Skip to main content
NCBI home page
ACS AuthorChoice logoLink to ACS AuthorChoice
. 2024 Jul 11;15(29):7363–7370. doi: 10.1021/acs.jpclett.4c01657

Overcoming the Pitfalls of Computing Reaction Selectivity from Ensembles of Transition States

Ruben Laplaza †,‡, Matthew D Wodrich †,‡, Clemence Corminboeuf †,‡,*
PMCID: PMC11284845  PMID: 38990895

Abstract

graphic file with name jz4c01657_0006.jpg

The prediction of reaction selectivity is a challenging task for computational chemistry, not only because many molecules adopt multiple conformations but also due to the exponential relationship between effective activation energies and rate constants. To account for molecular flexibility, an increasing number of methods exist that generate conformational ensembles of transition state (TS) structures. Typically, these TS ensembles are Boltzmann weighted and used to compute selectivity assuming Curtin-Hammett conditions. This strategy, however, can lead to erroneous predictions if the appropriate filtering of the conformer ensembles is not conducted. Here, we demonstrate how any possible selectivity can be obtained by processing the same sets of TS ensembles for a model reaction. To address the burdensome filtering task in a consistent and automated way, we introduce marc, a tool for the modular analysis of representative conformers that aids in avoiding human errors while minimizing the number of reoptimization computations needed to obtain correct reaction selectivity.

Relying on computational methods, such as density functional theory (DFT), to accurately predict reaction selectivity remains a key challenge for in silico catalyst design.1−5 Small errors in computed transition state (TS) energies, even those below chemical accuracy (1 kcal/mol), can result in a reversal of predicted selectivity6 due to the exponential relationship between effective activation energies and rate constants.7−12 Dealing with these accuracy issues can further be complicated when large and flexible functional groups used to impart asymmetry through noncovalent interactions are present,13−15 as these larger systems are likely to adopt multiple TS conformations.

Computational approaches for estimating selectivity often resort to choosing one (or a handful) of relevant conformations derived from either “chemical intuition” or experimental evidence. The relative free energies of the presumed reaction pathways are then computed and the resulting selectivity estimated.16−18 While this computationally inexpensive approach may work for simple systems, it becomes increasingly tricky for larger species and cannot be generalized to large pools of catalysts.19−23 On the other extreme, (ab initio) molecular dynamics combined with enhanced sampling techniques such as metadynamics24 or replica exchange25−28 (which may be powered by machine learning potentials28−32) can provide full conformational landscapes that would yield accurate selectivity predictions. In practice, however, such approaches are generally too computationally demanding, either in terms of directly modeling the system over long time frames, or in generating the amount of training data needed to create ML models, and are thus limited to smaller systems.30,32−34

One pragmatic approach for determining selectivity from DFT data is to assume a system operates under Curtin-Hammett conditions.35 In such cases, full conformational sampling of TS structures can be undertaken, and the resulting energies weighted to obtain final product ratios.17,18,36−41 The popularity of this “Curtin-Hammett Conformational Sampling” (CHCS) method has fostered an increasing number of tools that rely on rotamer libraries,42−45 inexpensive potentials combined with enhanced sampling techniques46−48 as popularized by the CREST program49−53 or distance geometry methods54−59 to generate conformational ensembles of a molecule. These approaches provide more complete pictures of selectivity but also require additional computations. As an example of the importance of conformational degrees of freedom, we recently demonstrated how on-the-fly conformational sampling can be used to accurately model enantioselectivity for a diverse set of catalysts with reduced human intervention.59−61

When using any of the aforementioned approaches for conformer sampling, particularly the automated variants, inadequate handling of the TS ensembles can lead to significant errors in selectivity estimations. This arises primarily due to two situations: (1) the counting of multiple equivalent transition states (Repeated conformers, Figure 1) and (2) not distinguishing interconvertible and noninterconvertible pathways (Non-interconvertible conformers, Figure 1). Here, we highlight potential pitfalls of using the CHCS strategy by demonstrating how processing the same ensemble of computed TS conformers in various ways leads to virtually any selectivity prediction, even for a simple organic reaction. We then introduce marc, a tool for the modular analysis of representative conformers that improves selectivity predictions by untangling conformational ensembles through automated conformer classification and filtering.

Figure 1.

Figure 1

Open in a new tab

Schematic representation of the two concealed error sources in transition state conformer weighting.

Concealed Error Sources in Transition State Conformer Weighting. “Repeated conformer” errors arise when the same (or fundamentally identical) TSs present within an ensemble are counted multiple times. Such errors have different effects depending on how the selectivity is determined. In Boltzmann weighting, for instance, a repeated high energy TS can artificially raise the TS barrier height toward that product. On the other hand, if selectivity is assessed directly from rate constants, then a repeated low energy TS can artificially lower the barrier toward that product. In theory, such errors are easily avoided by manually filtering redundant species. Automation, however, introduces its own set of problems as small numerical discrepancies in bond lengths/angles and/or energies cause symmetry related structures to not be recognized by the program. Equivalent structures with different atom indexing (for instance due to rotations of t-butyl or phenyl groups) can also lead to ineffective filtering, unless graph isomorphisms are considered.62

“Interconversion error”, the error associated with not distinguishing and properly treating interconvertible and noninterconvertible structures, is subtler and trickier to process. On the potential energy surface, interconvertibility between TSs (first order saddle points) is governed by temperature-dependent barrier heights (second order saddle points), which are hard to characterize63 but have been shown to affect reaction dynamics.64−67 Clearly, two TSs differing only by, for instance, a small rotation of a C–Ph single bond (i.e., rotamers) are connected by a negligibly small energetic barrier, making these species easily interconvertible as they readily adopt the most energetically favorable structure to bypass the TS. On the other hand, conformationally locked structures, such as C2-symmetric biaryl moieties (Figure 1), cannot intercovert due to a high barrier associated with significant steric repulsion, and ergodicity is broken. Conformer generation tools are not necessarily bound by realistic kinetics, which results in the presence of different TS conformers within an ensemble that may not be accessible to one another. In principle, noninterconvertible TSs should be treated as separate reaction pathways, with the rate constants associated with each pathway leading to the same product being summed. On the other hand, interconvertible TSs should be treated as a single pathway. Improper treatment resulting in “double counting” in this setting would lead to an artificial decrease in the effective activation energy. Note that differentiating between conformational isomers and rotamers is a recurrent challenge in automated reaction network exploration.11,12,68−71

To illustrate how these error sources quantitatively impact selectivity predictions, we examine the N-methylation of a tropane (1) with isotopically labeled 14CH3I (Figure 2).72−75 Two conformations of the system (1a and 1b) exist in equilibrium through a pyramidal inversion of the bridging nitrogen (TSi). An SN2 reaction with 14CH3I leads to two methylated isotopomers (2a and 2b) formed through TSa and TSb, respectively. As the activation energies associated with the SN2 reaction are significantly larger (>12 kcal/mol) than that of the pyramidal N-inversion (<5 kcal/mol through TSi), the system operates under Curtin-Hammett conditions and the product distribution exclusively depends on the free energy barriers of TSa and TSb, independently of the energy of 1a and 1b.76 Using this system as an illustrative model, we calculate selectivity (expressed as an isotopomer ratio), employing different strategies to account for repeated and (non)interconvertible conformer issues.

Figure 2.

Figure 2

Open in a new tab

Free energy profile of the N-methylation reaction of 3-(benzoyloxy)-8-methyl-8-azabicyclo[3.2.1]octane with isotopically labeled 14CH3I via SN2 transition states TSa and TSb.

Selectivity Prediction without Conformational Sampling. As a first approximation, we identified structures for TSa and TSb which were optimized at the ωB97XD/def2-TZVP//ωB97XD/def2-SVP level (see Computational Details for additional information). These computations showed TSa to be 2.21 kcal/mol lower in energy than TSb (Figure 2). Taking

graphic file with name jz4c01657_m001.jpg 1

where the 0 subscript indicates that both activation energies are taken with respect to the lowest energy intermediate (i.e., ΔΔG‡ is expressed as the difference between the absolute free energies of TSa and TSb). In this case, ΔΔG‡ = −2.21 kcal/mol and the major product at 298 K is predicted to be 2a following

graphic file with name jz4c01657_m002.jpg 2

as seen in Figure 3a.

Figure 3.

Figure 3

Open in a new tab

Isotopomer ratios were calculated with different strategies. (a) Without any conformational sampling. (b) Through Boltzmann weighting of the 10 lowest TS conformers per pathway. (c) As before, but considering the 20 lowest conformers per pathway. (d) Through addition of the rate constants from TS conformers, considering the 10 lowest TS conformers per pathway. (e) As before, but considering the 20 lowest conformers per pathway. (f) Calculated by manually examining and filtering the relevant TS conformers. (g) Calculated considering structures selected by marc.

Selectivity Prediction with Conformational Sampling. Figure 2 shows that the energies of TSa and TSb lie close to one another. Structurally, however, both the freely rotating benzoyloxy group and the 8-membered bicycle can adopt a multitude of orientations in the TS. To refine the prediction of selectivity, we performed a constrained conformational search of both TSa and TSb using CREST version 2.11,50 (note that we use CREST, as opposed to other programs,77 simply based on its popularity) keeping the relevant I–CH3–N distances fixed to ensure facile geometric convergence during subsequent ωB97XD reoptimizations. The resulting TS ensembles contained 86 (TSa) and 146 (TSb) structures (see Figure 4a/b for the superimposed structures). We now examine selectivity determined using Boltzmann weighting and summed rate constant approaches.

Figure 4.

Figure 4

Open in a new tab

Superimposition of transition state conformers based on the RMSD of the tropane moiety. Full conformational ensemble of (a) TSa (86 structures) and (b) TSb (146 structures). (c) Twenty lowest energy conformers of TSa after DFT reoptimization, where all conformers belong to the same (orange) cluster. (d) Twenty lowest energy conformers of TSb after DFT reoptimization, where conformers belong to two clusters (downward pointing C=O, orange, or upward pointing C=O (pink).

Boltzmann Weighting of Transition State Conformers. Boltzmann weighting treats TSs as ensembles in which all conformers leading to a specific product are assumed to be freely interconvertible. As a first assumption, we took the NTS = 10 lowest energy TS structures leading to each product (as predicted by their GFN2-xTB78 energies). The TS geometries were reoptimized at the ωB97XD/def2-TZVP//ωB97XD/def2-SVP level (see Computational Details for more information), during which some GFN2-xTB conformers converged to identical TSs (and some similar conformers diverged to different TSs, vide infra). We use ΔG‡j,0 to refer to the individual TS energies (relative to 1b) and add a second subscript to differentiate a from b when necessary. Boltzmann weighting79 the ωB97XD TS energies of the 10 lowest energy TSa and TSb conformers gives the ensemble energies (indicated by the ens. superscript which is followed by the number of transition states included, NTS) as

graphic file with name jz4c01657_m003.jpg 3

where

graphic file with name jz4c01657_m004.jpg 4

are the Boltzmann weights. As per eq 3, ΔGens0 is always higher than the lowest ΔGj,0. The weighting process is conducted separately for all TSa and TSb conformers.80 Substituting these values back into eq 2 yields

graphic file with name jz4c01657_m005.jpg 5

and

graphic file with name jz4c01657_m006.jpg 6

as seen in Figure 3b. Thus, Boltzmann weighting reverses selectivity from the single conformer result (98:2, Figure 3). Of course, using the lowest 10 energy TS conformers leading to each product is arbitrary as we do not know the “true” number of unique pathways prior to conformer generation. If we assume 20 conformers to be a better choice and repeat the same process by reoptimizing the NTS = 20 lowest energy TS structures, we obtain ΔΔG‡ens,20 = 1.75 kcal/mol, giving a 5:95 product ratio (Figure 3d).

Recall that once the lowest energy TS is found, each additional TS conformer identified increases selectivity toward the opposite product. As an example, the highest energy TSb conformer in our ensemble lies 17.5 kcal/mol above 1b, which is higher than all TSa conformers. If duplicates of this high energy conformer are mistakenly added to the ensemble, ΔGensb,0 (which is normally lower than ΔGensa,0) would slowly tend toward 17.5 kcal/mol, eventually reversing the predicted preferred product. For this reason, duplicate TSs can be problematic, leading to inaccurate selectivity predictions.

Summing Rate Constants of Transition State Conformers. Boltzmann weighting assumes free interconvertability of all TSs leading to a specific product. However, if this interconversion is precluded by a high energetic barrier, then these TSs are best characterized as belonging to different reaction valleys that proceed in parallel toward their products. Assuming this is the case for all TS conformers, the effective rate constant keff. is given as

graphic file with name jz4c01657_m007.jpg 7

where the mole fraction of the reference state (1b) is neglected for simplicity39 and −ΔGeff is the effective activation energy toward the product. In other words, the effective kinetic rate constant is now the sum of all individual rate constants for all pathways leading to a specific product. In this case, the product distribution is the ratio of the total rate constants in each direction, given by

graphic file with name jz4c01657_m008.jpg 8

which is analogous to eqs 1 and 2 with effective activation energies. Taking NaTS = NbTS = 10 and substituting in eq 8, we obtain ΔΔG‡eff,10 = 0.48 kcal/mol and a product ratio [2a]/[2b] = 0.44, slightly selective toward 2b (31:69, Figure 3d). Using NaTS = NbTS = 20 gives ΔΔG‡eff,20 = 0.07 kcal/mol, corresponding to no selectivity (49:51, Figure 3e).

Here, additional low energy conformers significantly accelerate the reaction rate by providing multiple parallel pathways toward the product, while higher energy conformers do not influence the rate. Thus, “double counting” of low energy conformers in this setting lowers the effective activation energy in eq 7 by adding extra terms. In an extreme case, double counting of the lowest energy TS will give a RTln(2) lower barrier. Importantly, this treatment fundamentally differs from Boltzmann weighting, where high energy conformers would decelerate the reaction.

From the same sets of TS ensembles, we have obtained selectivity predictions ranging from highly selective for 2a to highly selective for 2b, solely by postprocessing the same results in different ways. To obtain unbiased selectivity predictions, conformer ensemble sorting and selection are required to differentiate interconvertible and parallel TS structures, which obtain accurate results.

The Right Answer for the Right Reasons. As highlighted above, we lack information about the ability of the different TS conformers in our ensemble to freely interconvert, because they are generated without relying on any energetic criterion. For the exemplary case studied here, however, it is possible to manually sort the conformers to arrive at the correct selectivity. Inspecting the DFT-reoptimized TSa ensemble reveals only a single conformer family in which all structures are freely interconvertible. This family has a ΔG‡a,0,down ≈ 15.3 kcal/mol above 1b and is characterized most notably by a downward pointing C=O bond (Figure 4c). In contrast, the TSb ensemble consists of two distinct conformer families characterized by either upward or downward pointing C=O bonds (Figure 4d), having values of ΔG‡b,0,up ≈ 17.5 kcal/mol and ΔG‡b,0,down = 13.4 kcal/mol, respectively.81 Moving between these two clusters requires overcoming second order saddle points with non-negligible barriers of over 8 kcal/mol above the TSs. As a result, all TSs can be grouped into one of three clusters: either TSa with a downward pointing C=O bond, TSb with a downward pointing C=O bond, or TSb with an upward pointing C=O bond. Using eq 3 we obtain the ensemble energies for each cluster, which can then be separately added, as in eq 7, to calculate the final selectivity using eq 8. Doing so gives [2a]/[2b] ≈ 4:96 (Figure 3f). In the end, only three (of the initial 212) TS conformers actually dictate the selectivity of the reaction. Note that this result is in agreement with the expected reactivity of tropanes, which typically privilege attacks from the less encumbered Re-side.82,83

marc: An Automated Clustering Tool to Avoid Errors. Manually curating structures is time-consuming and unsuitable for all but the simplest systems. To automate this process, we developed marc (modular analysis of representative conformers) as a simple command line tool to process conformational ensembles. For a given ensemble, independent of its origin, marc uses a mix of geometric (both symmetry-informed heavy-atom RMDSs and dihedral angles) and energetic (if available) information to perform clustering designed to obtain an optimal number of structures needed to completely cover the conformational space. The general workflow of marc is shown in Figure 5.

Figure 5.

Figure 5

Open in a new tab

Workflow of marc. From a conformer ensemble, marc computes pairwise distances using different metrics (heavy atom RMSDs, relative energies, and dihedral angles) to construct a compound distance matrix, finds the optimal number of clusters using the silhouette method and k-means clustering, and samples the lowest energy structures belonging to each cluster.

Applying marc (using the default settings) to the (NaTS = 20) DFT-refined ensembles gives a single cluster containing all conformers. The largest heavy-atom RMSD between two structures is a non-negligible 0.87 Å, which is sufficiently large to be considered as unique structures based on simple RMSD filtering using predefined thresholds. On the other hand, the maximum energy difference is only 0.01 kcal/mol. By combining energetic and structural criteria, marc successfully identifies all TSa conformers as being interconvertible. In contrast, the (NbTS = 20) ensemble is found to have two clusters. The first contains only a single structure with a downward pointing C=O bond while the second contains the 19 other structures having upward pointing C=O bonds. Here, the maximum heavy-atom RMSD among the conformers is slightly larger (1.32 Å vs 0.87 Å) while the maximum energy difference is 3.2 kcal/mol with the lowest energy conformer belonging to the first cluster and 19 higher energy conformers belonging to the second. As marc selects the same conformers found using manual inspection in the previous section, we once again obtain an isotopomer ratio of 4:96 (Figure 3g).

The above results were obtained by processing the 20 lowest energy conformers for both TSa and TSb optimized at the DFT level. This raises an important question: could the 40 DFT reoptimizations have been completely avoided and the same results obtained? Running marc directly on the full CREST-generated ensembles dramatically reduces the number of computations needed. For the original 86 structure TSa ensemble, three clusters populated by 1, 1, and 84 structures are identified. The largest corresponds to the downward pointing C=O bond species discussed earlier, while two others contain species with upward pointing C=O bonds that are significantly higher in energy (such that they were not included in the 20 lowest energy structures selected for DFT refinement).84 For the 146-structure TSb ensemble, two clusters characterized by downward (101 structures) and upward pointing C=O bonds (45 structures) are found. If just five total TS structures (one structure from each of the three TSa and two TSb clusters) are reoptimized by using DFT, a 6:94 product ratio is found. This closely matches the 4:96 product ratio obtained by processing the 20 lowest energy DFT reoptimized conformers from TSa/TSb, at ∼1/8 of the computational cost.85 In short, marc also simplifies the postprocessing of conformational ensembles that have not yet been refined with DFT, reducing large ensembles to a handful of representative structures thereby, reducing computational expense. Such savings become increasingly important when the CHCS protocol is applied to large species and/or in high-throughput settings.

In conclusion, using the N-methylation of tropane with isotopically labeled 14CH3I leading to two isotopomers as a model reaction, we have shown how an ensemble of the same TS conformers can be processed in different ways to obtain any possible selectivity prediction under Curtin-Hammett conditions. These different selectivity predictions arise from errors associated with the presence of “repeated conformer” and “interconversion” errors associated with distinguishing when various TSs are freely able to interconvert among themselves. We then introduced marc, a simple command line tool designed to analyze conformational ensembles and select the most representative structures. Using marc, accurate predictions of selectivity can be obtained with a significantly reduced computational cost. Having identified the problem, as well as a path toward a solution, a more thorough benchmark of clustering and conformational ensemble processing methodologies across different systems will be the subject of future work.

Computational Details

Conformer ensemble generation was performed with CREST49,50,78 version 2.11 using the default settings except doubling the default metadynamics runtime and setting a 0.5 au harmonic constraint placed on the I–CH3–N atoms involved in the SN2 transition state. Selected geometries were optimized at the ωB97XD86/def2-SVP87 level of theory as implemented in Gaussian 16.88 Vibrational frequency analysis was used to confirm that stationary points were either minima (no imaginary frequencies), transition states (one imaginary frequency), or second order saddle points (two imaginary frequencies) on the potential energy surface. Refined energy estimates were obtained by single point computations at the ωB97XD/def2-TZVP level on ωB97XD/def2-SVP geometries. Free energy corrections were taken from the ωB97XD/def2-SVP computations using the GoodVibes program.38 Solvent effects were included in the single point computations using the SMD89 implicit solvation model (for acetonitrile). Reported free energies include the solvation-corrected ωB97XD/def2-TZVP electronic energies, and the ωB97XD/def2-SVP free energy corrections. All structures and computed energies are available in the “examples” directory of https://github.com/lcmd-epfl/marc.

marc and user instructions can be found at https://github.com/lcmd-epfl/marc with DOI 10.5281/zenodo.12569985. Clustering is performed using the k-means algorithm, with a multidimensional scaling of the averaged dissimilarity matrices as input, as implemented in the scikit-learn python library.90 The silhouette score method is used to assess the number of clusters. RMSDs are computed considering all isomorphisms between the molecular graphs, which accounts for molecular symmetry.62marc also allows users to use other clustering algorithms and dissimilarity matrix types, including agglomerative methods and dihedral angles as pioneered by the CENSO program.79

Acknowledgments

The authors are grateful to the EPFL for financial support and computational resources. This publication was created as part of NCCR Catalysis (grant number 180544), a National Centre of Competence in Research funded by the Swiss National Science Foundation. R.L. thanks Dr. Shubhajit Das and Dr. Simone Gallarati for insightful discussions, and Dr. Marc H. Garner for inspiration regarding acronyms.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpclett.4c01657.

  • Transparent Peer Review report available (PDF)

The authors declare no competing financial interest.

Supplementary Material

jz4c01657_si_001.pdf (233.5KB, pdf)

References

  1. Poree C.; Schoenebeck F. A Holy Grail in Chemistry: Computational Catalyst Design: Feasible or Fiction?. Acc. Chem. Res. 2017, 50, 605–608. 10.1021/acs.accounts.6b00606. [DOI] [PubMed] [Google Scholar]
  2. Ahn S.; Hong M.; Sundararajan M.; Ess D. H.; Baik M.-H. Design and Optimization of Catalysts Based on Mechanistic Insights Derived from Quantum Chemical Reaction Modeling. Chem. Rev. 2019, 119, 6509–6560. 10.1021/acs.chemrev.9b00073. [DOI] [PubMed] [Google Scholar]
  3. Funes-Ardoiz I.; Schoenebeck F. Established and Emerging Computational Tools to Study Homogeneous Catalysis—From Quantum Mechanics to Machine Learning. Chem 2020, 6, 1904–1913. 10.1016/j.chempr.2020.07.008. [DOI] [Google Scholar]
  4. Burai Patrascu M.; Pottel J.; Pinus S.; Bezanson M.; Norrby P.-O.; Moitessier N. From desktop to benchtop with automated computational workflows for computer-aided design in asymmetric catalysis. Nat. Catal. 2020, 3, 574–584. 10.1038/s41929-020-0468-3. [DOI] [Google Scholar]
  5. Fey N.; Lynam J. M. Computational mechanistic study in organometallic catalysis: Why prediction is still a challenge. WIREs Comput. Mol. Sci. 2022, 12, e1590 10.1002/wcms.1590. [DOI] [Google Scholar]
  6. Krieger A. M.; Pidko E. A. The Impact of Computational Uncertainties on the Enantioselectivity Predictions: A Microkinetic Modeling of Ketone Transfer Hydrogenation with a Noyori-type Mn-diamine Catalyst. ChemCatChem. 2021, 13, 3517–3524. 10.1002/cctc.202100341. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Lam Y.-h.; Grayson M. N.; Holland M. C.; Simon A.; Houk K. N. Theory and Modeling of Asymmetric Catalytic Reactions. Acc. Chem. Res. 2016, 49, 750–762. 10.1021/acs.accounts.6b00006. [DOI] [PubMed] [Google Scholar]
  8. Thiel W. Computational Catalysis-Past, Present, and Future. Angew. Chem., Int. Ed. 2014, 53, 8605–8613. 10.1002/anie.201402118. [DOI] [PubMed] [Google Scholar]
  9. Vogiatzis K. D.; Polynski M. V.; Kirkland J. K.; Townsend J.; Hashemi A.; Liu C.; Pidko E. A. Computational Approach to Molecular Catalysis by 3d Transition Metals: Challenges and Opportunities. Chem. Rev. 2019, 119, 2453–2523. 10.1021/acs.chemrev.8b00361. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Harvey J. N.; Himo F.; Maseras F.; Perrin L. Scope and Challenge of Computational Methods for Studying Mechanism and Reactivity in Homogeneous Catalysis. ACS Catal. 2019, 9, 6803–6813. 10.1021/acscatal.9b01537. [DOI] [Google Scholar]
  11. Steiner M.; Reiher M. Autonomous Reaction Network Exploration in Homogeneous and Heterogeneous Catalysis. Top. Catal. 2022, 65, 6–39. 10.1007/s11244-021-01543-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Steiner M.; Reiher M. A human-machine interface for automatic exploration of chemical reaction networks. Nat. Commun. 2024, 15, 3680. 10.1038/s41467-024-47997-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Crawford J.; Sigman M. Conformational Dynamics in Asymmetric Catalysis: Is Catalyst Flexibility a Design Element?. Synthesis 2019, 51, 1021–1036. 10.1055/s-0037-1611636. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Cen S.; Huang N.; Lian D.; Shen A.; Zhao M.-X.; Zhang Z. Conformational enantiodiscrimination for asymmetric construction of atropisomers. Nat. Commun. 2022, 13, 4735. 10.1038/s41467-022-32432-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Maji R.; Mallojjala S. C.; Wheeler S. E. Electrostatic Interactions in Asymmetric Organocatalysis. Acc. Chem. Res. 2023, 56, 1990–2000. 10.1021/acs.accounts.3c00198. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Wodrich M. D.; Ye B.; Gonthier J. F.; Corminboeuf C.; Cramer N. Ligand-Controlled Regiodivergent Pathways of Rhodium(III)-Catalyzed Dihydroisoquinolone Synthesis: Experimental and Computational Studies of Different Cyclopentadienyl Ligands. Chem. - Eur. J. 2014, 20, 15409–15418. 10.1002/chem.201404515. [DOI] [PubMed] [Google Scholar]
  17. Peng Q.; Duarte F.; Paton R. S. Computing organic stereoselectivity– from concepts to quantitative calculations and predictions. Chem. Soc. Rev. 2016, 45, 6093–6107. 10.1039/C6CS00573J. [DOI] [PubMed] [Google Scholar]
  18. Guan Y.; Wheeler S. E. Automated Quantum Mechanical Predictions of Enantioselectivity in a Rhodium-Catalyzed Asymmetric Hydrogenation. Angew. Chem., Int. Ed. 2017, 56, 9101–9105. 10.1002/anie.201704663. [DOI] [PubMed] [Google Scholar]
  19. Gensch T.; dos Passos Gomes G.; Friederich P.; Peters E.; Gaudin T.; Pollice R.; Jorner K.; Nigam A.; Lindner-D’Addario M.; Sigman M. S.; et al. A Comprehensive Discovery Platform for Organophosphorus Ligands for Catalysis. J. Am. Chem. Soc. 2022, 144, 1205–1217. 10.1021/jacs.1c09718. [DOI] [PubMed] [Google Scholar]
  20. Alegre-Requena J. V.; Sowndarya S. V. S.; Perez-Soto R.; Alturaifi T. M.; Paton R. S. AQME: Automated quantum mechanical environments for researchers and educators. WIREs Comp. Mol. Sci. 2023, 13, e1663 10.1002/wcms.1663. [DOI] [Google Scholar]
  21. Hashemi A.; Bougueroua S.; Gaigeot M.-P.; Pidko E. A. HiREX: High-Throughput Reactivity Exploration for Extended Databases of Transition-Metal Catalysts. J. Chem. Inf. Model. 2023, 63, 6081–6094. 10.1021/acs.jcim.3c00660. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Kammeraad J. A.; Das S.; Argüelles A. J.; Sayyed F. B.; Zimmerman P. M. Conformational Sampling over Transition-Metal-Catalyzed Reaction Pathways: Toward Revealing Atroposelectivity. Org. Lett. 2024, 26, 2867–2871. 10.1021/acs.orglett.3c04047. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Kalikadien A. V.; Mirza A.; Hossaini A. N.; Sreenithya A.; Pidko E. A. Paving the road towards automated homogeneous catalyst design. ChemPlusChem. 2024, e202300702 10.1002/cplu.202300702. [DOI] [PubMed] [Google Scholar]
  24. Fleming K. L.; Tiwary P.; Pfaendtner J. New Approach for Investigating Reaction Dynamics and Rates with Ab Initio Calculations. J. Phys. Chem. A 2016, 120, 299–305. 10.1021/acs.jpca.5b10667. [DOI] [PubMed] [Google Scholar]
  25. Okamoto Y.; Sugita Y. Replica-exchange molecular dynamics method for protein folding. Chem. Phys. Lett. 1999, 314, 141–151. 10.1016/S0009-2614(99)01123-9. [DOI] [Google Scholar]
  26. Petraglia R.; Nicolai A.; Wodrich M. D.; Ceriotti M.; Corminboeuf C. Beyond static structures: Putting forth REMD as a tool to solve problems in computational organic chemistry. J. Comput. Chem. 2016, 37, 83–92. 10.1002/jcc.24025. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Fabregat R.; Fabrizio A.; Meyer B.; Hollas D.; Corminboeuf C. Hamiltonian-Reservoir Replica Exchange and Machine Learning Potentials for Computational Organic Chemistry. J. Chem. Theory Comput. 2020, 16, 3084–3094. 10.1021/acs.jctc.0c00100. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Gallarati S.; Fabregat R.; Juraskova V.; Inizan T. J.; Corminboeuf C. How Robust Is the Reversible Steric Shielding Strategy for Photoswitchable Organocatalysts?. J. Org. Chem. 2022, 87, 8849–8857. 10.1021/acs.joc.1c02991. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Behler J. First Principles Neural Network Potentials for Reactive Simulations of Large Molecular and Condensed Systems. Angew. Chem., Int. Ed. 2017, 56, 12828–12840. 10.1002/anie.201703114. [DOI] [PubMed] [Google Scholar]
  30. Young T. A.; Johnston-Wood T.; Deringer V. L.; Duarte F. A transferable active-learning strategy for reactive molecular force fields. Chem. Sci. 2021, 12, 10944–10955. 10.1039/D1SC01825F. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Jurásková V.; Célerse F.; Laplaza R.; Corminboeuf C. Assessing the persistence of chalcogen bonds in solution with neural network potentials. J. Chem. Phys. 2022, 156, 154112. 10.1063/5.0085153. [DOI] [PubMed] [Google Scholar]
  32. Young T. A.; Johnston-Wood T.; Zhang H.; Duarte F. Reaction dynamics of Diels–Alder reactions from machine learned potentials. Phys. Chem. Chem. Phys. 2022, 24, 20820–20827. 10.1039/D2CP02978B. [DOI] [PubMed] [Google Scholar]
  33. Yang M.; Bonati L.; Polino D.; Parrinello M. Using metadynamics to build neural network potentials for reactive events: the case of urea decomposition in water. Catal. Today 2022, 387, 143–149. 10.1016/j.cattod.2021.03.018. [DOI] [Google Scholar]
  34. Margraf J. T.; Jung H.; Scheurer C.; Reuter K. Exploring catalytic reaction networks with machine learning. Nat. Catal. 2023, 6, 112–121. 10.1038/s41929-022-00896-y. [DOI] [Google Scholar]
  35. Seeman J. I. Effect of conformational change on reactivity in organic chemistry. Evaluations, applications, and extensions of Curtin-Hammett Winstein-Holness kinetics. Chem. Rev. 1983, 83, 83–134. 10.1021/cr00054a001. [DOI] [Google Scholar]
  36. Suleimanov Y. V.; Green W. H. Automated Discovery of Elementary Chemical Reaction Steps Using Freezing String and Berny Optimization Methods. J. Chem. Theory Comput. 2015, 11, 4248–4259. 10.1021/acs.jctc.5b00407. [DOI] [PubMed] [Google Scholar]
  37. Husch T.; Seebach D.; Beck A. K.; Reiher M. Rigorous Conformational Analysis of Pyrrolidine Enamines with Relevance to Organocatalysis. Helv. Chim. Acta 2017, 100, e1700182 10.1002/hlca.201700182. [DOI] [Google Scholar]
  38. Luchini G.; Alegre-Requena J. V.; Funes-Ardoiz I.; Paton R. S. GoodVibes: automated thermochemistry for heterogeneous computational chemistry data. F1000Research 2020, 9, 291. 10.12688/f1000research.22758.1. [DOI] [Google Scholar]
  39. Williams I. H. Gibbs energies of activation for reacting systems with multiple reactant-state and transition-state conformations. J. Phys. Org. Chem. 2022, 35, e4312 10.1002/poc.4312. [DOI] [Google Scholar]
  40. Zhao Q.; Savoie B. M. Simultaneously improving reaction coverage and computational cost in automated reaction prediction tasks. Nature Comput. Sci. 2021, 1, 479–490. 10.1038/s43588-021-00101-3. [DOI] [PubMed] [Google Scholar]
  41. Zhao Q.; Hsu H.-H.; Savoie B. M. Conformational Sampling for Transition State Searches on a Computational Budget. J. Chem. Theory Comput. 2022, 18, 3006–3016. 10.1021/acs.jctc.2c00081. [DOI] [PubMed] [Google Scholar]
  42. Kothiwale S.; Mendenhall J. L.; Meiler J. BCL::Conf: small molecule conformational sampling using a knowledge based rotamer library. J. Cheminform. 2015, 7, 47. 10.1186/s13321-015-0095-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Hawkins P. C. D.; Skillman A. G.; Warren G. L.; Ellingson B. A.; Stahl M. T. Conformer Generation with OMEGA: Algorithm and Validation Using High Quality Structures from the Protein Databank and Cambridge Structural Database. J. Chem. Inf. Model. 2010, 50, 572–584. 10.1021/ci100031x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Rosales A. R.; Wahlers J.; Limé E.; Meadows R. E.; Leslie K. W.; Savin R.; Bell F.; Hansen E.; Helquist P.; Munday R. H.; et al. Rapid virtual screening of enantioselective catalysts using CatVS. Nat. Catal. 2019, 2, 41–45. 10.1038/s41929-018-0193-3. [DOI] [Google Scholar]
  45. Ingman V. M.; Schaefer A. J.; Andreola L. R.; Wheeler S. E. QChASM: Quantum chemistry automation and structure manipulation. WIREs Comput. Mol. Sci. 2021, 11, e1510 10.1002/wcms.1510. [DOI] [Google Scholar]
  46. Hansen E.; Rosales A. R.; Tutkowski B.; Norrby P.-O.; Wiest O. Prediction of Stereochemistry using Q2MM. Acc. Chem. Res. 2016, 49, 996–1005. 10.1021/acs.accounts.6b00037. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Rosales A. R.; Quinn T. R.; Wahlers J.; Tomberg A.; Zhang X.; Helquist P.; Wiest O.; Norrby P.-O. Application of Q2MM to predictions in stereoselective synthesis. Chem. Commun. 2018, 54, 8294–8311. 10.1039/C8CC03695K. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Petraglia R.; Nicolaï A.; Wodrich M. D.; Ceriotti M.; Corminboeuf C. Beyond static structures: Putting forth REMD as a tool to solve problems in computational organic chemistry. J. Comput. Chem. 2016, 37, 83–92. 10.1002/jcc.24025. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Grimme S. Exploration of Chemical Compound, Conformer, and Reaction Space with Meta-Dynamics Simulations Based on Tight-Binding Quantum Chemical Calculations. J. Chem. Theory Comput. 2019, 15, 2847–2862. 10.1021/acs.jctc.9b00143. [DOI] [PubMed] [Google Scholar]
  50. Pracht P.; Bohle F.; Grimme S. Automated exploration of the low-energy chemical space with fast quantum chemical methods. Phys. Chem. Chem. Phys. 2020, 22, 7169–7192. 10.1039/C9CP06869D. [DOI] [PubMed] [Google Scholar]
  51. Pracht P.; Grimme S. Calculation of absolute molecular entropies and heat capacities made simple. Chem. Sci. 2021, 12, 6551–6568. 10.1039/D1SC00621E. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Seidel T.; Permann C.; Wieder O.; Kohlbacher S. M.; Langer T. High-Quality Conformer Generation with CONFORGE: Algorithm and Performance Assessment. J. Chem. Inf. Model. 2023, 63, 5549–5570. 10.1021/acs.jcim.3c00563. [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Pracht P.; Grimme S.; Bannwarth C.; Bohle F.; Ehlert S.; Feldmann G.; Gorges J.; Müller M.; Neudecker T.; Plett C.; et al. CREST—A program for the exploration of low-energy molecular chemical space. J. Chem. Phys. 2024, 160, 114110. 10.1063/5.0197592. [DOI] [PubMed] [Google Scholar]
  54. Blaney J. M.; Dixon J. S.. Distance Geometry in Molecular Modeling. In Reviews in Computational Chemistry; 1994. 10.1002/9780470125823.ch6. [DOI] [Google Scholar]
  55. Vainio M. J.; Johnson M. S. Generating Conformer Ensembles Using a Multiobjective Genetic Algorithm. J. Chem. Inf. Model. 2007, 47, 2462–2474. 10.1021/ci6005646. [DOI] [PubMed] [Google Scholar]
  56. O'Boyle N. M.; Vandermeersch T.; Flynn C. J.; Maguire A. R.; Hutchison G. R. Confab - Systematic generation of diverse low-energy conformers. J. Cheminform. 2011, 3, 8. 10.1186/1758-2946-3-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  57. Hawkins P. C. D. Conformation Generation: The State of the Art. J. Chem. Inf. Model. 2017, 57, 1747–1756. 10.1021/acs.jcim.7b00221. [DOI] [PubMed] [Google Scholar]
  58. Laplaza R.; Gallarati S.; Corminboeuf C. Genetic Optimization of Homogeneous Catalysts. Chem. Methods 2022, 2, e202100107 10.1002/cmtd.202100107. [DOI] [Google Scholar]
  59. Sobez J.-G.; Reiher M. Molassembler: Molecular Graph Construction, Modification, and Conformer Generation for Inorganic and Organic Molecules. J. Chem. Inf. Model. 2020, 60, 3884–3900. 10.1021/acs.jcim.0c00503. [DOI] [PMC free article] [PubMed] [Google Scholar]
  60. Laplaza R.; Sobez J.-G.; Wodrich M. D.; Reiher M.; Corminboeuf C. The (not so) simple prediction of enantioselectivity– a pipeline for high-fidelity computations. Chem. Sci. 2022, 13, 6858–6864. 10.1039/D2SC01714H. [DOI] [PMC free article] [PubMed] [Google Scholar]
  61. Wodrich M. D.; Laplaza R.; Cramer N.; Reiher M.; Corminboeuf C. Toward in silico Catalyst Optimization. CHIMIA 2023, 77, 139. 10.2533/chimia.2023.139. [DOI] [PubMed] [Google Scholar]
  62. Meli R.; Biggin P. C. spyrmsd: symmetry-corrected RMSD calculations in Python. J. Cheminform. 2020, 12, 49. 10.1186/s13321-020-00455-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
  63. Ess D. H.; Wheeler S. E.; Iafe R. G.; Xu L.; Çelebi-Ölçüm N.; Houk K. N. Bifurcations on Potential Energy Surfaces of Organic Reactions. Angew. Chem., Int. Ed. 2008, 47, 7592–7601. 10.1002/anie.200800918. [DOI] [PMC free article] [PubMed] [Google Scholar]
  64. Heidrich D.; Quapp W. Saddle points of index 2 on potential energy surfaces and their role in theoretical reactivity investigations. Theor. Chim. Acta 1986, 70, 89–98. 10.1007/BF00532206. [DOI] [Google Scholar]
  65. Collins P.; Ezra G. S.; Wiggins S. Index k saddles and dividing surfaces in phase space with applications to isomerization dynamics. J. Chem. Phys. 2011, 134, 244105. 10.1063/1.3602465. [DOI] [PubMed] [Google Scholar]
  66. Harabuchi Y.; Ono Y.; Maeda S.; Taketsugu T.; Keipert K.; Gordon M. S. Nontotally symmetric trifurcation of an SN2 reaction pathway. J. Comput. Chem. 2016, 37, 487–493. 10.1002/jcc.24241. [DOI] [PubMed] [Google Scholar]
  67. Yadav K.; Pradhan R.; Lourderaj U. Influence of second-order saddles on reaction mechanisms. Faraday Discuss. 2022, 238, 183–203. 10.1039/D2FD00026A. [DOI] [PubMed] [Google Scholar]
  68. Gao C. W.; Allen J. W.; Green W. H.; West R. H. Reaction Mechanism Generator: Automatic construction of chemical kinetic mechanisms. Comput. Phys. Commun. 2016, 203, 212–225. 10.1016/j.cpc.2016.02.013. [DOI] [Google Scholar]
  69. Unsleber J. P.; Grimmel S. A.; Reiher M. Chemoton 2.0: Autonomous Exploration of Chemical Reaction Networks. J. Chem. Theory Comp. 2022, 18, 5393–5409. 10.1021/acs.jctc.2c00193. [DOI] [PMC free article] [PubMed] [Google Scholar]
  70. Zhao Q.; Savoie B. M. Deep reaction network exploration of glucose pyrolysis. Proc. Natl. Acad. Sci. U. S. A. 2023, 120, e2305884120. 10.1073/pnas.2305884120. [DOI] [PMC free article] [PubMed] [Google Scholar]
  71. Gilkes J.; Storr M. T.; Maurer R. J.; Habershon S. Predicting Long-Time-Scale Kinetics under Variable Experimental Conditions with Kinetica.jl. J. Chem. Theory Comp. 2024, 20, 5196. 10.1021/acs.jctc.4c00333. [DOI] [PMC free article] [PubMed] [Google Scholar]
  72. MacGillavry C. H.; Fodor G. 120. The stereochemistry of the tropane alkaloids. Part XVI. The configuration of the N-epimeric ethylmethylnortropinium salts. J. Chem. Soc. 1964, 597. 10.1039/jr9640000597. [DOI] [Google Scholar]
  73. Bottini A. T.; De la Camp U. O.; Thut C. C.; Gal J.; Bellettini A. G. Stereochemistry of tropane quaternizations. J. Org. Chem. 1972, 37, 324–326. 10.1021/jo00967a034. [DOI] [Google Scholar]
  74. Otzenberger R. D.; Lipkowitz K. B.; Mundy B. P. Quaternizations in the 8-azabicyclo[4.3.0]non-3-ene series. J. Org. Chem. 1974, 39, 319–321. 10.1021/jo00917a008. [DOI] [Google Scholar]
  75. Babin V.; Taran F.; Audisio D. Late-Stage Carbon-14 Labeling and Isotope Exchange: Emerging Opportunities and Future Challenges. JACS Au 2022, 2, 1234–1251. 10.1021/jacsau.2c00030. [DOI] [PMC free article] [PubMed] [Google Scholar]
  76. Due to their ion-pair character, the energies of TSa and TSb may be dependent on the solvent employed. Here, we have carried out our computations in the gas phase for demonstrative purposes.
  77. Iribarren I.; Trujillo C. Efficiency and Suitability when Exploring the Conformational Space of Phase-Transfer Catalysts. J. Chem. Inf. Model. 2022, 62, 5568–5580. 10.1021/acs.jcim.2c00934. [DOI] [PMC free article] [PubMed] [Google Scholar]
  78. Bannwarth C.; Ehlert S.; Grimme S. GFN2-xTB—An Accurate and Broadly Parametrized Self-Consistent Tight-Binding Quantum Chemical Method with Multipole Electrostatics and Density-Dependent Dispersion Contributions. J. Chem. Theory Comput. 2019, 15, 1652–1671. 10.1021/acs.jctc.8b01176. [DOI] [PubMed] [Google Scholar]
  79. Grimme S.; Bohle F.; Hansen A.; Pracht P.; Spicher S.; Stahn M. Efficient Quantum Chemical Calculation of Structure Ensembles and Free Energies for Nonrigid Molecules. J. Phys. Chem. A 2021, 125, 4039–4054. 10.1021/acs.jpca.1c00971. [DOI] [PubMed] [Google Scholar]
  80. Note that free energy corrections could be adjusted to account for the number of available microstates (see ref (38)) which is not considered here.
  81. The fact that we do not find more conformers pointing down resulting from CREST, which uses biased metadynamics, immediately signals that the interconversion between up and down is not trivial.
  82. Hodgson D. M.; Charlton A.; Paton R. S.; Thompson A. L. C-Alkylation of Chiral Tropane- and Homotropane-Derived Enamines. J. Org. Chem. 2013, 78, 1508–1518. 10.1021/jo3025972. [DOI] [PubMed] [Google Scholar]
  83. Li Y.; Jackson K. E.; Charlton A.; Le Neve-Foster B.; Khurshid A.; Rudy H.-K. A.; Thompson A. L.; Paton R. S.; Hodgson D. M. Asymmetric Induction in C-Alkylation of Tropane-Derived Enamines: Congruence Between Computation and Experiment. J. Org. Chem. 2017, 82, 10479–10488. 10.1021/acs.joc.7b01954. [DOI] [PubMed] [Google Scholar]
  84. These structures correspond to the upward pointing conformations of TSa that are missing from the energetically pruned ensembles selected for DFT reoptimzation. Reoptimization of these conformers using DFT gave a ΔG‡a,0,up ≈ 16.2 kcal/mol for both, well above the 15.3 kcal/mol found for the lowest energy species in the TSa ensemble.
  85. The small difference is due to the identification of an additional high-energy TSa carbonyl-pointing-up conformer that is missing in the DFT refined ensembles.
  86. Chai J.-D.; Head-Gordon M. Long-range corrected hybrid density functionals with damped atom–atom dispersion corrections. Phys. Chem. Chem. Phys. 2008, 10, 6615. 10.1039/b810189b. [DOI] [PubMed] [Google Scholar]
  87. Weigend F.; Ahlrichs R. Balanced basis sets of split valence, triple zeta valence and quadruple zeta valence quality for H to Rn: Design and assessment of accuracy. Phys. Chem. Chem. Phys. 2005, 7, 3297. 10.1039/b508541a. [DOI] [PubMed] [Google Scholar]
  88. Frisch M. J.; Trucks G. W.; Schlegel H. B.; Scuseria G. E.; Robb M. A.; Cheeseman J. R.; Scalmani G.; Barone V.; Petersson G. A.; Nakatsuji H.. et al. Gaussian 16, Revision C.01; Gaussian Inc.: Wallingford, CT, 2016.
  89. Marenich A. V.; Cramer C. J.; Truhlar D. G. Universal Solvation Model Based on Solute Electron Density and on a Continuum Model of the Solvent Defined by the Bulk Dielectric Constant and Atomic Surface Tensions. J. Phys. Chem. B 2009, 113, 6378–6396. 10.1021/jp810292n. [DOI] [PubMed] [Google Scholar]
  90. Pedregosa F.; Varoquaux G.; Gramfort A.; Michel V.; Thirion B.; Grisel O.; Blondel M.; Prettenhofer P.; Weiss R.; Dubourg V.; et al. Scikit-learn: Machine Learning in Python. J. Mach. Learn. Res. 2011, 12, 2825–2830. [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

jz4c01657_si_001.pdf (233.5KB, pdf)

Articles from The Journal of Physical Chemistry Letters are provided here courtesy of American Chemical Society

RESOURCES