Machine-Assisted Preparation of a Chiral Diamine Ligand Library and In Silico Screening Using Ab Initio Structural Parameters for Heterogeneous Chiral Catalysts

Article abstract of DOI:10.1002/adsc.202100798

A ligand library containing 31 chiral diamines was synthesized using a flow-based semiautomatic reductive amination system. These ligands were evaluated in a continuous-flow asymmetric 1,4-addition reaction with a heterogeneous Ni catalyst. Based on the experimental results of ab initio DFT calculations, a prediction model for enantioselectivities was successfully constructed. Furthermore, virtual screening of possible ligands was conducted to identify promising structures, which showed good enantioselectivities in experiments. (Figure presented.).

Full text of DOI:10.1002/adsc.202100798

COMMUNICATIONS
doi.org/10.1002/adsc.202100798
Machine-Assisted Preparation of a Chiral Diamine Ligand
Library and In Silico Screening Using Ab Initio Structural
Parameters for Heterogeneous Chiral Catalysts
Tatsuya Kuremoto,a Ren Sadatsune,a Tomohiro Yasukawa,a Yasuhiro Yamashita,a
and Shū Kobayashia,*
a
Department of Chemistry, School of Science, The University of Tokyo, Hongo, Bunkyo-ku, Tokyo 113-0033 (Japan)
E-mail: shu_kobayashi@chem.s.u-tokyo.ac.jp
Manuscript received: June 26, 2021; Version of record online: ■■■, ■■■■
Supporting information for this article is available on the WWW under https://doi.org/10.1002/adsc.202100798
as humans.[4,5] However, the overall system tends to be
large, expensive, and limited in possible operations,
Abstract: A ligand library containing 31 chiral
diamines was synthesized using a flow-based semi-
which hamper the widespread use of such automated
automatic reductive amination system. These li-
methods. By contrast, one of the merits of conducting
gands were evaluated in a continuous-flow asym-
chemical reactions in continuous-flow systems is their
metric 1,4-addition reaction with a heterogeneous
suitability for automation. If a continuous-flow system
Ni catalyst. Based on the experimental results of ab
is simply combined with an autosampler and a fraction
initio DFT calculations, a prediction model for
collector at the inlet and the outlet, respectively, the
enantioselectivities was successfully constructed.
system could be used to conduct multiple reactions
Furthermore, virtual screening of possible ligands
sequentially.[6,7]
was conducted to identify promising structures,
which showed good enantioselectivities in experi-
The experimental evaluation of the performance of
designed metal/ligand catalyst combinations is another
ments.
important part of research and development. Recently,
data-intensive approaches have attracted much atten-
tion from all scientific fields, including synthetic
Keywords: flow chemistry; heterogeneous catalysis;
automatic synthesis; machine learning; virtual screen-
organic chemistry.[8,9] To predict the performance of
catalysts, many models have been developed based on
ing
various molecular parameters such as Hammett,[10]
Taft,[11] Charton,[12] and Sterimol parameters.[13] To
include synergistic nonlinear variations, Sigman et al.
Toward a sustainable society, a paradigm shift from proposed that interlinked effects of parameters be
homogeneous catalysts to heterogeneous catalysts in embedded in molecular vibration modes, and they
organic synthesis is desired since recovery and reuse of demonstrated the utility of this approach for many
catalysts can reduce the environmental burden signifi- chemical systems.[14] We hypothesized that slight
cantly. Furthermore, continuous-flow reactions using variations of the optimized molecular structure also
heterogeneous catalysts have advantages over current reflected complex interactions between substituents
batch reactions in terms of environmental compatibil- and served as relevant parameters. These parameters
ity, efficiency, and safety, and are expected to play can be included in the standard output of molecular
important roles in future fine-chemical production.[1] structure optimization and can be input directly to fit
Therefore, rapid research and development of hetero- the model.
geneous catalysts that can be used for flow reactions is
Catalytic asymmetric CÀ C bond-forming reactions
in strong demand.[2]
are essential for the production of fine chemicals;
The synthesis of catalysts and their associated however, heterogeneous catalysts are less explored
ligands usually relies on time-consuming and laborious than homogeneous catalysts. We recently reported
manual processes, which prevent chemists from con- heterogeneous chiral Ni catalysts[15] for asymmetric
structing large datasets that can be analyzed.[3] There 1,4-addition reactions of 1,3-dicarbonyl compounds
have been several efforts to make “robot chemists” in with nitroalkenes.[16,17] The catalysts showed high
batch systems, which could set up reactions the same activity and enantioselectivity and were applicable for
Adv. Synth. Catal. 2021, 363, 1–6
1
© 2021 Wiley-VCH GmbH
��
These are not the final page numbers!

COMMUNICATIONS
asc.wiley-vch.de
continuous-flow reactions. To further improve the could be synthesized using the same column and
catalytic performance, further screening of chiral repeating the reaction and wash cycle. Changing
ligands on the heterogeneous catalyst is necessary. solutions of starting materials, washing the column,
However, this is time-consuming because catalysts and collecting the solutions after the reactions were
with each ligand need to be prepared individually. We fully automated. Collected fractions were combined,
think that rapid screening could be performed more concentrated, and purified by an automatic column
efficiently using a method in which chiral ligands are chromatography system. Using this reaction system,
flowed to a column packed with a ligand-free we have synthesized using aminosilica (NH-SiO2)
heterogeneous Ni catalyst to construct an active chiral diamine ligands (Figure 1).
species in the column.
We then evaluated the chiral diamine ligand library
Herein, we describe the construction of a semi- in an asymmetric 1,4-addition reaction with a hetero-
automatic system for the synthesis of a chiral diamine geneous chiral Ni catalyst. A Ni-diamine complex
ligand library based on continuous-flow reductive prepared from nickel acetate and N,N’-dibenzylethyle-
amination[18] and evaluate the library in continuous- nediamine was used as a precursor (Scheme 2). N,N’-
flow asymmetric 1,4-addition reactions with heteroge- Dibenzylethylenediamine acted as a dummy ligand and
neous chiral Ni catalysts. Experimental yields and was removed during the calcination step. This complex
enantioselectivities were determined with 31 ligands, and mesoporous silica (MCM-41) were mixed in
which could be synthesized from commercially avail- MeCN for 2 h and then filtered. Obtained solids were
°
able inexpensive aldehydes. Moreover, we have shown calcinated at 450 C for 6 h to afford Ni@MCM-41. Ni
that a reliable prediction model could be constructed loading of the catalyst was determined to be
based on nonempirical DFT calculations without 0.295 mmol/g by ICP-OES analysis. This catalyst did
human interpretation. We have virtually screened all
possible ca. 103 substitution patterns, most of which
are commercially unavailable or highly expensive, in
silico and identified promising ligands, which have
been confirmed to give good enantioselectivities by
additional experiments.
The chiral diamine ligand library was synthesized
by direct reductive amination of (1R,2R)-(À )-1,2-
cyclohexanediamine and various aromatic aldehydes in
a continuous-flow method based on our previous report
(Scheme 1).[18] A column (SUS, 4.6×250 mm diame-
°
ter) packed with Pt/C was heated at 40 C, and a
solution of an aldehyde and a diamine in a mixture of
toluene and ethanol (95:5) was pumped into the
column at 0.2 mL/min. A stream of 20 mL/min of H2,
regulated by a mass-flow controller, was connected to
the system through a T-shaped mixer at the inlet of the
column. The reaction was conducted for 100 min, and
then the column was subsequently washed by flowing
solvent at 1 mL/min for 20 min. Due to the robustness
and long lifetime of the Pt/C catalyst, all the ligands
Figure 1. Experimentally Synthesized Chiral Diamine Ligand
Library.
Scheme 1. Synthesis of a Chiral Diamine Ligand Library.
Scheme 2. Synthesis of Ni@MCM-41.
Adv. Synth. Catal. 2021, 363, 1–6
2
© 2021 Wiley-VCH GmbH
��
These are not the final page numbers!

COMMUNICATIONS
asc.wiley-vch.de
not have activity because the dummy ligand was
burned out.
After optimizing the reaction conditions, a flow
system to evaluate chiral ligands was set up as shown
in Scheme 3. A toluene solution of starting material
and chiral diamine ligand was pumped to a flow
reactor (SUS column, 5×50 mm diameter). The col-
umn was packed with a mixture of Ni@MCM-41
(13.84 μmol) and Celite (0.35 g), and was heated at
Scheme 3. Flow System to Evaluate Chiral Diamine Ligands.
°
45 C. It was found that the results could be affected
by the ligand of previous run even after extensive
washing of the column. Therefore, we prepared the
column packed with a mixture of fresh catalyst and ized in Table 1. Our model could predict the outcome
Celite in each reaction to avoid the contamination of with total seven parameters, two HOMO energy levels,
ligand. To generate active species inside the column, two LUMO energy levels, one bond length, and two
pretreatment of the column with a chiral diamine bond angles (Figure 3). The validity of the model was
ligand solution was performed before starting the confirmed by leave-one-out cross validation to ensure
reaction. The ligand was supplied to the column to the prediction power for the out of training dataset. No
maintain catalytic activity. We performed the reaction Mulliken charges were chosen. We also tried to
for 19 h, and fractions were collected every hour. The construct a model for yield prediction. However, we
fractions collected at 1–2, 2–3, 3–4, and 18–19 h were failed to obtain a valid model, which might indicate
analyzed by 1H NMR spectroscopy to determine yields, that other parameters in addition to aromatic moieties
and all fractions were analyzed by HPLC to estimate of the ligands would be needed to be considered, or
yields and enantioselectivities (Figure 2).
that some of the reactions did not reach steady states.
After constructing the prediction model, we per-
With experimental data in hand, we studied
regression analysis to establish the prediction model. formed virtual screening of possible ligands. Five
We chose the LASSO regression model because it kinds of substituents (H, Me, CF3, F, and Cl) at five
simplified the interpretation of the resulting model. In positions could be assumed based on our training data.
this study, minimal molecular descriptors were selected DFT calculation inputs for all 55 =3125 ligands were
based on their applicability to virtual screening. generated by the program and virtually evaluated
HOMO and LUMO energies (25 energies each), (Figure 4). Among top-performing ligands, we selected
Mulliken charges of atoms, bond lengths, and angles three valid ligands, namely 3,5-dimethyl, 2,3-dimethyl,
were collected by DFT calculations (B3LYP/6-31G(d)) and 3-fluoro-5-methyl substituted ligands, based on
of aromatic moieties of ligands as parameters of chiral commercial availability, for further experimental vali-
ligands. We constructed a model to predict enantiose- dation of the prediction model (Table 2). All three
lectivities, and coefficients of parameters are summar- ligands showed good enantioselectivities among the
Figure 2. Yield and Enantiomeric Excess of the Fraction Collected at 18–19 h. For 3Br the fraction was collected at 12–13 h. a)
Determined by 1H NMR analysis. b) Determined by HPLC analysis.
Adv. Synth. Catal. 2021, 363, 1–6
3
© 2021 Wiley-VCH GmbH
��
These are not the final page numbers!

COMMUNICATIONS
asc.wiley-vch.de
ical reactions that is based on ab initio theoretical
calculation results and is suitable for computational
generation of a vast set of parameters. With this
protocol, we could readily prepare a large dataset of
imaginary ligands for the heterogeneous chiral cata-
lysts and virtually screened many possible ligands to
successfully identify promising ligands worth evaluat-
ing experimentally. To our knowledge, this is the first
study that allowed suitable heterogeneous chiral cata-
lysts to be predicted based on machine-assisted library
synthesis and in silico virtual screening. Our total
system, readily available continuous-flow-based auto-
matic synthesis, combined with analysis and in silico
screening, paves the way for next-generation research
in synthetic organic chemistry.
Table 1. Coefficients of Parameters in the LASSO Model.
Parameter
Coefficient
Parameter
Coefficient
HOMO21
HOMO22
LUMO7
À 0.03
À 0.33
5.52
R5
A3
A5
À 7.83
À 1.85
3.98
LUMO14
À 4.03
Experimental Section
Procedure for the Synthesis of Chiral Diamine
Ligand Library
Figure 3. Correlation between Experimental and Predicted
A 250×4.6 mm diameter SUS column was packed with a
previously prepared mixture of Pt/C (5%, 0.39 g) and Celite
(1.00 g). Blank solvent (toluene/ethanol=95:5) was flowed at
1.0 mL/min at room temperature for several minutes until the
system had been stabilized. A H2 gas stream (20 mL/min) was
connected through a T-shape mixer and the flow rate of blank
solution was set at 0.2 mL/min. The column was heated at
Enantioselectivities.
°
40 C. After stabilization of the system, inlet of the flow was
put into the reaction mixture, solution of chiral diamine
(50 mM) and aromatic aldehyde (2.2 eq.) in toluene/ethanol=
95:5. The reaction was conducted for 100 minutes and the
collected eluent was concentrated and purified with automatic
chromatography system (NHÀ SiO2, hexane/EtOAc gradient).
Immediately after the reaction, the inlet of the flow was put into
the blank solvent to wash the column for reuse. The blank
solvent was flowed at 1.0 mL/min for 20 minutes. After the
washing, the inlet of the flow was put into the next reaction
solution to conduct the following synthesis.
Figure 4. Histogram of the Predicted Enantioselectivities of
Virtual Screening Set.
Procedure for the Synthesis of Ni@MCM-41
Table 2. Evaluation of New Ligands.
R
Experimental ee (%)
Predicted ee (%)
In a 50 mL two-necked round-bottom flask, NiBr2 (0.8 mmol)
and N,N’-dibenzylethylenediamine (0.176 mmol) were stirred in
3,5-Me
2,3-Me
3-F-5-Me
73
80
75
86.01
76.66
78.71
°
MeCN (40 mL) at 85 C for 3 h. MCM-41 (2.0 g), which was
°
used after drying at 100 C for 1 h under reduced pressure, was
added and stirred for 1 h. The reaction mixture were filtrated
and obtained solid was washed with DCM and dried under
°
reduced pressure. Solid material was calcinated at 450 C for
6 h under air. The Ni loading of obtained catalyst was
determined by ICP analysis.
dataset based on experimental results. Furthermore, it
was also predicted that the highest enantioselectivity
using these types of chiral ligands was ca. 90% ee, and
that other types of ligands would be needed to be
found to further improve the enantioselectivity.
Procedure for the Evaluation of Ligand Library in
Asymmetric 1,4-Addition Reaction
In conclusion, we have set up a semiautomatic
synthesis and evaluation system for continuous-flow
reactions with heterogeneous chiral catalysts. The
system is inexpensive and configurable. Moreover, we
have developed a novel analysis procedure for chem-
A 5×50 mm diameter SUS column was packed with a
previously prepared mixture of Ni@MCM-41 (Ni: 13.84 μmol)
and Celite (0.35 g). Blank toluene was flowed at 1.0 mL/min. at
°
45 C for several minutes. Ligand solution (in toluene, 2 mM,
Adv. Synth. Catal. 2021, 363, 1–6
4
© 2021 Wiley-VCH GmbH
��
These are not the final page numbers!

COMMUNICATIONS
asc.wiley-vch.de
°
15 mL) was flowed at 1.0 mL/min. at 45 C for 2 h in a
circulation manner. After pretreatment of column with ligand,
inlet of the flow was put into the reaction mixture, solution of
dimethyl malonate (1.5 eq.), β-nitrostyrene (0.1 M), chiral
ligand (1 mol%), and 1,2,4,5-tetramethylbenzene (internal
standard, 0.5 eq.) in toluene. The reaction was performed for
19 h and fractions were collected every hour. Fractions collected
at 1–2, 2–3, 3–4, and 18–19 h after starting the reaction were
[5] S. Steiner, J. Wolf, S. Glatzel, A. Andreou, J. M. Granda,
G. Keenan, T. Hinkley, G. Aragon-Camarasa, P. J.
Kitson, D. Angelone, et al., Science 2019, 363,
eaav2211.
[6] C. W. Coley, D. A. Thomas, J. A. M. Lummiss, J. N.
Jaworski, C. P. Breen, V. Schultz, T. Hart, J. S. Fishman,
L. Rogers, H. Gao, et al., Science 2019, 365, 557.
[7] T. Hardwick, N. Ahmed, Chem. Sci. 2020, 11, 11973–
11988.
1
analysed by H NMR to determine yield. All fractions, after the
treatment with silica gel, were analysed by HPLC to determine
ee.
[8] A. F. Zahrt, S. V. Athavale, S. E. Denmark, Chem. Rev.
2020, 120, 1620–1689.
[9] F. Strieth-Kalthoff, F. Sandfort, M. H. S. Segler, F.
Glorius, Chem. Soc. Rev. 2020, 49, 6154–6168.
[10] L. P. Hammett, Chem. Rev. 1935, 17, 125–136.
[11] R. W. Taft, J. Am. Chem. Soc. 1952, 74, 2729–2732.
[12] M. Charton, J. Am. Chem. Soc. 1975, 97, 1552–1556.
[13] K. C. Harper, E. N. Bess, M. S. Sigman, Nat. Chem.
2012, 4, 366–374.
Acknowledgements
This work was partially supported by a Grant-in-Aid for
Science Research from the Japan Society for the Promotion of
Science, the Ministry of Education, Culture, Sports, Science and
Technology, and the Japan Science and Technology Agency. T.
K. acknowledges the MERIT program, The University of Tokyo.
[14] A. Milo, E. N. Bess, M. S. Sigman, Nature 2014, 507,
210–214.
[15] H. Ishitani, K. Kanai, W. J. Yoo, T. Yoshida, S.
Kobayashi, Angew. Chem. Int. Ed. 2019, 58, 13313–
13317; Angew. Chem. 2019, 131, 13447–13451.
[16] D. A. Evans, D. Seidel, J. Am. Chem. Soc. 2005, 127,
9958–9959.
[17] D. A. Evans, S. Mito, D. Seidel, J. Am. Chem. Soc. 2007,
129, 11583–11592.
[18] B. Laroche, H. Ishitani, S. Kobayashi, Adv. Synth. Catal.
2018, 360, 4699–4704.
References
[1] S. Kobayashi, Chem. Asian J. 2016, 11, 425–436.
[2] K. Masuda, T. Ichitsuka, N. Koumura, K. Sato, S.
Kobayashi, Tetrahedron 2018, 74, 1705–1730.
[3] Y. Shi, P. L. Prieto, T. Zepel, S. Grunert, J. E. Hein, Acc.
Chem. Res. 2021, 54, 546–555.
[4] K. Machida, Y. Hirose, S. Fuse, T. Sugawara, T.
Takahashi, Chem. Pharm. Bull. 2010, 58, 87–93.
Adv. Synth. Catal. 2021, 363, 1–6
5
© 2021 Wiley-VCH GmbH
��
These are not the final page numbers!

COMMUNICATIONS
Machine-Assisted Preparation of a Chiral Diamine Ligand
Library and In Silico Screening Using Ab Initio Structural
Parameters for Heterogeneous Chiral Catalysts
Adv. Synth. Catal. 2021, 363, 1–6
T. Kuremoto, R. Sadatsune, T. Yasukawa, Y. Yamashita, S.
Kobayashi*