔGH‡) obtained for the 15 boranes shown in Fig. 3A as training data. This GPR analysis resulted in three distinct models, labelled Model I (ΔGw° vs ΔGH‡; Q2 = 0.75), II (LUMO level vs ΔGH‡; Q2 = 0.76), and III (LUMO level vs ΔGw°; Q2 = 0.21) (Fig. 3B). A significantly lower Q2 of Model III would indicate its insufficient reliability. Subsequently, we used the theoretical parameters of the other 39 triarylboranes to predict their TOF values using these three models. We found intriguing inconsistencies among the TOF values for the B4 derivatives that contain meta-CF3 groups predicted using Model I and those predicted using Model II (Fig. 3C), albeit that the Q2 values of Model I and II were identical at this stage. For example, when Model I was applied, the TOF values of B4b, B4c, and B4e were predicted to be 3.50, 3.66, and 3.59, respectively; however, using Model II, the corresponding TOF values were predicted to be 0.55, 0.23, and 0.55 respectively (Fig. 3C, D). A critical difference between Model I and Model II is the use of the LUMO energy levels in the GPR analysis. Thus, to evaluate whether the LUMO levels can serve here as a critical parameter for the prediction of the TOF for the production of 3aa, we additionally synthesized B4b, B4c, and B4e, and confirmed that these boranes demonstrate excellent activity for the reductive alkylation of 1a with 2a under the model reaction conditions as predicted by Model I (Fig. 3D). We updated each model additionally using the experimental TOF values of B4b, B4c, and B4e to construct Model I′ (Q2 = 0.73), II′ (Q2 = 0.28), and III′ (Q2 = 0.28) (Supplementary Table 3). These results convinced us of the superiority of Model I′ for the prediction of the catalytic activity of Bxy under the applied conditions. Based on synthetic accessibility considerations, we decided to employ B4b as the optimal catalyst in the following experiments. These results also suggest that the employment of a combination of theoretical parameters related to the rate-determining step (e.g., ΔGH‡ in this work) and the (potential) catalyst deactivation step (e.g., ΔGw in this work) is decisive, whereas a parameter related to intrinsic Lewis acidity of triarylboranes, such as the LUMO energy level, is inappropriate for the prediction of the catalytic activity of triarylboranes under the applied conditions.
We also aimed to investigate relatively unexplored theoretical parameters for the construction of a regression-based model for the prediction of the catalytic activity of triarylboranes. In this context, we explored the deformation energy (EDEF) [kcal mol‒1] that can be used to evaluate the degree of remote back-strain49, where EDEF represents the energetic penalty associated with the change in the conformation at the boron center from trigonal planar to tetrahedral upon the formation of adducts between Lewis bases (LBs) and triarylboranes48,53,54. We calculated the EDEF(LB) values for 18 boranes shown in Fig. 4A via the gas-phase optimization of their H2O or THF adducts (i.e., EDEF(H2O) and EDEF(THF)) followed by energy-decomposition analysis at the RI-DSD-PBEP86-D3BJ/ma-Def2-QZVPP//PBEh-3c/Def2-SVP level. The GPR analysis with EDEF(LB) and ΔGH‡ using the experimental TOF values resulted in the construction of Model IV (LB = H2O; Q2 = 0.59) and V (LB = THF; Q2 = 0.51) with an acceptable reliability (Fig. 4B). Interestingly, these Q2 values were found to be insensitive to the differences in H2O and THF. For further comparison, we also prepared Model VI (q(B) vs ΔGH‡; Q2 = 0.09) and VII (q(C) vs ΔGH‡; Q2 = ‒0.31), wherein Mulliken charges [e] on the boron atoms and their average values on three ipso-carbon atoms in Bxy are given as q(B) and q(C), respectively; however, the reliability of these models was found to be insufficient. These results thus suggest that EDEF should be a valuable parameter to explore during the ML-based optimization of triarylboranes, as the estimation of EDEF through the optimization of structurally simple molecules (i.e., boranes, Lewis bases such as H2O, and their adducts) is technically easier than the calculation of the activation energies, which is only feasible after the optimization of transition states.
With the optimal triarylborane B4b in hand, we modified the reaction conditions to reduce its environmental impact and thus establish a greener and more sustainable system. In this context, the use of an alternative reaction solvent that could also act as a Lewis base to generate an FLP with B4b was initially explored, given the recent demand for the replacement of hazardous THF with alternative ethereal compounds that exhibit lower toxicity combined with high chemical and thermal stability32. In the presence of 40 atm H2, the reductive alkylation of 1a with 2a was carried out using THF, 2-methyltetrahydrofuran (2-MeTHF), cyclopentyl methyl ether (CPME), 4-methyltetrahydropyran (MTHP), or 2,2,4-trimethyl-1,3-dioxolane (TMD) (Fig. 5A). While THF provided a superior result (50%) compared to 2-MeTHF (38%) and CPME (18%), 3aa was generated in 59% yield when MTHP was used. Prolongation of the reaction time to 24 h resulted in the formation of 3aa in 72% yield; however, the removal of the 4 Å MS caused a decrease in the yield of 3aa to 42%. Finally, increasing the H2 pressure to 60 atm resulted in the formation of 3aa in 95% yield after the period of 24 h. The use of TMD did not furnish any 3aa. It should be noted that MTHP can be easily separated from water (its solubility in H2O is ~1.5 wt%) and removed under reduced pressure due to its strong hydrophobicity and low heat of vaporization, although its employment as a greener solvent has been limited in organic synthesis compared with the use of 2-MeTHF and CPME32,33. Moreover, to clarify the benefit of using MTHP over THF, we compared the activation energies for the heterolytic cleavage of H2 by the combination of B4b and THF or MTHP at the ωB97X-D/6-311+G(d,p)//ωB97X-D/6-31G(d,p) level (Fig. 5B). A possible transition state was found in both cases, and that in the case of MTHP was found to be more stabilized (TSMTHP = +21.8 kcal mol−1) than that in the case involving THF (TSTHF = +23.2 kcal mol−1). We attribute this stabilization to the increased structural flexibility of the tetrahydropyrane motif relative to THF, which allows the formation of efficient non-covalent interactions (NCIs) between the F/Cl atoms in B4b and the H atoms in MTHP. The participation of such NCIs was confirmed using the quantum theory of atoms in molecules (AIM) method (for details, see Supplementary Fig. 13)55,56.
Functional-group compatibility
The B4b-catalyzed reductive alkylation of 1a with 2a in MTHP using H2 (40 atm) demonstrated remarkable compatibility toward a variety of additives (A0–A21) (Fig. 6). All these experiments were carried out twice, and mean values [%] are given for the yield of 3aa, the recovered additive, and the imine intermediate 4-(benzylideneamino)benzoic acid formed in situ through the B4b-catalyzed condensation of 1a and 2a (left cycle in Fig. 1B). Initially, we carried out a control experiment using A0, which confirmed that the yields of 3aa and the remaining imine were consistent (72% and 28%, respectively) with those of the reaction conducted without A0 (Figs. 5A and 6). Relative to the control experiment, the reductive alkylation among 1a, 2a, and H2 proceeded without a significant change in the yield of 3aa, the recovered additives, or imine intermediates for additives with ketone (A3/A13), primary amide (A4), aryl bromide/iodide including alkyl ether (A6/A7), allylic ether (A8), terminal alkyne (A9), enone (A12), nitrile (A15), and ester (A20) moieties. It is also noteworthy that additives including sulfhydryl (A16) and sulfide (A17) moieties did not affect the present reaction, whereas such sulfur-containing compounds can be critical inhibitors in transition-metal-based catalysis and organocatalysis42. On the other hand, the hydrogenation of the imine intermediates was suppressed in the presence of aliphatic/aromatic carboxy (A1/A14), aliphatic hydroxyl (A2), bulky silyl ether (A10), pinacolatoboryl (A18), and N-tert-butoxycarbonyl (Boc) (A21) moieties, as these functional groups include either a Lewis-basic or -acidic site that can kinetically inhibit the formation of the FLP consisting of B4b and MTHP. In fact, the quantitative recovery of the additives after a period of 24 h was confirmed, without the generation of any other significant byproduct, i.e., the sum of the yields of 3aa and the imine was always ~95%. In contrast, the formation of 3aa was largely suppressed under reaction conditions including N-heteroaromatic moieties such as an imidazole (A5) and an indole (A19), as these heteroaromatic units can react with B4b to form Lewis adducts and/or with aldehydes to complicate the system. In the case of A11, which includes an epoxide moiety, 3aa was only produced in 13% yield, and a significant loss of A11 was confirmed after the reaction. Given that Lewis-acidic triarylboranes mediate the ring-opening transformation of epoxides57,58, the B4b-catalyzed ring-opening reaction of A11 to give the corresponding aldehyde can be expected to compete with the targeted reaction (see Supplementary Information for details).
Scope study
Finally, we applied the combination of B4b and MTHP for the reductive alkylation of aniline-derived amino acids and peptide derivatives in the presence of H2 (Fig. 7). Aminosalicylic acids 1b and 1c were effectively alkylated under the optimized conditions, and 3ba and 3ca were obtained in 90% and 93% yield, respectively; 60 atm of H2 was used in the latter case. For comparison, under a pressure of 80 atm of H2 in THF, 3ba and 3ca were furnished in 47% and 70% yield in the presence of 10 mol% and 15 mol% B1a, respectively, which again demonstrates the advantages of the present system using B4b and MTHP in terms of synthetic efficiency and sustainability. The reductive alkylation of anthranilic acid (1d), which is also known as vitamin L1, 5-aminoisophthalic acid (1e), and 4-aminophenylacetic acid (1f) afforded 3da, 3ea, and 3fa in excellent yield using 40–60 atm of H2. In contrast, we recognized that aliphatic amino acids (or their imine derivatives) and substrates insoluble in MTHP were not suitable. For example, aspartic acid (1g) is insoluble in MTHP, and no reaction took place when 1g was employed under otherwise identical conditions. Esterification of the carboxy group in 3-amino-4,4-dimethylpentanoic acid effectively improved its solubility in MTHP; however, the hydrogenation of the imine derived from 1h and 2a did not occur. Based on these results, we prepared alanine- and S-methylcysteine-based peptides 1i and 1j, and subjected them to the optimal reaction conditions; alkylated peptides 3ia and 3ja were obtained in 95% and 55% yield, respectively. In terms of the scope of aldehydes, 3,5-bis(trifluoromethyl)benzaldehyde (2b) furnished 3ab in 93% yield, but 3,5-di-tert-butylbenzaldehyde (2c) gave 3ac in merely 38%. In the latter case, a significant amount of 2c remained unreacted, indicating difficulties associated with the formation of the imine intermediate due to the decreased electrophilicity of the aldehyde moiety. A comparable result was obtained when p-tolualdehyde (2d) was used with respect to the case using 2a, and 3ad was afforded in 80% yield. As confirmed by the aforementioned robustness screening, the B4b/MTHP system exhibited remarkable functional-group compatibility in the reductive alkylation of aniline derivatives to afford 3ka−3sa in excellent yield. Given that harsh conditions, including 80 atm of H2, 10 mol% borane, and/or a longer reaction time were required for the reactions with 1l, 1 m, and 1q in the reported B1a/THF system, the present B4b/MTHP system clearly demonstrates its advantages.
The present study demonstrates an in-silico-assisted approach to designing triarylboranes that exhibit promising reactivity as main-group catalysts for the reductive alkylation of aniline-derived amino acids with H2. We have constructed an in-silico library of triarylboranes and obtained their theoretical parameters using DFT calculations. Guided by Gaussian process regression (GPR) using theoretical and experimental parameters, we identified the optimal triarylborane, i.e., B(2,3,5,6-Cl4-C6H)(2,6-F2-3,5-(CF3)2-C6H)2 (B4b). Through the evaluation of the regression-based models, we confirmed that the use of a parameter related to an intrinsic Lewis acidity of the triarylboranes (e.g., LUMO energy level and Mulliken charge on the boron atom) as one of the variables for the GPR analysis may lead to an underestimation when predicting the catalyst activity (TOF in h−1) under the optimized reaction conditions. Moreover, we propose that the deformation energy (EDEF) may serve as a potentially useful parameter to construct an adequate model. We also identified that 4-methyltetrahydropyran (MTHP) is a superior Lewis-basic solvent for not only the generation of FLP species with B4b, but also for the realization of a more practical and less-hazardous reaction system compared to a system using THF. In fact, the B4b-catalyzed reductive alkylation using aldehydes as an alkylating reagent and H2 in MTHP proceeded efficiently even in the presence of a variety of additives, showcasing its broad functional-group compatibility. Aniline-derived amino acids and C-terminal-protected peptides were alkylated in good-to-excellent yields under the optimized conditions with the concomitant generation of H2O as the sole byproduct.
Methods
General procedures for reductive alkylation of 1x′ with 2y′ affording 3x′y′: A 30 mL autoclave was charged with 1x′ (0.40 mmol), 2 y′ (0.40 mmol), B4b (0.02 mmol), 4 Å MS (100 mg), and MTHP (8 mL). Once sealed, the vessel was pressurized with H2 (40 or 60 atm), and the reaction mixture was stirred at 100 °C for 24 h. Then, degassed at rt followed by the addition of acetone, the resultant mixture was filtered to remove MS and other solids when generated. Subsequently, all volatiles were removed in vacuo to give 3x′y′, which was purified by flash column chromatography on silica gel.
Data availability
Data generated or analyzed during this study are provided in full within the published article and its supplementary materials. Metrical data for the solid-state structures are available from the Cambridge Crystallographic Data Centre (CCDC) under reference numbers 2295627 (B1f), 2295628 (B2b), 2295633 (B2c), 2295634 (B2e), 2295629 (B3b), 2295635 (B3c), 2295631 (B3s), 2295632 (B4b), and 2295630 (B4e). These data can be obtained free of charge from the CCDC via www.ccdc.cam.ac.uk/data_request/cif. Coordinates of the optimized structures are provided as source data. All other data are available from the corresponding author. Source data are provided with this paper.
Code availability
The code used in this work can be found in the Zenodo repository at https://doi.org/10.5281/zenodo.8420294.
References
Vogiatzis, K. D. et al. Computational approach to molecular catalysis by 3d transition metals: challenges and opportunities. Chem. Rev. 119, 2453–2523 (2019).
Toyao, T. et al. Machine learning for catalysis informatics: recent applications and prospects. ACS Catal. 10, 2260–2297 (2020).
Foscato, M. & Jensen, V. R. Automated in silico design of homogeneous catalysts. ACS Catal. 10, 2354–2377 (2020).
Matsubara, S. Digitization of organic synthesis — how synthetic organic chemists use AI technology —. Chem. Lett. 50, 475–481 (2021).
Zahrt, A. F., Athavale, S. V. & Denmark, S. E. Quantitative structure–selectivity relationships in enantioselective catalysis: past, present, and future. Chem. Rev. 120, 1620–1689 (2020).
Rinehart, N. I., Zahrt, A. F., Henle, J. J. & Denmark, S. E. Dreams, false starts, dead ends, and redemption: a chronicle of the evolution of a chemoinformatic workflow for the optimization of enantioselective catalysts. Acc. Chem. Res. 54, 2041–2054 (2021).
Crawford, J. M., Kingston, C., Toste, F. D. & Sigman, M. S. Data science meets physical organic chemistry. Acc. Chem. Res. 54, 3136–3148 (2021).
Liles, J. P. et al. Data science enables the development of a new class of chiral phosphoric acid catalysts. Chem. 9, 1518–1537 (2023).
Chen, H. et al. Data-driven catalyst optimization for stereodivergent asymmetric synthesis by iridium/boron hybrid catalysis. Cell Rep. Phys. Sci. 2, 100679 (2021).
Gensch, T. et al. A comprehensive discovery platform for organophosphorus ligands for catalysis. J. Am. Chem. Soc. 144, 1205–1217 (2022).
Xu, J. et al. Atroposelective Negishi coupling optimization guided by multivariate linear regression analysis: asymmetric synthesis of KRAS G12C covalent inhibitor GDC-6036. J. Am. Chem. Soc. 144, 20955–20963 (2022).
Karl, T. M. et al. Machine learning-guided development of trialkylphosphine Ni(I) dimers and applications in site-selective catalysis. J. Am. Chem. Soc. 145, 15414–15424 (2023).
Goebel, J. F. et al. Computer-driven development of ylide functionalized phosphines for palladium-catalyzed Hiyama couplings. Angew. Chem. Int. Ed. 62, e202216160 (2023).
Huang, H., Zong, H., Bian, G., Yue, H. & Song, L. Correlating the effects of the N-substituent sizes of chiral 1,2-amino phosphinamide ligands on enantioselectivities in catalytic asymmetric Henry reaction using physical steric parameters. J. Org. Chem. 79, 9455–9464 (2014).
See, X. Y. et al. Iterative supervised principal component analysis driven ligand design for regioselective Ti-catalyzed pyrrole synthesis. ACS Catal. 10, 13504–13517 (2020).
Mukai, M., Nagao, K., Yamaguchi, S. & Ohmiya, H. Molecular field analysis using computational-screening data in asymmetric N-heterocyclic carbene-copper catalysis toward data-driven in silico catalyst optimization. Bull. Chem. Soc. Jpn. 95, 271–277 (2022).
Wang, J.-W. et al. Nickel-catalyzed remote asymmetric hydroalkylation of alkenyl ethers to access ethers of chiral dialkyl carbinols. J. Am. Chem. Soc. 145, 10411–10421 (2023).
Cordova, M., Wodrich, M. D., Meyer, B., Sawatlon, B. & Corminboeuf, C. Data-driven advancement of homogeneous nickel catalyst activity for aryl ether cleavage. ACS Catal. 10, 7021–7031 (2020).
Stephan, D. W. & Erker, G. Frustrated Lewis pair chemistry: development and perspectives. Angew. Chem. Int. Ed. 54, 6400–6441 (2015).
Jupp, A. R. & Stephan, D. W. New directions for frustrated Lewis pair chemistry. Trends Chem. 1, 35–48 (2019).
Weicker, S. A. & Stephan, D. W. Main group Lewis acids in frustrated Lewis pair chemistry: beyond electrophilic boranes. Bull. Chem. Soc. Jpn. 88, 1003–1016 (2015).
Scott, D. J., Fuchter, M. J. & Ashley, A. E. Designing effective ‘frustrated Lewis pair’ hydrogenation catalysts. Chem. Soc. Rev. 46, 5689–5700 (2017).
Hoshimoto, Y. & Ogoshi, S. Triarylborane-catalyzed reductive N-alkylation of amines: a perspective. ACS Catal. 9, 5439–5444 (2019).
Fasano, V. & Ingleson, M. J. Recent advances in water-tolerance in frustrated Lewis pair chemistry. Synthesis 50, 1783–1795 (2018).
Paradies, J. Structure-reactivity relationships in borane-based FLP-catalyzed hydrogenations, dehydrogenations, and cycloisomerizations. Acc. Chem. Res. 56, 821–834 (2022).
Feng, X. & Du, H. Metal-free asymmetric hydrogenation and hydrosilylation catalyzed by frustrated Lewis pairs. Tetrahedron Lett. 55, 6959–6964 (2014).
Gyömöre, Á. et al. Moisture-tolerant frustrated Lewis pair catalyst for hydrogenation of aldehydes and ketones. ACS Catal. 5, 5366–5372 (2015).
Dorkó, É. et al. Expanding the boundaries of water-tolerant frustrated Lewis pair hydrogenation: enhanced back strain in the Lewis acid enables the reductive amination of carbonyls. Angew. Chem. Int. Ed. 56, 9512–9516 (2017).
Sapsford, J. S. et al. Direct reductive amination of carbonyl compounds catalyzed by a moisture tolerant tin(IV) Lewis acid. Adv. Synth. Catal. 360, 1066–1071 (2018).
Hoshimoto, Y., Kinoshita, T., Hazra, S., Ohashi, M. & Ogoshi, S. Main-group-catalyzed reductive alkylation of multiply substituted amines with aldehydes using H2. J. Am. Chem. Soc. 140, 7292–7300 (2018).
Zhao, J. et al. A theoretical study on the borane-catalyzed reductive amination of aniline and benzaldehyde with dihydrogen: the origins of chemoselectivity. J. Org. Chem. 87, 1194–1207 (2022).
Bijoy, R., Agarwala, P., Roy, L. & Thorat, B. N. Unconventional ethereal solvents in organic chemistry: a perspective on applications of 2-methyltetrahydrofuran, cyclopentyl methyl ether, and 4-methyltetrahydropyran. Org. Process. Res. Dev. 26, 480–492 (2022).
Samoilov, V. et al. Bio-based solvents and gasoline components from renewable 2,3-butanediol and 1,2-propanediol: Synthesis and characterization. Molecules 25, 1723 (2020).
Bryan, M. C. et al. Key green chemistry research areas from a pharmaceutical manufacturers’ perspective revisited. Green Chem. 20, 5082–5103 (2018).
Murugesan, K. et al. Catalytic reductive aminations using molecular hydrogen for synthesis of different kinds of amines. Chem. Soc. Rev. 49, 6273–6328 (2020).
Roughley, S. D. & Jordan, A. M. The medicinal chemist’s toolbox: an analysis of reactions used in the pursuit of drug candidates. J. Med. Chem. 54, 3451–3479 (2011).
Carden, J. L., Dasgupta, A. & Melen, R. L. Halogenated triarylboranes: synthesis, properties and applications in catalysis. Chem. Soc. Rev. 49, 1706–1725 (2020).
He, J., Rauch, F., Finze, M. & Marder, T. B. (Hetero)arene-fused boroles: a broad spectrum of applications. Chem. Sci. 12, 128–147 (2021).
Kumar, G., Roy, S. & Chatterjee, I. Tris(pentafluorophenyl)borane catalyzed C–C and C–heteroatom bond formation. Org. Biomol. Chem. 19, 1230–1267 (2021).
Berger, S. M., Ferger, M. & Marder, T. B. Synthetic approaches to triarylboranes from 1885 to 2020. Chem. Eur. J. 27, 7043–7058 (2021).
Das, S., Turnell-Ritson, R. C., Dyson, P. J. & Corminboeuf, C. Design of frustrated Lewis pair catalysts for direct hydrogenation of CO2. Angew. Chem. Int. Ed. 61, e202208987 (2022).
Saito, N. et al. Functional group evaluation kit for digitalization of information on the functional group compatibility and chemoselectivity of organic reactions. Bull. Chem. Soc. Jpn. 96, 465–474 (2023).
Collins, K. D. & Glorius, F. A robustness screen for the rapid assessment of chemical reactions. Nat. Chem. 5, 597–601 (2013).
Collins, K. D. & Glorius, F. Intermolecular reaction screening as a tool for reaction evaluation. Acc. Chem. Res. 48, 619–627 (2015).
Hashimoto, T., Asada, T., Ogoshi, S. & Hoshimoto, Y. Main group catalysis for H2 purification based on liquid organic hydrogen carriers. Sci. Adv. 8, eade0189 (2022).
Erös, G. et al. Catalytic hydrogenation with frustrated Lewis pairs: selectivity achieved by size-exclusion design of Lewis acids. Chem. Eur. J. 18, 574–585 (2012).
Dorkó, É. et al. Correlating electronic and catalytic properties of frustrated Lewis pairs for imine hydrogenation. J. Organomet. Chem. 847, 258–262 (2017).
Erdmann, P. & Greb, L. What distinguishes the strength and the effect of a Lewis acid: analysis of the Gutmann–Beckett method. Angew. Chem. Int. Ed. 61, e202114550 (2022).
Sakuraba, M., Morishita, T., Hashimoto, T., Ogoshi, S. & Hoshimoto, Y. Remote back strain: a strategy for modulating the reactivity of triarylboranes. Synlett 34, 2187–2192 (2023).
Fasano, V., Radcliffe, J. E. & Ingleson, M. J. B(C6F5)3-catalyzed reductive amination using hydrosilanes. ACS Catal. 6, 1793–1798 (2016).
The GPy authors. GPy: A Gaussian process framework in python. http://github.com/SheffieldML/Gpy.
Kondo, M. et al. Exploration of flow reaction conditions using machine-learning for enantioselective organocatalyzed Rauhut-Currier and [3+2] annulation sequence. Chem. Commun. 56, 1259–1262 (2020).
Silva, D. R., Santos, L. A., Freitas, M. P., Guerra, C. F. & Hamlin, T. A. Nature and strength of Lewis acid/base interaction in boron and nitrogen trihalides. Chem. Asian J. 15, 4043–4054 (2020).
Timoshkin, A. Y., Davydova, E. I., Sevastianova, T. N., Suvorov, A. V. & Schaefer, H. F. Relationship between the energy of donor–acceptor bond and the reorganization energy in molecular complexes. Int. J. Quantum Chem. 88, 436–440 (2002).
Kumar, P. S. V., Raghavendra, V. & Subramanian, V. Bader’s theory of atoms in molecules (AIM) and its applications to chemical bonding. J. Chem. Sci. 128, 1527–1536 (2016).
Macchi, P. & Sironi, A. Chemical bonding in transition metal carbonyl clusters: complementary analysis of theoretical and experimental electron densities. Coord. Chem. Rev. 238–239, 383–412 (2003).
Bhagat, M. N. et al. Enhancing the regioselectivity of B(C6F5)3-catalyzed epoxide alcoholysis reactions using hydrogen-bond acceptors. ACS Catal. 9, 9663–9670 (2019).
Andrea, K. A. & Kerton, F. M. Triarylborane-catalyzed formation of cyclic organic carbonates and polycarbonates. ACS Catal. 9, 1799–1809 (2019).
Acknowledgements
We thank Kuraray Co., Ltd., for providing 4-methyltetrahydropyran. This project was supported by Grants-in-Aid for Transformative Research Area (A) Digitalization-Driven Transformative Organic Synthesis (JSPS KAKENHI grants 22H05363 to Y.Ho. and 21H05217 to S.T.) as well as the Environment Research and Technology Development Fund (JPMEERF20211R01 to Y.Ho.) of the Environmental Restoration and Conservation Agency of the Ministry of the Environment of Japan. Part of this work was supported by JST SPRING (grant JPMJSP2138 to Y.Hi.). Parts of the calculations were performed using resources from the Research Center for Computational Science, Okazaki, Japan (22-IMS-C107 and 23-IMS-C094).
Author information
Authors and Affiliations
Contributions
Y.Ho. conceived and directed this project. Y.Hi. and Y.Ho. performed the experiments and theoretical calculations. Y.Hi., T.W., and S.T. performed the Gaussian process regression. Y.Hi. prepared an initial draft of the manuscript, which was finalized by Y.Ho. with feedback from T.W., S.T., and S.O.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Peer review
Peer review information
Nature Communications thanks Tiago Rodrigues, Jiyang Zhao and the other, anonymous, reviewer for their contribution to the peer review of this work. A peer review file is available.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Source data
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Hisata, Y., Washio, T., Takizawa, S. et al. In-silico-assisted derivatization of triarylboranes for the catalytic reductive functionalization of aniline-derived amino acids and peptides with H2. Nat Commun 15, 3708 (2024). https://doi.org/10.1038/s41467-024-47984-0
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41467-024-47984-0
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.