Accelerated Reactivity Mechanism and Interpretable Machine Learning Model of N-Sulfonylimines toward Fast Multicomponent Reactions

Krupal P. Jethava, Jonathan Fine, Yingqi Chen, Ahad Hossain, and Gaurav Chopra*

Letter

Accelerated Reactivity Mechanism and Interpretable Machine

Learning Model of N‑Sulfonylimines toward Fast Multicomponent

Reactions

Cite This: https://dx.doi.org/10.1021/acs.orglett.0c03083

Read Online

ACCESS

Metrics & More

Article Recommendations

sı Supporting Information

ABSTRACT: We introduce chemical reactivity ﬂowcharts to help chemists interpret

reaction outcomes using statistically robust machine learning models trained on a

small number of reactions. We developed fast N-sulfonylimine multicomponent

reactions for understanding reactivity and to generate training data. Accelerated

reactivity mechanisms were investigated using density functional theory. Intuitive

chemical features learned by the model accurately predicted heterogeneous reactivity

of N-sulfonylimine with diﬀerent carboxylic acids. Validation of the predictions shows

that reaction outcome interpretation is useful for human chemists.

omputer-assisted organic chemistry has huge potential

for predicting chemical reaction conditions and for

multicomponent reaction to explore heterogeneous reactivity

of N-sulfonylimines by training a human interpretable ML

model that identiﬁes chemical patterns of reactivity to predict

and test the reactivity of new substrates in the multicomponent

reactions prospectively.

−3

automating synthetic chemistry.

In recent years, machine

learning (ML)-based approaches have been successfully

applied to screen libraries of druglike molecules,

for

quantitative structure−activity relationships (QSARs), for

We selected N-sulfonylimines as our model substrate

because N-sulfonylimines are one of the important synthons

in organic chemistry that is used for a variety of chemical

transformations. N-Sulfonylimine is a good source of an

−12

retrosynthetic planning,

for reaction condition predic-

tion, and for reaction prediction. Reactivity prediction is a

hard problem that requires speciﬁc experimental data sets to

14,15

21,22

train ML models.

Traditionally, creating such experimental

electrophilic carbon for radical and nucleophilic addition

databases requires a large number of manual experiments to

check the feasibility of available starting materials to determine

if it reacts together. It is important to note that most ML

methods use existing databases and use only successful

reactions published in the literature as reporting of

unsuccessful reaction is scarcely available. Thus, for a speciﬁc

case study through the careful training of ML models using

both successful and unsuccessful reaction data, it is possible to

train on smaller data sets to test speciﬁc synthetic objectives.

Recently, smaller data sets were used for prediction but

without experimental validation of regiostereoisomers, site

reactions. Several reports have described N-sulfonylimine

reactions in which the reactivity of a carbon−nitrogen double

bond is exploited. Notably, sulfamidate, a cyclic N-

sulfonylimine, has been used to prepare interesting hetero-

cyclic scaﬀolds. Sulfamidate is transformed into a fused

26−31

heterocycle using a Michael addition, cycloaddition,

32−34

35−37

arylation,

alkenylation,

or alkynylation strategy by

leveraging the electrophilicity of sulfamidate (Scheme 1).

However, among the reported synthetic strategies, con-

struction of a direct C−C bond between the imine carbon and

the (het)aromatic partner is underrepresented in the literature.

Speciﬁcally, a synthetic strategy for the direct C−C bond

stereoisomers, diastereoisomers in Diels−Alder reaction, and

radical C−H functionalization of heterocycles. ML models

are helpful in building a chemical library that is otherwise

tedious to explore by screening each reaction to check

substrate feasibility under certain reaction conditions. To

date, there is limited literature precedence for prospective

validation of desired chemical reactions and interpreting its

Received: September 14, 2020

18,19

reactivity using ML methods.

We provide the ﬁrst report,

to the best of our knowledge, of a fast and one-pot

XXXX American Chemical Society

Org. Lett. XXXX, XXX, XXX−XXX

Organic Letters

Letter

Scheme 1. Strategy for Exploring N-Sulfonylimine

from the amine and aldehyde. It is noteworthy that in situ

Reactivity toward Multicomponent Reactions

formation of imines is not always favorable for several reasons.

(

i) A reaction could be partially reversible. (ii) The product

formed could be unstable under certain reaction conditions.

(

iii) Imine formation is highly dependent upon its starting

materials, an aldehyde and an amine, limiting the use of the

previous approaches when we want to use sulfonamides.

Alternatively, aryl chloride can be used instead of carboxylic

acids but has a limited substrate scope. To address these issues,

we provide the ﬁrst report of using N-sulfonylimine as a

substrate for a fast and single-step approach to synthesize

sulfamidate-embedded 1,3,4-oxadiazole using MCR.

We started our investigation with the idea that several types

of cyclic N-sulfonylimines, acyclic N-sulfonylimines, and

aromatic imines are known in the literature and easily

synthesized. To determine the pattern of electrophilicity of

various imines with carboxylic acids, we used Fukui reaction

linkage between sulfamidate and oxadiazole has not been

parameters calculated using density functional theory (DFT)

explored to date. The oxadiazole scaﬀold is a unique presence

and identiﬁed the most suitable imines using the electro-

philicity of the carbon atom (Figure S1). Both cyclic and

acyclic N-sulfonylimines are highly susceptible to the

nucleophilic attack of carboxylic acids according to the

calculation. Therefore, we started our investigation using the

model substrate cyclic N-sulfonylimine (sulfamidate) 1a, which

can be easily synthesized from substituted salicylaldehydes. We

initially selected benzoic acid as the second reaction partner

because of its moderate nucleophilic tendency (Figure S1) and

because it would help with the selection of optimized

conditions for the future use of a chemically diverse range of

carboxylic acids. In addition, the synthesis of other derivatives

with the optimized condition would serve as a training data set

to develop a robust bootstrapped machine learning model that

is human interpretable to guide prospective synthetic experi-

ments using fast MCRs of N-sulfonylimines.

38−40

in many biologically active compounds

and pharmaceut-

ical agents and is a privileged scaﬀold in material science

(

Figure 1). Therefore, sulfamidate-linked 1,3,4-oxadiazole

could be useful for the chemical library design of new druglike

scaﬀolds.

Initially, we performed an optimization study using

sulfamidate (1a) and benzoic acid (2a) to form the desired

product 3a (Table 1 for Scheme 2). The optimized reaction

condition was obtained when dichloromethane (DCM) was

added at the end after taking all starting materials at −10 °C

Figure 1. Compounds with 1,3,4-oxadiazole in medicinal chemistry.

(

see the Supporting Information for Reaction condition

Multicomponent reactions (MCRs) have attracted medici-

nal chemists to prepare chemical libraries of biologically

optimization study).

important molecules and drugs in a rapid manner using two

43−46

Table 1. Optimization of the Synthesis of 1,3,4-Oxadiazole

or more building blocks.

However, MCRs are highly

dependent on the reactivity of starting materials as well as the

presence of the solvents, catalysts, concentrations, and

equivalents of reagents being used. The understanding of

the chemical reactivity of the starting materials for a particular

MCR would be useful for identifying speciﬁc starting materials

for reaction outcomes. A machine learning model for

interpreting chemical reactivity and reaction outcome can

suggest a type of starting material to be used successfully to

obtain the desired product, thereby reducing the waste of

valuable reagents, time, and eﬀort. Keeping this in mind, we

developed a MCR of cyclic or acyclic N-sulfonylimine with

carboxylic acid-containing starting materials to generate

training data and understand chemical reactivity. Previously,

entry

solvent(s)

T (°C)

t (min)

yield (%)

DCE:MeCN

DCE

DCM

Ice-bath

−10

120

30−120

5−30

trace

DCM

48,49

Ramazani et al. and Yudin et al.

individually reported a

four-component reaction yielding a 1,3,4-oxadiazole scaﬀold

using aromatic reactants and peptides, respectively, using N-

isocyanoimino triphenylphosphorane (PINC) as a reaction

partner. In the previous studies, the imine was generated in situ

Reactions at a 0.1 mmol scale. DCE, dichloroethane; MeCN,

acetonitrile. Reaction condition followed as per the literature.

Benzoic acid added at the end. All solid components were taken

together, and solvent was added at the end. Isolated yield.

Org. Lett. XXXX, XXX, XXX−XXX

Organic Letters

Letter

Scheme 2. Synthesis of 1,3,4-Oxadiazole Using Cyclic and

Acyclic Imine with Benzoic Acid with Optimized Reaction

Conditions

eﬀectively, giving 1,3,4-oxadiazole 3e in poor yield. Notably,

bromo derivatives of sulfamidate 1e with benzoic acid (2a) did

not aﬀord the desired product (3f). Nonetheless, when

sulfamidate 1c was reacted with pyridine carboxylic acid 2c,

it formed the expected product with an inseparable isomer in

poor yield. In addition, 4-hydroxybenzoic acid (2d) did not

react with sulfamidate 1c to form the desired product 3h. Next,

we also sought to study the reactivity of other carboxylic acids

with sulfamidates. Therefore, apart from the products shown in

Scheme 3, we also attempted other reactions to study the

Scheme 3. Substrate Scope for Representative Cyclic N-

Sulfonyl-imine with Various Carboxylic Acids Used as

Training Data

Next, we applied the optimized reaction condition to the

acyclic N -sulfonylimine selected using DFT calculations

containing product (see Figures S2 −S7 and Tables S1, S1a,

Figure S1) as it was the second most reactive imine. Also,

we sought to develop a one-pot strategy for preparing 1,3,4-

oxadiazole-linked sulfonamide compared to the four-step

synthesis reported previously. Interestingly, when acyclic

N-sulfonylimine (4a) was reacted with benzoic acid (2a) under

optimized conditions, the reaction aﬀorded the desired

product (5a) in good yield (Scheme 2), but with a longer

reaction time (10 min) for the complete conversion compared

to that of sulfamidate (<5 min). This led us to investigate the

mechanism and the energy proﬁle of plausible intermediates

and transition states formed in this reaction.

To gain mechanistic insight into the chemical reactions, we

used DFT with a polarized continuum model for DCM

solvation at −10 °C to identify transition states and

intermediates for acyclic (Figure S2 ) and cyclic N-sulfonyli-

mines (Figure S3 ). In the proposed mechanism, the

nucleophilic attack by the negatively charged carbon atom of

PINC on the electrophilic center of N-sulfonylimine yields

Intermediate-1. The subsequent Intermediate-2 is formed by a

nucleophilic attack of benzoic acid. Next, intramolecular

cyclization at the carbonyl carbon, formation of the

oxazaphosphetane intermediate, and subsequent removal of

triphenylphosphine oxide yield the desired 1,3,4-oxadiazole-

S1b, and S2 ). All ﬁ les and animation are available at https://

chopralab.github.io/n_sulfonylimine_reactions/.

The energy proﬁle in the proposed mechanism also suggests

why cyclic N-sulfonylimine would be faster than the acyclic N-

sulfonylimine. The ﬁrst half of the reaction mechanism, until

formation of Intermediate-2, is dependent on the chemistry of

acyclic or cyclic N-sulfonylimines (Figures S2 and S3). In both

cases, the rate-limiting step is the attack of the PINC reagent

on the N-sulfonylimine with changes in free energy (ΔG_b_a_r_r_i_e_r)

of 16.3 and 12.6 kcal/mol for the acyclic and cyclic N-

sulfonylimines, respectively. This suggests that both reactions

will occur quickly, and it is expected that the cyclic N-

sulfonylimine would be faster than the acyclic N-sulfonylimine,

which has been shown experimentally.

reactivity of sulfamidate with other carboxylic acids (see

Scheme S1). For example, diﬂuoro arylacetic acid, pyrimidine-

2-carboxylic acid, terephthalic acid, etc., did not react well with

sulfamidates.

The heterogeneous reactivity of sulfamidates made us

intrigued to study the reactivity of acyclic N-sulfonylimines

with carboxylic acids after the successful model reaction shown

in Scheme 2. As shown in Scheme 4, acyclic N-sulfonylimine

substrates were reacted with benzoic acids. Unlike halogenated

sulfamidates, the reaction of halogenated acyclic N-sulfonyli-

mine 4b reacted well with benzoic acid (2a) and 4-bromo-3-

methyl benzoic acid (2e), giving the desired products 5b and

Scheme 4. Substrate Scope for Acyclic N-Sulfonylimine with

Carboxylic Acids Used as Training Data

Next, using the optimized conditions, we started investigat-

ing various sulfamidates and carboxylic acid derivatives. The

reaction of the diethylamine-containing sulfamidate (1b) with

benzoic acid (2a) aﬀorded the desired product 3b in 46%

yield. The reaction of sulfamidate 1b with p-toluic acid (2b)

also formed product 3c but in low yield (17%). In addition,

reaction of methoxy-substituted sulfamidate 1c with benzoic

acid (2a) formed expected product 3d in moderate yield

(

52%). However, naphthyl sulfamidate (1d) did not react

Org. Lett. XXXX, XXX, XXX−XXX

Organic Letters

Letter

Figure 2. Chemical reactivity ﬂowchart. Decision tree model for the substrate scope of the reaction between imine and acid. (A−C) Pictorial

explanation of how the model assigns rules for predicting reactivity. (D) Final bootstrapped model trained on all data with details for each rule

shown in colored boxes. (E−H) Examples of each of these rules using the training data. Box colors represent features shown in panel D, and the

yellow line of the ﬂowchart shows the reaction outcome based on chemical features.

c in 53% and 37% yields, respectively. In addition, the

chemical reactivity ﬂowcharts with features shown in Figure

2A−D that can be used to investigate additional reactions to

ensure that the model is predictive while maintaining

interpretability.

syntheses of 5d and 5e were achieved successfully using

trimethoxy-substituted N-sulfonylimine (4c) and 4-cyano-

substituted N-sulfonylimine (4d), respectively, and they were

well tolerated to aﬀord the desired products 5d and 5e,

respectively (70% and 64% yields, respectively).

The heterogeneous reactivity of cyclic and acyclic N-

sulfonylimines motivated us to develop a machine learning

model using the successful and unsuccessful reactions. We

The ﬁnal model’s chemical reactivity ﬂowchart shown in

Figure 2D is used to interpret and evaluate the substrate scope

of the reaction given the limited number of reactions available

for training. The ﬁrst decision of the model is a check for a

carboxylic acid group adjacent to an aromatic ring where the

lack of this feature leads to a prediction of no reaction (an

example of a failing case when the feature is lacking is shown in

Figure 2E). Next, the model checks for the substitution at

position 6 of the cyclic imine where a substitution at this

position results in the prediction of no reaction (example of a

feature passing case given in Figure 2F). Next, the model

checks for the presence of a substitution at the para position of

the carboxylic acid where the lack of this feature results in the

prediction that a reaction will occur. An example of this feature

failing case is given in Figure 2G. Note that any substitution

other than a para substitution, including heteroaromatics, will

result in the failing case. Finally, the model checks for a

diethylamine substitution at position 7 of sulfonylimine or

whether the carboxylic acid is p-toluic acid. All decisions made

by the ML model were highly conﬁdent except for the ﬁnal

decision (green box in Figure 2H where the only reaction

passing this condition is shown). This decision is supported by

only a p-toluic acid or an amine substitution. Therefore, the

model is unable to distinguish between speciﬁc features that

resulted in a successful reaction. To elucidate chemistry at this

step, we tested the reaction between 1c (N-sulfonylimine

without an amine substitution) and 2b (p-toluic acid) and

noted that the reaction occurred. Conversely, the reaction

trained decision tree models using the extended connectivity

ﬁngerprints (ECFP) of the carboxylic acids and N-

sulfonylimines determined separately. For a ﬁngerprint radius

of 0−3, no major bit collisions were observed. Each reaction is

assigned a reaction ID and a binary condition (“Worked” in

Table S3 ) to represent whether a reaction occurs between the

N-sulfonylimine and carboxylic acid. The goal of the decision

tree models is to predict the “Worked” response using the

ﬁngerprints. Due to the limited amount of reactions available

for training (20 reactions), multiple ﬁngerprint features may

represent the same split in the decision. To address this issue,

the validation of the decision trees was performed 1000 times

to sample the diﬀerent possible models that can be created for

a given ﬁngerprint radius (shown in Figures S8 and S9). We

used bootstrapping of several decision tree models to ensure

the robustness of our model for predicting prospective

experimental outcomes (see the Supporting Information,

ﬁgures and tables for bootstrapped decision tree models, for

details that include an introduction of the machine learning

method used in this work). A Cohen κ statistic of 0.706 was

obtained with an ECFP ﬁngerprint radius of 3 for the ﬁnal

model, suggesting strong intermodel reliability on limited

56,57

training data (20 reactions).

This methodology yields the

Org. Lett. XXXX, XXX, XXX−XXX

Organic Letters

The Supporting Information is available free of charge at

Letter

between 1b (N-sulfonylimine with an amine substitution) and

ASSOCIATED CONTENT

sı Supporting Information

■

d (4-hydroxy benzoic acid) did not occur. These results show

that the ﬁnal decision should check for p-toluic acid and not an

amine substitution. Finally, we tested 2d with the acyclic N-

sulfonylimine 4a to see if this rule applied to acyclic N-

sulfonylimines and noted that the reaction does occur. These

reactions are shown in Scheme 5 and show how our ML

https://pubs.acs.org/doi/10.1021/acs.orglett.0c03083.

Copies of ¹H and ¹³C NMR spectra for all new

compounds, supporting text for the reaction mechanism,

supporting text for the bootstrapping of decision tree

models, and experimental section for synthesis for

validation of the machine learning model (PDF)

Scheme 5. Reactions Performed to Test the ML Model

AUTHOR INFORMATION

Corresponding Author

■

Gaurav Chopra − Department of Chemistry, Purdue Institute for

Drug Discovery, Purdue Institute for Integrative Neuroscience,

Purdue Institute for Inﬂammation, Immunology and Infectious

Disease, Purdue Center for Cancer Research, and Purdue

University Integrative Data Science Initiative, Purdue University,

West Lafayette, Indiana 47907, United States; orcid.org/

000-0003-0942-7898 ; Email: gchopra@purdue.edu

Authors

Krupal P. Jethava − Department of Chemistry, Purdue

University, West Lafayette, Indiana 47907, United States;

orcid.org/0000-0002-4440-3333

Jonathan Fine − Department of Chemistry, Purdue University,

West Lafayette, Indiana 47907, United States

Yingqi Chen − Department of Chemistry, Purdue University,

West Lafayette, Indiana 47907, United States

Ahad Hossain − Department of Chemistry, Purdue University,

West Lafayette, Indiana 47907, United States

strategy can be used to interpret the reactions by MCR similar

to a human chemist. These reactions elucidate that the ﬁnal

decision of the model is supported by two diﬀerent features to

predict reaction outcome. The additional reactions clarify this

rule as they show that a p-methyl substitution in benzoic acid is

responsible for reactivity and that a p-hydroxy substitution

leads to decreased reactivity for cyclic N-sulfonylimine. It also

suggests that acyclic N-sulfonylimines are more reactive than

the cyclic forms as they also react with p-hydroxy benzoic acid.

Therefore, the use of bootstrapping is essential for developing

reliable models for smaller data sets, and the judicious use of

decision tree models on positive and negative reactions will

help human and machine chemists interpret reaction out-

comes.

Complete contact information is available at:

https://pubs.acs.org/10.1021/acs.orglett.0c03083

Author Contributions

K.P.J. and J.F. contributed equally to this work.

Notes

The authors declare no competing ﬁnancial interest.

ACKNOWLEDGMENTS

■

This work was supported, in part, by a Purdue University start-

up package from the Department of Chemistry at Purdue

University, a Ralph W. and Grace M. Showalter Research Trust

award, the Integrative Data Science Initiative award, the Jim

and Diann Robbers Cancer Research Grant for New

Investigators award, and NIH NCATS ASPIRE Design

Challenge awards to G.C. and a Lynn Fellowship to J.F.

Additional support, in part, by the NSF I/UCRC Center for

Bioanalytical Metrology (Award 1916691), a NCATS Clinical

and Translational Sciences Award from the Indiana Clinical

and Translational Sciences Institute (UL1TR002529), and the

Purdue University Center for Cancer Research (NIH Grant

P30 CA023168) is also acknowledged. ChemAxon software

support is acknowledged.

In summary, we have developed a fast MCR of acyclic or

cyclic N-sulfonylimines that was used as a representative

reaction type to develop ML models for predicting reaction

outcomes in a blind prospective manner. The fast and

peculiar reactivity mechanism of N-sulfonylimines was

explained using DFT calculation to understand the critical

role of transition states and intermediates. Bootstrapped

decision tree-based ML models resulted in a chemical

reactivity ﬂowchart that explained the choices made by the

model to predict reaction outcomes. The human interpretable

ML approach can be extended to explore any MCR or any

chemical reaction used to synthesize a library of compounds in

a quick and eﬃcient manner. In addition, the generated

chemical library could be expanded by Suzuki coupling, ring

opening through -SO extrusion, or tosyl group removal

REFERENCES

■

using known reaction conditions. This work provides a

framework for developing fast MCRs, understanding the

underlying reaction mechanism, and identifying chemical

features for predicting the reactivity of components that

results in successful reactions to save valuable time for chemists

to avoid chasing dead-end leads.

1) Corey, E. J.; Wipke, W. T. Computer-Assisted Design of

Complex Organic Syntheses Published by: American Association for

the Advancement of Science Stable URL: Http://Www.Jstor.Org/

Stable/1727162 Digitize, Preserve and Extend Access to Science of

New Computer-Assisted Design of C. Science (Washington, DC, U. S.)

1969, 166 (3902), 178−192.

Org. Lett. XXXX, XXX, XXX−XXX

Organic Letters

Michael Addition/Cyclization Reactions in Batch and Flow. ACS

Letter

(

2) Gao, H.; Struble, T. J.; Coley, C. W.; Wang, Y.; Green, W. H.;

Catal. 2016, 6 (1), 348−356.

(22) Moore, L. C.; Lo, A.; Fell, J. S.; Duong, M. R.; Moreno, J. A.;

Rich, B. E.; Bravo, M.; Fettinger, J. C.; Souza, L. W.; Olmstead, M.

M.; Houk, K. N.; Shaw, J. T. Acyclic Stereocontrol in the Additions of

Nucleophilic Alkenes to Α -Chiral N -Sulfonyl Imines. Chem. - Eur. J.

2019, 25 (52), 12214−12220.

(23) Stubbing, L. A.; Savage, G. P.; Brimble, M. A. N-

Alkylsulfonylimines as Dipolarophiles in Cycloaddition Reactions.

Chem. - Asian J. 2013, 8, 42−48.

(24) Luo, Y.; Hepburn, H. B.; Chotsaeng, N.; Lam, H. W.

Enantioselective Rhodium-Catalyzed Nucleophilic Allylation of Cyclic

Imines with Allylboron Reagents. Angew. Chem., Int. Ed. 2012, 51

Jensen, K. F. Using Machine Learning to Predict Suitable Conditions

for Organic Reactions. ACS Cent. Sci. 2018, 4 (11), 1465−1476.

Controlling an Organic Synthesis Robot with Machine Learning to

Search for New Reactivity. Nature 2018 , 559 (7714), 377 −381.

3) Granda, J. M.; Donina, L.; Dragone, V.; Long, D. L.; Cronin, L.

4) Lavecchia, A. Machine-Learning Approaches in Drug Discovery:

Methods and Applications. Drug Discovery Today 2015, 20 (3), 318−

31.

5) Rodríguez-Per

ez, R.; Miyao, T.; Jasial, S.; Vogt, M.; Bajorath, J.

Prediction of Compound Profiling Matrices Using Machine Learning.

ACS Omega 2018, 3 (4), 4713−4723.

Neural Nets as a Method for Quantitative Structure-Activity

Relationships. J. Chem. Inf. Model. 2015, 55 (2), 263 −274.

6) Ma, J.; Sheridan, R. P.; Liaw, A.; Dahl, G. E.; Svetnik, V. Deep

(

33), 8309−8313.

25) Xiong, X. F.; Zhang, H.; Peng, J.; Chen, Y. C. Direct

Asymmetric Michael Addition of Cyclic N-Sulfonylimines to α ,β -

Unsaturated Aldehydes. Chem. - Eur. J. 2011 , 17 (8), 2358 −2360.

7) Badowski, T.; Molga, K.; Grzybowski, B. A. Selection of Cost-

Effective yet Chemically Diverse Pathways from the Networks of

Computer-Generated Retrosynthetic Plans. Chem. Sci. 2019, 10 (17),

(26) Laha, J. K.; Jethava, K. P. Access to Imidazolidine-Fused

Sulfamidates and Sulfamides Bearing a Quaternary Center via 1,3-

Dipolar Cycloaddition of Nonstabilized Azomethine Ylides. J. Org.

Chem. 2017, 82 (7), 3597−3604.

(

640−4651.

8) Fooshee, D.; Mood, A.; Gutman, E.; Tavakoli, M.; Urban, G.;

Liu, F.; Huynh, N.; Van Vranken, D.; Baldi, P. Deep Learning for

(27) Laha, J. K.; Jethava, K. P.; Tummalapalli, K. S. S.; Sharma, S.

Chemical Reaction Prediction. Mol. Syst. Des. Eng. 2018, 3 (3), 442−

Synthesis of Mono-N-Sulfonylimidazolidines by a 1,3-Dipolar Cyclo-

addition Strategy, as an Alternative to Selective N-Sulfonylation, and

Their Ring Cleavage To Afford 1,2-Diamines. Eur. J. Org. Chem. 2017,

452.

(

9) Szymkuc, S.; Gajewska, E. P.; Klucznik, T.; Molga, K.; Dittwald,

P.; Startek, M.; Bajczyk, M.; Grzybowski, B. A. Computer-Assisted

Synthetic Planning: The End of the Beginning. Angew. Chem., Int. Ed.

28) Hu, P.; Hu, J.; Jiao, J.; Tong, X. Amine-Promoted Asymmetric

017 (31), 4617−4624.

4 + 2) Annulations for the Enantioselective Synthesis of

(

016, 55, 5904−5937.

10) Liu, B.; Ramsundar, B.; Kawthekar, P.; Shi, J.; Gomes, J.; Luu

Tetrahydropyridines: A Traceless and Recoverable Auxiliary Strategy.

Nguyen, Q.; Ho, S.; Sloane, J.; Wender, P.; Pande, V. Retrosynthetic

Reaction Prediction Using Neural Sequence-to-Sequence Models.

ACS Cent. Sci. 2017, 3 (10), 1103−1113.

Angew. Chem., Int. Ed. 2013, 52 (20), 5319−5322.

(29) Kravina, A. G.; Mahatthananchai, J.; Bode, J. W. Enantiose-

lective, NHC-Catalyzed Annulations of Trisubstituted Enals and

Cyclic n-Sulfonylimines via α ,β -Unsaturated Acyl Azoliums. Angew.

Chem., Int. Ed. 2012, 51 (37), 9433−9436.

(

11) Klucznik, T.; Mikulak-Klucznik, B.; Mccormack, M. P.;

Mrksich, M.; Trice, S. L. J.; Grzybowski, B. A.; et al. Efficient

Syntheses of Diverse, Medicinally Relevant Targets Planned by

Computer and Executed in the Laboratory. Chem 2018 , 4 , 522 −532.

(30) Chen, X. Y.; Lin, R. C.; Ye, S. Catalytic [2 + 2] and [3 + 2]

Cycloaddition Reactions of Allenoates with Cyclic Ketimines. Chem.

Commun. 2012, 48 (9), 1317−1319.

(12) Wei, J. N.; Duvenaud, D.; Aspuru-Guzik, A. Neural Networks

for the Prediction of Organic Chemistry Reactions. ACS Cent. Sci.

016, 2 (10), 725−732.

13) Gao, H.; Struble, T. J.; Coley, C. W.; Wang, Y.; Green, W. H.;

(31) Spielmann, K.; Van Der Lee, A.; De Figueiredo, R. M.;

(

Campagne, J. M. Diastereoselective Palladium-Catalyzed (3 + 2)-

Cycloadditions from Cyclic Imines and Vinyl Aziridines. Org. Lett.

Jensen, K. F. Using Machine Learning to Predict Suitable Conditions

for Organic Reactions. ACS Cent. Sci. 2018 , 4 (11), 1465 −1476.

32) Nishimura, T.; Noishiki, A.; Chit Tsui, G.; Hayashi, T.

018, 20 (5), 1444−1447.

14) Segler, M. H. S.; Waller, M. P. Neural-Symbolic Machine

Asymmetric Synthesis of (Triaryl)Methylamines by Rhodium-

Catalyzed Addition of Arylboroxines to Cyclic N-Sulfonyl Ketimines.

J. Am. Chem. Soc. 2012, 134 (11), 5056 −5059.

Learning for Retrosynthesis and Reaction Prediction. Chem. - Eur. J.

017, 23 (25), 5966−5971.

15) Coley, C. W.; Green, W. H.; Jensen, K. F. Machine Learning in

Computer-Aided Synthesis Planning. Acc. Chem. Res. 2018, 51 (5),

281−1289.

16) Beker, W.; Gajewska, E. P.; Badowski, T.; Grzybowski, B. A.

(33) Wang, H.; Jiang, T.; Xu, M. H. Simple Branched Sulfur-Olefins

as Chiral Ligands for Rh-Catalyzed Asymmetric Arylation of Cyclic

Ketimines: Highly Enantioselective Construction of Tetrasubstituted

Carbon Stereocenters. J. Am. Chem. Soc. 2013 , 135 (3), 971 −974.

Prediction of Major Regio-, Site-, and Diastereoisomers in Diels −

Alder Reactions by Using Machine-Learning: The Importance of

Physically Meaningful Descriptors. Angew. Chem., Int. Ed. 2019, 58

14), 4515−4519.

(34) Li, Y.; Yu, Y. N.; Xu, M. H. Simple Open-Chain Phosphite-

Olefin as Ligand for Rh-Catalyzed Asymmetric Arylation of Cyclic

Ketimines: Enantioselective Access to Gem-Diaryl α -Amino Acid

Derivatives. ACS Catal. 2016, 6 (2), 661−665.

(

17) Li, X.; Zhang, S.; Xu, L.; Hong, X. Predicting Regioselectivity in

35) De Munck, L.; Monleon, A.; Vila, C.; Pedro, J. R. Diarylprolinol

Radical C −H Functionalization of Heterocycles through Machine

as a Ligand for Enantioselective Alkynylation of Cyclic Imines. Adv.

Learning. Angew. Chem., Int. Ed. 2020, 59 (32), 13253−13259.

Synth. Catal. 2017, 359 (9), 1582−1587.

18) Ahneman, D. T.; Estrada, J. G.; Lin, S.; Dreher, S. D.; Doyle, A.

36) Huang, Y.; Huang, R. Z.; Zhao, Y. Cobalt-Catalyzed

G. Predicting Reaction Performance in C −N Cross-Coupling Using

Machine Learning. Science (Washington, DC, U. S.) 2018, 360 (6385),

Enantioselective Vinylation of Activated Ketones and Imines. J. Am.

Chem. Soc. 2016, 138 (20), 6571−6576.

(37) Quan, M.; Wang, X.; Wu, L.; Gridnev, I. D.; Yang, G.; Zhang,

86−190.

W. Ni(II)-Catalyzed Asymmetric Alkenylations of Ketimines. Nat.

Commun. 2018, 9 (1), 2258.

19) Nielsen, M. K.; Ahneman, D. T.; Riera, O.; Doyle, A. G.

Deoxyfluorination with Sulfonyl Fluorides: Navigating Reaction Space

with Machine Learning. J. Am. Chem. Soc. 2018, 140 (15), 5004−

(38) Vaidya, A.; Jain, S.; Jain, P.; Jain, P.; Tiwari, N.; Jain, R.; Jain,

R.; K. Jain, A.; K. Agrawal, R. Synthesis and Biological Activities of

Oxadiazole Derivatives: A Review. Mini-Rev. Med. Chem. 2016, 16,

825−845.

(39) Khalilullah, H.; J. Ahsan, M.; Hedaitullah, M.; Khan, S.; Ahmed,

B. 1,3,4-Oxadiazole: A Biologically Active Scaffold. Mini-Rev. Med.

Chem. 2012, 12 (8), 789−801.

5008.

20) Miyabe, H.; Ueda, M.; Naito, T. N-Sulfonylimines as an

Excellent Acceptor for Intermolecular Radical Reactions. Chem.

Commun. 2000, No. 20, 2059−2060.

21) Izquierdo, J.; Pericas, M. A. A Recyclable, Immobilized

Analogue of Benzotetramisole for Catalytic Enantioselective Domino

Org. Lett. XXXX, XXX, XXX−XXX

Organic Letters

Shaquiquzzaman, M. A Review Exploring Therapeutic Worth of 1,3,4-

Letter

(40) Verma, G.; Khan, M. F.; Akhtar, W.; Alam, M. M.; Akhter, M.;

Oxadiazole Tailored Compounds. Mini-Rev. Med. Chem. 2019, 19 (6),

77−509.

41) Katritzky, A. R.; Rees, C. W. Comprehensive Heterocyclic

Chemistry; Elsevier, 2009; Vols. 1−7.

42) Slobbe, P.; Ruijter, E.; Orru, R. V. A. Recent Applications of

Multicomponent Reactions in Medicinal Chemistry. MedChemComm

012, 3, 1189−1218.

43) Rotstein, B. H.; Zaretsky, S.; Rai, V.; Yudin, A. K. Small

Heterocycles in Multicomponent Reactions. Chem. Rev. 2014, 114,

323−8359.

44) Domling, A.; Wang, W.; Wang, K. Chemistry and Biology of

Multicomponent Reactions. Chem. Rev. 2012, 112, 3083 −3135.

45) Saiz, C.; Wipf, P.; Manta, E.; Mahler, G. Reversible

(

Thiazolidine Exchange: A New Reaction Suitable for Dynamic

Combinatorial Chemistry. Org. Lett. 2009, 11 (15), 3170−3173.

(46) Maskrey, T. S.; Frischling, M. C.; Rice, M. L.; Wipf, P. A Five-

Component Biginelli-Diels-Alder Cascade Reaction. Front. Chem.

018, 6, 376.

47) Ramazani, A.; Rezaei, A. Novel One-Pot, Four-Component

Condensation Reaction: An Efficient Approach for the Synthesis of

,5-Disubstituted 1,3,4-Oxadiazole Derivatives by a Ugi-4CR/Aza-

Wittig Sequence. Org. Lett. 2010, 12 (12), 2852−2855.

48) Appavoo, S. D.; Kaji, T.; Frost, J. R.; Scully, C. C. G.; Yudin, A.

K. Development of Endocyclic Control Elements for Peptide

Macrocycles. J. Am. Chem. Soc. 2018, 140 (28), 8763 −8770.

(49) Frost, J. R.; Scully, C. C. G.; Yudin, A. K. Oxadiazole Grafts in

Peptide Macrocycles. Nat. Chem. 2016, 8 (12), 1105 −1111.

50) Belowich, M. E.; Stoddart, J. F. Dynamic Imine Chemistry.

Chem. Soc. Rev. 2012, 41, 2003−2024.

51) Wang, Q.; Mgimpatsang, K. C.; Konstantinidou, M.; Shishkina,

S. V.; Domling, A. 1,3,4-Oxadiazoles by Ugi-Tetrazole and Huisgen

Reaction. Org. Lett. 2019, 21 (18), 7320−7323.

52) Contreras, R. R.; Fuentealba, P.; Galvan, M.; Perez, P. A Direct

(

Evaluation of Regional Fukui Functions in Molecules. Chem. Phys.

Lett. 1999, 304 (5−6), 405−413.

(53) Chang, D. J.; Jeong, M. Y.; Song, J.; Jin, C. Y.; Suh, Y. G.; Kim,

H. J.; Min, K. H. Discovery of Small Molecules That Enhance

Astrocyte Differentiation in Rat Fetal Neural Stem Cells. Bioorg. Med.

Chem. Lett. 2011, 21 (23), 7050 −7053.

(54) Quinlan, J. R. Induction of Decision Trees. Machine Learning

(

986, 1 (1), 81−106.

55) Rogers, D.; Hahn, M. Extended-Connectivity Fingerprints. J.

Chem. Inf. Model. 2010 , 50 (5), 742 −754.

56) Cohen, J. Weighted Kappa: Nominal Scale Agreement

Provision for Scaled Disagreement or Partial Credit. Psychol. Bull.

968, 70 (4), 213−220.

57) McHugh, M. L. Interrater Reliability: The Kappa Statistic.

Biochem. Medica 2012, 22 (3), 276−282.

58) Jethava, K. P.; Fine, J. A.; Chen, Y.; Hossain, A.; Chopra, G.

Accelerated Reactivity Mechanism and Interpretable Machine

Learning Model of N-Sulfonylimines Towards Fast Multicomponent

Reactions. ChemRxiv 2020, DOI: 10.26434/CHEM-

RXIV.12116163.V1.