Electro-Descriptors for the Performance Prediction of Electro-Organic Synthesis

Article abstract of DOI:10.1002/anie.202014072

Electrochemical organic synthesis has attracted increasing attentions as a sustainable and versatile synthetic platform. Quantitative assessment of the electro-organic reactions, including reaction thermodynamics, electro-kinetics, and coupled chemical pr

Full text of DOI:10.1002/anie.202014072

A Journal of the Gesellschaft Deutscher Chemiker
Angewandte
Chemie
International Edition
www.angewandte.org
Accepted Article
Title: Electro-descriptors for the performance prediction of electro-
organic synthesis
Authors: Yuxuan Chen, Bailin Tian, Zheng Cheng, Xiaoshan Li, Min
Huang, Yuxia Sun, Shuai Liu, Xu Cheng, Shuhua Li, and
Mengning Ding
This manuscript has been accepted after peer review and appears as an
Accepted Article online prior to editing, proofing, and formal publication
of the final Version of Record (VoR). This work is currently citable by
using the Digital Object Identifier (DOI) given below. The VoR will be
published online in Early View as soon as possible and may be different
to this Accepted Article as a result of editing. Readers should obtain
the VoR from the journal website shown below when it is published
to ensure accuracy of information. The authors are responsible for the
content of this Accepted Article.
To be cited as: Angew. Chem. Int. Ed. 10.1002/anie.202014072
Link to VoR: https://doi.org/10.1002/anie.202014072

10.1002/anie.202014072
Angewandte Chemie International Edition
RESEARCH ARTICLE
Electro-descriptors for the performance prediction of electro-
organic synthesis
Yuxuan Chen[a]†, Bailin Tian[a]†, Zheng Cheng[a],[b], Xiaoshan Li[a], Min Huang[a], Yuxia Sun[a], Shuai
Liu[c], Xu Cheng[c], Shuhua Li*[a],[b] and Mengning Ding*[a]
[a]
Yuxuan Chen, Bailin Tian, Zheng Cheng, Xiaoshan Li, Min Huang, Yuxia Sun, Prof. Shuhua Li, Prof. Mengning Ding
Key Laboratory of Mesoscopic Chemistry, School of Chemistry and Chemical Engineering, Nanjing University, Nanjing 210023, China
E-mail: shuhua@nju.edu.cn and mding@nju.edu.cn
[b]
[c]
Zheng Cheng, Prof. Shuhua Li
Institute of Theoretical and Computational Chemistry, School of Chemistry and Chemical Engineering, Nanjing University, Nanjing 210023, China
Shuai Liu, Prof. Xu Cheng
Jiangsu Key Laboratory of Advanced Organic Materials, School of Chemistry and Chemical Engineering, Nanjing University, Nanjing, 210023, China
†These authors contribute equally to this work
Supporting information for this article is given via a link at the end of the document.
Abstract:
time consuming and potentially expensive in the industrial scenarios.
In addition, fewer efforts have been directed to the mechanistic
aspects of electro-organic reactions, such as interfacial electron
transfer kinetics[5b, 8], coupling of homogenous chemical process and
the potential development of heterogeneous electro-catalysts[9], which
could significantly benefit the development of novel synthetic
approaches from the fundamental perspective.
Electrochemical organic synthesis has attracted increasing attentions
from both scientific and industrial communities as a sustainable and
versatile synthetic platform. Quantitative assessment of the electro-
organic reactions, including reaction thermodynamics, interfacial
kinetics and coupled chemical processes, highlights the uniqueness
of electro-synthesis and can lead to the development of analytical tool
to guide their future design. Here we study the thermodynamics,
electro-kinetics and performance (yield) of electro-organic reactions
with different mechanisms, and conclude that electrochemical
parameters such as onset potential, Tafel slope, and effective voltage
can be utilized as multi electro-descriptors for the evaluation of
reaction conditions and the prediction of corresponding reactivities
(reaction yields). An “electro-descriptor-diagram” is generated, where
reactive and non-reactive conditions/substances show distinct
boundary. Successful predictions for reaction outcomes have been
demonstrated using electro-descriptor diagram, or from Machine
Learning (ML) algorithms with experimentally-derived electro-
descriptors. This method represents a promising tool for the data-
acquisition, mechanistic clarification, reaction prediction, and high-
throughput screening of feasible substance and optimal conditions
(solvents, electrolytes, additives and electrode materials) for general
organic electro-synthesis.
Electro-organic research could further benefit from electrochemistry
as an efficient tool for both quantitative analysis and mechanistic
investigation. For example, with mature theories and experimental
practice of Cyclic Voltammetry (CV)[10]
,
a
comprehensive
understanding of electro-kinetics[11], including mass transportation,
interfacial electron transfer and coupled chemical reaction[12], can be
acquired for complicated electro-organic systems to give insights into
the synthetic pathway that helps to optimize the reaction. There are
reported examples utilizing electrochemical analysis such as CV to
determine the redox potentials of different functional groups and
investigate the reaction mechanisms[13], but it still calls for new
strategy that further promote the correlated studies and optimization.
In this work, we have developed a novel analytical method based on
voltammetry for the quantitative and systematic evaluation of electro-
organic reactions, and more importantly to predict substance
reactivities and the reaction yield under specific set of reaction
conditions. We report three experimentally derived electro-descriptors
from easy-to-obtain CV diagram of the electro-organic systems: onset
potential, Tafel slope, and effective voltage, in correlation to the
thermodynamic, electro-kinetic and experimental (empirical) aspects
of the synthetic reaction. An “electro-descriptor diagram” is developed,
in which all high-yield substrates and conditions form a reaction “hot
zone” that can be clearly distinguished from non-reactive sets. This
electro-descriptor approach successfully applies to three electro-
organic reactions with different pathways, with clear identification of
the reactive “hot zone”, enabling accurate prediction (through either
electro-descriptor diagram or ML algorithm) of the reaction yield from
unknown substances, electrode materials and reaction conditions.
The electro-descriptors represent the first experimentally derived
descriptor system for the performance prediction of the organic
reactions, as an alternative to the in silico structure-related descriptors.
Overall, this method represents a promising tool for the prediction and
screening of reactive substrates and optimized reaction conditions.
Introduction
Since Faraday’s pioneering work using current to drive
nonspontaneous reactions[1], electro-organic synthesis has emerged
as
a popular methodology offering advantages in synthetic
sustainability and industrial safety, including mild reaction conditions,
green chemical process, high selectivity, and fine control over reaction
parameters[2]. Numerous electro-organic methodologies have been
developed[3], enabling facile synthesis of building blocks of
pharmaceutical molecules[4], functional materials[5] and chemical
intermediates[6]; and are therefore of critical significance to the
chemical and pharmaceutical industry. However, typical development
of a new synthetic methodology requires the optimization of reaction
conditions (voltage, solvent, additives and temperature) and calls for
high-volume screening in search of feasible substrates[7], both are
1
This article is protected by copyright. All rights reserved.

10.1002/anie.202014072
Angewandte Chemie International Edition
RESEARCH ARTICLE
a
b
j
j
p
E
p
E
eff
eff
Tafel Slope
(E=alog|j|+b)
Onset Potential E
onset
E (V vs. Ref)
Figure 1| Derivation of the electro-descriptors. a. Experimental set-up for an anodic electro-organic synthesis. R (substance in its reduced
form) is first oxidized to O (oxidized from) on the anode (through interfacial electron transfer process, E), which then reacts with X (through
homogenous chemical reaction, C) to form target molecule OX. b. Three electro-descriptors, onset potential, Tafel slope and effective voltage,
in correspondence to the thermodynamic, electro-kinetic and operando (empirical reaction rate) perspectives of the electro-organic systems
respectively, are identified from the CV diagram of different substrates and reaction conditions.
proof-of-concept reaction as it represents a common synthetic
methodology with typical reaction mechanism. Figure 2a
Results and Discussion
demonstrates the electro-oxidation of 1 under controlled-voltage
electrolysis (6.6 V cell voltage), following an EEC mechanism (E
represents heterogeneous electron transfer on electrode and C
Electrochemical parameters from Cyclic Voltammogram. Figure
represents a homogenous chemical reaction). To generate the
1b shows a representative CV curve of an anodic electro-synthetic
electro-descriptors, CV tests were conducted on 29 sets of reaction
system with typical reaction conditions. Three electro-descriptors,
conditions including 17 substrates and 12 entries with different
onset potential, Tafel slope and effective voltage are derived,
solvents and additives for the same model substrate[15h], as listed in
representing three distinct yet correlated dimensions of the
Fig. 2b (also see Supplementary Table S1, S4 and Fig. S1).
electrochemical system: thermodynamic aspect, electro-kinetics and
Voltammograms shown in Fig. 2c and 2d depict CV characteristics of
empirical (in operando) parameters, respectively. For an anodic
typical reactive and non-reactive electro-organic systems. A clear
reaction, onset potential (Eonset) indicates the necessary potential to
distinction can be observed in onset potential (Eonset) of the reactive
oxidize a substrate on the electrode surface: reaction with a higher
and nonreactive ones (Fig. 2c). Entry 1-2 has an Eonset (1.76 V vs.
onset potential is thermodynamically more unfavorable. Tafel slope
Ag/AgCl) that is much higher than Entry 1-4 (1.20 V vs. Ag/AgCl),
serves as an electro-kinetic descriptor for the electron transfer on
which indicates that the oxidation of Entry 1-2 is thermodynamically
electrode. In an EC process (shown in Equation 1 and 2), Tafel slope
unfavorable. From an electro-kinetic perspective, we also noticed that
measures the kinetics of the interfacial electron transfer (Equation 1).
Tafel slopes of all the reactive systems demonstrate an optimal range
from 0.4 V/dec to 0.7 V/dec, in distinct contrast to the non-reactive
substrates (Fig. 2b), indicating the apparent existence of an ideal
electro-kinetics to achieve desired products in electro-organic
A large Tafel slope is correlated with a relatively sluggish kinetic of the
synthesis. A large Tafel slope observed in certain substrate suggests
interfacial electron transfer. Finally, as typical bench research often
a sluggish kinetics that is unfavourable to its oxidation. On the other
followed the protocol of optimizing synthetic conditions for a “model
hand, a small Tafel slope showing fast oxidation kinetics, although
reaction” and consequent exploring the scope of substrates, we define
favouring the production of R and its following reaction, is not always
“effective voltage (current)” that can be extracted from actual electro-
good as it might indicate the occurrence of side reactions (e.g.,
synthetic data for current (voltage) normalization to the model reaction.
oxidation of aryl ring rather than amine) with distinct kinetics. From a
A higher effective voltage indicates that the electrolytic cell requires a
quick assessment on these electrochemical parameters, the distinct
higher driving force to reach the same operando reaction rate, which
features for optimal Eonset and Tafel slope indicate that one simple
factors in empirical synthetic details.
descriptor is not sufficient enough to evaluate the reactivity of an
electro-organic system.
Electro-descriptor analysis of an electro-organic system. We first
studied the phosphonylation of tetrahydroisoquinoline[15] (1) as a
2
This article is protected by copyright. All rights reserved.

10.1002/anie.202014072
Angewandte Chemie International Edition
RESEARCH ARTICLE
b
a
Entry1-1 76%
1.26
1.63
Entry1-2 0%
1.83
1.96
Entry1-3 0%
0.84
1.03
Entry1-4 80%
1.20
1.58
Entry1-5 0%
Entry1-6 89%
1.28
1.96
Eonset
Eeffective
1.11
1.50
0.70
Tafel slope
0.66
0.50
0.27
0.59
0.63
Entry1-7 53%
1.24
1.48
Entry1-8 0%
1.29
1.52
Entry1-9 67%
1.34
2.00
Entry1-10 35%
Entry1-11 trace
Entry1-12 21%
Eonset
Eeffective
1.26
1.55
0.47
2.25
2.60
1.19
1.33
1.86
0.46
Tafel slope
0.43
0.41
0.51
Different P source
55%
Entry1-13 0%
Entry1-14 0%
Entry1-15 30%
Entry1-16 trace
Entry1-17 0%
Eonset
Eeffective
Tafel slope
0.85
1.14
0.28
0.86
1.11
0.23
1.68
2.35
0.85
1.35
1.42
0.45
1.84
3.00
0.65
38%
c
d
1200
Entry1-2
Entry1-4
200
150
100
50
Entry 1-30
Entry 1-33
1000
800
600
400
200
0
Yield: 64%
Eeffective=1.64 V
Yield: 0
irxn
Yield: 80%
Eonset1
0
Yield: 0
Eeffective=2.59 V
-50
Eonset2
-200
0
1
2
3
4
5
0
1
2
3
E (V vs. Ag/AgCl)
E (V vs. Ag/AgCl)
Yield
Yield
e
f
g
6
5
4
3
2
1
0
0.89
0.78
0.67
0.56
0.45
0.33
0.22
0.11
0.0
0.89
0.78
0.67
0.56
0.45
0.33
0.22
0.11
0.0
Entry 1-17
2.8
2.4
2.0
1.6
1.2
0.89
Entry1-17
Reactive
Trace
Nonreactive
Reactive
Trace
Nonreactive
Entry1-5
Entry1-5
0.3
0.6
0.9
1.2
1.5
1.8
0.3
0.6
0.9
1.2
1.5
1.8
Tafel Slope (V/dec)
Tafel Slope (V/dec)
Figure 2| Electro-descriptor diagram of the EEC reaction. a. Schematic illustration of the electro-oxidation of tetrahydroisoquinoline (1). The
quinoline substances went through a two-step electron transfer processes (EE) followed by the reaction with phosphates (C), and the first
electron transfer is the rate determining step. b. The reaction yield and electro-descriptors of different substrates under same reaction condition
optimized for model reaction with 1, different reaction conditions were also screened for model reaction (see Supplementary Table 1). c. CV of
Entry 1-2 and Entry 1-4 demonstrating the importance of onset potential (Eonset): the nonreactive species have a higher onset potential. d. CV
of Entry 1-30 and Entry 1-33 showing that despite close Eonset, the nonreactive entries (Entry 1-30) have a high effective voltage and a low
reaction rate. e. The 3-D electro-descriptor diagram built with three electro-descriptors. A reactive zone is observed, with obvious boundary
between reactive and nonreactive substances. f, g. 2-D electro-descriptor diagram, Onset potential—Tafel slope diagram (f) and Effective
voltage—Tafel slope diagram (g) both show distinct reactive and non-reactive areas defined by reaction conditions with different yields, reaction
yields are quantified by different color in the diagram (color bar on the right).
while effective voltage (current) factors in actual experimental details
Effective voltage (or current) is defined as the voltage (current)
and serves as an effective empirical descriptor.
needed to achieve the same operando current (voltage) as to the
model reactions in a constant potential (current) scenario, which
serves as a normalization to the model reaction. In some cases,
Construction of multi electro-descriptor diagram. With all three
effective voltage (see Fig. S2) leads to similar conclusions as onset
electro-descriptors derived from the voltammogram of each electro-
potential, which is not surprising given that thermodynamically
organic system (with specific substances and reaction conditions, see
favorable reactions often show high overall charge transfer rate (i.e.,
Fig. 2b and Fig. S4), they were then placed into the three coordinates
high faradaic current) under operando conditions. However, in some
system to provide a 3D “electro-descriptor diagram”. As shown in Fig.
cases where Eonset is feeble to discriminate reactivity (see Fig. 2d),
2e, all reactive substances show a clear boundary with non-reactive
effective voltage could well distinguish the operando reaction rate
ones, leading to a distinct reactive “hot zone”, indicating that for high
difference. We later demonstrate that effective voltage offers
yield reaction systems, these corresponding descriptors all fall into an
independent evaluation to certain reactions where Eonset and Tafel
optimal range. This apparent observation in the form of a diagram
slope are difficult to offer ideal distinction in the performance analysis,
provides a convenient and potentially powerful tool to predict the
3
This article is protected by copyright. All rights reserved.

10.1002/anie.202014072
Angewandte Chemie International Edition
RESEARCH ARTICLE
reactivity of a target electro-organic reaction just by acquiring the
voltammogram. Previously, single value of onset potential (or the
apparent current increase) extracted from CV has been generally
discriminating reactive and non-reactive systems. It can be easily
rationalized that the reaction rate of non-reactive Entry 1-30 is slow.
The introduction of effective voltage successfully addresses the
abnormality issues with the Eonset—Tafel slope combination in this
reaction system (Fig. 2f). With Effective voltage—Tafel slope diagram
(Fig. 2g), a better defined reactive hot zone is observed for
phosphonylation of unprotected secondary amine, and “non-reactive”
points (such as Entry 1-5 & 1-17) can be better separated. Therefore,
the combination of Tafel slope and effective voltage can better serve
as the two electro-descriptor system for reactivity evaluation and
prediction in this case.
employed for reactivity evaluation in the electro-organic synthesis[16]
.
Our results reveal that data analysis with multi-dimensional electro-
descriptors provide more accuracy for the reactivity prediction based
on different fundamental perspectives. The 3D electro-descriptor
diagram can further break down to two 2D diagrams, as shown in Figs.
2f and 2g, with additional color representation of reaction yield for
more convenient visualization. Clear boundaries were first noted,
defined by the most “out-bound”, typically “non-reactive” points in the
diagram (shown as white points in Figs. 2f and 2g). The reaction “hot
zone” can still be easily concluded from the 2D electro-descriptor
diagram (Figs. 2f and 2g), where all the reactive substrates and
conditions fall into an optimal 2D area. Reactive point that lies in the
hot zone typically has a higher yield. The shape of reaction hot zone
in Fig. 2f indicates that the combination of two electro-descriptors
(each represents kinetic and thermodynamic aspects), can be visually
more effective to evaluate the reactivity of the system.
Despite the clear separation of “non-reactive” substance/conditions
(with zero yield of desired product) from “reactive” ones, it is also
informative from a synthetic perspective to distinguish whether the
“non-reactive” substance was electrochemically inert or was
converted in a different pathway to give side products. To this end,
crude product analysis was conducted for the typical “non-reactive”
entries (see Supplementary Unreactive entry product analysis
Section). Interestingly, the difference between “zero conversion” and
“side reaction” can be reflected in the electro-descriptor diagram. For
“0 yield” Entry 1-3, 1-13 and 1-14, product analysis show that the aryl
rings in these substances were actually oxidized, rather than the
intended oxidation of amine. As shown in Fig. 2f and 2g (for more
detailed demonstrations in diagram, see Fig. S6), these batches
appear in the bottom-left corner in electro-descriptor-diagrams, due to
the significant small Tafel slopes. It is reasonable to conclude that
such overly fast electro-kinetics indicate an alternative reaction
pathway that leads to unintended side products. On the other hand,
for “0 yield” Entry 1-11 and 1-17, most starting materials remain intact.
Correspondingly, their position in the electro-descriptor diagram is
located more to the top right corner, as a result of relatively large Tafel
slope or effective voltage (see Fig. S6, effective voltages are more
efficient to separate these points as their corresponding CV currents
are small). For these two scenarios, their distinctive characteristics in
the electro-descriptor diagram further demonstrate the application of
this analytical approach to evaluate different reaction pathways for
reactions without intended product. Additionally, for other cases such
as Entry 1-2, 1-5 and 1-20, we also observed the oxidation of starting
materials. These batches correspond to the cases where the electro-
kinetics of side reactions is similar to the desired reactions, which
require other electro-descriptors in addition to Tafel slope (onset
potential for Entry 1-2 and effective voltage for Entry 1-5 and 1-20) for
predictive accuracy. Overall, the product analysis of the “unreacted”
trials in System 1 provides interesting information regarding the
reaction pathways, with complementary perspectives that extend the
analysis/prediction using this electro-descriptors system.
In addition, a linear correlation between Tafel slopes and onset
potentials (as indicated by white arrows in Fig. 2f) was observed. This
correlation indicates an interesting principle in this electro-organic
system: a more thermodynamically favorable reaction, to a certain
extent, relates to a faster electro-kinetic process. Although this is not
theoretically necessary in electrochemical synthesis, the possible
underlying correlation between electro-kinetic and thermodynamic
parameters, most likely appears in systems where the first charge
transfer process on electrode surface is the sore rate-determine step,
raises up a question whether their combination is sufficient to
evaluate/predict electro-organic reactions with varying conditions. We
later demonstrate that for different reaction mechanisms,
thermodynamic, electro-kinetic, and overall reaction rate do not
always have strict correlations. In Fig. 2f, we did notice that the
reactive points are crowded near the edge of the “hot zone”, which
might bring uncertainty when used for prediction in unknown tests.
Moreover, some nonreactive points (such as Entry 1-5 and Entry 1-
17, labeled in Fig. 2f) are located too close to the reactive ones,
indicating the inefficiency using Eonset and Tafel slope to predict the
reaction yield in certain “abnormal” cases.
For bench-top synthesis, it is relevant to measure and compare the
actual reaction rate (operando current) to give the most accurate yield
prediction. To this end, we introduce an empirical descriptor —
effective voltage, which is derived from operando current
normalization, to assess the rate of substrate’s conversion (if no side
reaction). Fig. 2d demonstrates the effective voltage of two entries
with similar Eonset, Entry 1-33 (reactive, 1.64 V vs. Ag/AgCl) and Entry
1-30 (nonreactive, 2.59 V vs. Ag/AgCl), revealing its role in
4
This article is protected by copyright. All rights reserved.

10.1002/anie.202014072
Angewandte Chemie International Edition
RESEARCH ARTICLE
a
b
c
d
Yield
Yield
0.92
2.6
2.4
2.2
2.0
1.8
1.6
1.4
1.2
0.92
0.81
0.69
0.58
0.46
0.35
0.23
0.12
0.0
0
0
2.2
2.0
0.81
0
0.69
0.58
0.46
0.35
0.23
0
0
0
1.8
0
0.83
0.57
0
0.49
1.6
1.4
1.2
0.83
0
0.71
0.82
0.89
0.83
0.92
0.86
0.82
0.83
0
0
0.73
0.47
Reactive
Reactive
Trace
Nonreactive
0.89
0.12
0.0
0
0.73
Trace
0.44
0.44
Nonreactive
0.0
0.2
0.4
0.6
0.8
1.0
0.0
0.2
0.4
0.6
0.8
1.0
Tafel Slope (V/dec)
Tafel Slope (V/dec)
Figure 3| Electro-descriptor diagram of the ECEC reaction. a. A reported case of electrochemical oxidation and following aziridination of
alkenes. The reaction undergoes an ECEC mechanism and the first electron transfer is the rate determining step. b. The electro-descriptor
diagram is built using three electro-descriptors, showing a reactive hot zone with a distinct boundary. c. Onset potential—Tafel slope diagram
shows a reactive zone where the high yield points are at the center of the reactive zone. The reactive substrates tend to have smaller Tafel
slopes. d. Effective voltage—Tafel slope diagram also shows a reactive zone and high-yield points are close to the center. Nonreactive substrate
generally has a higher effective voltage which indicates slow reaction rate in constant voltage electrolysis.
effective voltage is observed in Fig. 3d. For similar reasons, we
Electro-descriptor diagram for ECEC and catalytic EC reactions.
attribute the correlation to that the heterogeneous electron transfer is
We further applied this electro-descriptor tool to investigate the
the rate determining step for this specific reaction. In the dual-
electrochemical aziridination of alkene[17], which is useful protocol to
descriptor diagram (Figs. 3c and 3d), it can also be concluded that
synthesize aziridines for pharmaceutical use[18] and functional
Tafel slope along is sufficient enough to discriminate the reactivity of
materials[19]. The reaction undergoes an ECEC mechanism as shown
different substrates and conditions (in distinction to System 1). This
in Figure 3a. CV tests were conducted at the reaction conditions
result sheds further light on the mechanism of the electrochemical
reported in the reference[5b], and three electro-descriptors were
aziridination of alkene: the interfacial electron transfer is the rate
derived from CV diagrams. As shown in Fig. 3b, even more distinct
determining step, the following chemical reaction and the second
reactive hot zone can be observed in its 3D electro-descriptor diagram.
electron transfer proceeds at a much faster rate, so Tafel slope
Fig 3c shows relatively more diverging correlation between the Eonset
appears to be the dominating electro-kinetic descriptor in this reaction.
and Tafel slope, indicating that even though certain nonreactive points
are thermodynamically favorable for the reaction, they demonstrate
no reactivity (or poor selectivity) due to a sluggish electron-kinetics. In
comparison, a relatively stronger correlation between Tafel slope and
5
This article is protected by copyright. All rights reserved.

10.1002/anie.202014072
Angewandte Chemie International Edition
RESEARCH ARTICLE
Figure 4| Electro-descriptor diagram of the EC reaction. a. Electrochemical dehydrogenation of N‑Heterocycles utilizing organic electro-
catalyst as reported. The reaction undergoes an EC mechanism: mediator 5 is oxidized on the electrode (E) and then diffuses back to the
solution to dehydrate substance 4 (C). b. The electro-descriptor diagram of this EC reaction shows that efficient electro-catalysts form a reactive
zone; medium electro-catalysts are close to the reactive zone and nonreactive points are far away from the reactive area. c. Onset potential—
Tafel slope diagram shows that reactive points have a narrow distribution in onset potential. d. The distribution of reactive and non-reactive
points shows less regularity in Effective voltage—Tafel slope diagram for this mediator mechanism.
organic reactions catalyzed by a redox mediator, the thermodynamic
effect dominates the overall progress. Once the first reactive barrier
of the mediator is satisfied, the reaction proceeds well with different
electrode kinetics. We also noticed that one point with a low effective
voltage and low Eonset is nonreactive. An overly fast electro-kinetics
might indicate side reactions such as destruction/coupling of mediator,
leading to the loss of its catalytic activity. It should be noted that this
electro-organic reaction is performed under constant current mode
instead of constant voltage mode. While it does not significantly alter
the result of Eonset—Tafel slope diagram, the distribution of reactive
points is irregular in Effective voltage—Tafel slope diagram (Fig. 4d).
In a galvanostatic electrolysis, the overall reaction rate is leveled to
the same, thus effective voltage has no strong correlation with the
reaction yield. The different observations from the electro-descriptor
diagram call for a more systematic evaluation of the reactivity from
both constant voltage and galvanostatic modes, when developing a
new electro-organic methodology.
In addition to the direct electrolysis, a large number of electro-organic
synthesis have been conducted in an indirect fashion, with the
addition of organic mediator to reduce the barrier for interfacial charge
transfer[20]. A better understanding of their electro-kinetics helps to sift
out effective mediators under different conditions. Lei and co-workers
reported an electrochemical dehydrogenation of N-Heterocycles
utilizing TEMPO as electro-catalyst under constant current condition,
following a typical EC mechanism (Figure 4a)[21]. CV tests were run to
10 sets of electro-catalysts and conditions reported in the reference
work[21]. A hot zone is also observed in the 3D electro-descriptor
diagram, with trace yield points at the periphery and nonreactive
points in a distant area (Fig. 4b). A band-like reactive hot zone in
Eonset—Tafel slope diagram (Fig. 4c), where all reactive points fall into
a narrow range of onset potential but a wide range of Tafel slope,
shows a significantly distinct reaction kinetics and mechanism (in
comparison to Fig. 2f and Fig. 3c). It reveals that for the electro-
6
This article is protected by copyright. All rights reserved.

10.1002/anie.202014072
Angewandte Chemie International Edition
RESEARCH ARTICLE
Figure 5| Electro-descriptors for machine learning and visualized (diagram) prediction of the reaction performance. a. Machine learning
(ML) analysis using decision tree algorithm to predict the yield of electro-phosphonylation of secondary amine (System 1). A training set of
amines/conditions were randomly chosen to predict the performance of the test set. b. Observed versus predicted plots for System 1 (decision
tree algorithm) and System 2 (k-nearest neighbor algorithm). Leave-one-out approach is applied to evaluate the performance of ML model, all
predicted values are shown in the plot. As observed/predicted yields for non-reactive entries have the same value at (0, 0), uniform random
noise with a size of 4.5 is added to the data (using a python library seaborn) to slightly separate them for better visualization. c. The activity
diagram using three variables derived from Principle Component Analysis (PCA), in which red balls represent the reactive (yield ≥ 20%) entries
while blue balls represent the nonreactive entries. d. Prediction of the reaction yields from different substances (both quinolines and phosphates
in this bimolecular reaction) using pre-established electro-descriptors (effective voltage—Tafel slope) diagram for System 1. e. Prediction of the
reaction yields from different substances and additives using pre-established electro-descriptors (onset potential—Tafel slope) diagram for
System 2. f. Prediction of the reaction yields from different electrode materials using pre-established electro-descriptors (onset potential—Tafel
slope) diagram for System 3. The actual yields (determined from bulk electro-synthesis) in d-f are listed on the side (white points represent zero
yield reactions), which match well to the predictions from their positions in the electro-descriptor diagram. The most reactive points are listed
for visual reference.
and vibrational descriptors to enable yield predictions for Pd-catalyzed
Machine learning and visualized prediction of synthetic
C-N cross-coupling[23a]. The electro-descriptors defined in this work
performance using electro-descriptors. Machine learning (ML)
contain thermodynamic, kinetic and operando information, and
methodologies are becoming effective in the design of synthetic route
therefore could serve as a group of natural descriptors for ML
for various chemicals or in the prediction of product of a given
applications in electro-organic synthesis, with potential compatibility
reaction[14b, 22]. However, due to the complex multidimensionality in an
to the high-throughput analytical technologies. We have utilized
organic synthesis, it is more challenging to efficiently generate enough
different ML algorithms (see Methods) to predict the performance of
data for the employment of ML in reaction performance prediction[23]
Doyle and co-workers has recently applied ML with atomic, molecular
.
reactions (yield) using electro-descriptors as input (combined with
typical structural descriptors) in the phosphonylation of secondary
7
This article is protected by copyright. All rights reserved.

10.1002/anie.202014072
Angewandte Chemie International Edition
RESEARCH ARTICLE
amines (System 1) and aziridination of alkenes (System 2). CUR
decomposition[24] was used to construct uniformly-spaced training set,
in order to guarantee that the training sets are evenly spread over the
range of yields/conditions for more accurate ML prediction (see
Methods for detail). In each case, 60% of data were used as training
set to predict the yield in the remaining 40% test set, as shown in
Figure 5a (System 1). The predictive accuracies with different ML
algorithms were evaluated and the k-nearest neighbor algorithm was
found to provide better performance over random forest, gradient tree
boosting, linear regression and decisoin tree algorithms. The best
results are shown in Fig. S8a (R=0.85, RMSE=17.08) for System 1
and Fig. S8b (R=0.98, RMSE=7.34) for System 2. In order to avoid
overfitting as the size of dataset is relatively small, the performances
of ML models were further evaluated with the leave-one-out approach
(see Table S11 and corresponding discussions). The decision tree
algorithm shows the best performance for System 1, whereas the k-
nearest neighbor algorithm performs best for System 2. This
difference is presumably due to the larger feature dimension in
System 1 than in System 2, and k-nearest neighbor algorithm is more
suitable with low dimensional situation. Moreover, both k-nearest
neighbor and decision tree method are relatively insensitive to
collinearity (see Table S8 and S9). The best results are shown in Fig.
5b (R = 0.77, RMSE = 18.82 for System 1 and R = 0.97, RMSE =
10.16 for System 2). The relative importance of the electro-descriptors
was also evaluated with the random forest algorithm[23a], results are
shown in Fig. S9 and Table S10 (with corresponding discussions).
Overall, these results indicate that our electro-descriptors can be
successfully applied for ML prediction of the reaction performance in
a given electro-organic synthesis.
from newly tested substances in System 1 can be accurately
predicted according to their positions in the different reactivity zones
in diagram. Excitingly, different quinolines and phosphates were both
tested as unknown substances, and results showed that the electro-
descriptor diagram can efficiently predict the performance from either
reactant in a bimolecular reaction. In addition, this approach also
works well in System 2 (used for yield prediction of new substances
and new additives) and System 3 (used for yield prediction of new
electrode materials) with pre-established electro-descriptor diagrams,
results are shown in Fig. 5e and 5f. Other pre-established electro-
descriptor diagrams of system 1-3 are listed in Fig. S11 for reference.
These successful yield predictions from different reaction variables
demonstrate that the electro-descriptor diagram can serve as an
alternative, effective and general tool for the prediction of outcomes
in any electro-organic synthesis with given set of substances and
reaction conditions.
Conclusion
In summary, we have identified and defined three electrochemical
parameters, representing thermodynamic, kinetic and operando
aspects, as multi electro-descriptors for electro-organic reactions.
Electro-descriptor diagram with clear reactive hot zone has been
developed, which can serve as an effective and powerful tool to
evaluate and predict the outcome of a given electro-organic system.
The electro-descriptor diagram applies to three electro-organic
reactions with distinct mechanisms and provide different mechanistic
insights. With electro-descriptors, successful predictions of reaction
yields have been demonstrated with ML algorithms or direct visual
prediction from electro-descriptor diagram. Such analysis and
prediction could help avoid the vast expenses of time and resource
using typical trial-and-error approach in the development of novel
synthetic methods. We expect that with potential integration to the
high-throughput experimentation and fast voltammetry to trace
samplings, this electro-descriptor system will prove to be a universal
and efficient tool for high-throughput screening of substances and
optimization of conditions in future investigations.
To further confirm the correlations between electro-descriptors and
the necessity of multi-descriptor system, Principal Component
Analysis (PCA)[25] was performed to the three electro-descriptors with
additional electrochemical and structural characteristics used in ML
model (see Methods and Table S4-S7). Principal component’s
composition reveals that all three electro-descriptors are crucial in
discriminating reactivities of electro-organic reactions, and Pearson
coefficient shows that there is no self-correlation among the reduced
components. It is worth noticing that with successfully reduced
correlation from PCA, the visualized separation of different reaction
yields (Fig. 5c) is less obvious compared to the electro-descriptor-
diagram (Fig. 2e). This result indicates the uniqueness of electro-
descriptors in the evaluation of reactivities. Despite partial correlations,
the thermodynamic, electro-kinetic, and operando aspect they each
represents fundamentally determines the outcome of an electro-
organic reaction under any given set of reaction conditions. Through
accumulation of reaction data points, advanced computational
methods can be potentially incorporated to help spot reaction hot zone
and discriminate reactivity more efficiently through ML.
Acknowledgements
The authors acknowledge the support by the Fundamental Research
Funds for the Central Universities in China (020514380224), Natural
Science Foundation of Jiangsu Province (BK20180321), and National
Natural Science Foundation of China (21833002 and 21673110).
Keywords: electrochemistry • electro-descriptors•
electroorganic synthesis • machine learning • reaction prediction
In principle, ML algorithms often require large volume of training sets
for optimized precision, which is challenging for typical time-
consuming bench-top reactions such as organic synthesis (recent
advances in microreactors and high-throughput experimentation also
offer tremendous potentials). With easy-to-obtain experimental
electro-descriptors and the reactivity diagram, we further conducted
visualized prediction using the electro-descriptor diagram as an
alternative approach to the ML prediction. As shown in Fig. 5d, with
pre-established electro-descriptor diagram (Fig. 2g), the reaction yield
[1]
[2]
A. S. Lindsey, H. Jeskey, Chem. Rev. 1957, 57, 583-620.
a) M. Yan, Y. Kawamata, P. S. Baran, Chem. Rev. 2017,
117, 13230-13319; b) E. J. Horn, B. R. Rosen, P. S. Baran,
ACS Cent. Sci. 2016, 2, 302-308; c) R. Francke, R. D. Little,
Chem. Soc. Rev. 2014, 43, 2492-2521; d) J. Yoshida, K.
Kataoka, R. Horcajada, A. Nagaki, Chem. Rev. 2008, 108,
2265-2299.
[3]
a) C. A. Paddon, M. Atobe, T. Fuchigami, P. He, P. Watts,
S. J. Haswell, G. J. Pritchard, S. D. Bull, F. Marken, J. Appl.
8
This article is protected by copyright. All rights reserved.

10.1002/anie.202014072
Angewandte Chemie International Edition
RESEARCH ARTICLE
Electrochem. 2006, 36, 617-634; b) J. C. Siu, N. Fu, S. Lin,
Acc. Chem. Res. 2020, 53, 547-560; c) M. Elsherbini, T.
Wirth, Acc. Chem. Res. 2019, 52, 3287-3296; d) K. Mitsudo,
Y. Kurimoto, K. Yoshioka, S. Suga, Chem. Rev. 2018, 118,
5985-5999.
a) B. R. Rosen, E. W. Werner, A. G. O'Brien, P. S. Baran,
J. Am. Chem. Soc. 2014, 136, 5571-5574; b) F. Xu, H. Long,
J. Song, H.-C. Xu, Angew. Chem. Int. Ed. 2019, 58, 9017-
9021.
a) I. C. Stewart, C. C. Lee, R. G. Bergman, F. D. Toste, J.
Am. Chem. Soc. 2005, 127, 17616-17617; b) M. N. Jackson,
S. Oh, C. J. Kaminsky, S. B. Chu, G. Zhang, J. T. Miller, Y.
Surendranath, J. Am. Chem. Soc. 2018, 140, 1004-1010.
a) J. Li, L. He, X. Liu, X. Cheng, G. Li, Angew. Chem. Int.
Ed. 2019, 58, 1759-1763; b) J. H. Simons, J. Electrochem.
Soc. 1949, 95, 47-47.
A. R. Rosales, J. Wahlers, E. Limé, R. E. Meadows, K. W.
Leslie, R. Savin, F. Bell, E. Hansen, P. Helquist, R. H.
Munday, O. Wiest, P.-O. Norrby, Nat. Catal. 2018, 2, 41-45.
C. J. Kaminsky, J. Wright, Y. Surendranath, ACS Catal.
2019, 9, 3667-3671.
a) S. Li, W. Deng, S. Wang, P. Wang, D. An, Y. Li, Q. Zhang,
Y. Wang, ChemSusChem 2018, 11, 1995-2028; b) W. Zhou,
K. Cheng, J. Kang, C. Zhou, V. Subramanian, Q. Zhang, Y.
Wang, Chem. Soc. Rev. 2019, 48, 3193-3228.
a) J. Heypovsky, Proc. Royal Soc. Lon. 1923, 102, 628-640;
b) J. Heyrovsky, C. R. Hebd. Seances Acad. Sci. 1924, 179,
1267-1268; c) R. S. Nicholson, Anal. Chem. 1965, 37,
1351-1355.
D. A. Buttry, M. D. Ward, Chem. Rev. 1992, 92, 1355-1379.
N. Elgrishi, K. J. Rountree, B. D. McCarthy, E. S. Rountree,
T. T. Eisenhart, J. L. Dempsey, J. Chem. Edu. 2017, 95,
197-206.
a) F. Wang, S. S. Stahl, Angew. Chem. Int. Ed. 2019, 58,
6385-6390; b) C. Amatore, A. Jutand, G. Le Duc, Chem.
Eur. J. 2011, 17, 2492-2503; c) C. Amatore, G. Le Duc, A.
Jutand, Chem. Eur. J. 2013, 19, 10082-10093; d) D. P.
Hickey, C. Sandford, Z. Rhodes, T. Gensch, L. R. Fries, M.
S. Sigman, S. D. Minteer, J. Am. Chem. Soc. 2019, 141,
1382-1392; e) Y. Kawamata, J. C. Vantourout, D. P. Hickey,
P. Bai, L. Chen, Q. Hou, W. Qiao, K. Barman, M. A.
Edwards, A. F. Garrido-Castro, J. N. deGruyter, H.
Nakamura, K. Knouse, C. Qin, K. J. Clay, D. Bao, C. Li, J.
T. Starr, C. Garcia-Irizarry, N. Sach, H. S. White, M.
Neurock, S. D. Minteer, P. S. Baran, J. Am. Chem. Soc.
2019, 141, 6392-6402; f) B. K. Peters, K. X. Rodriguez, S.
H. Reisberg, S. B. Beil, D. P. Hickey, Y. Kawamata, M.
Collins, J. Starr, L. Chen, S. Udyavara, K. Klunder, T. J.
Gorey, S. L. Anderson, M. Neurock, S. D. Minteer, P. S.
Baran, Science 2019, 363, 838-845.
[17]
[18]
a) J. Chen, W.-Q. Yan, C. M. Lam, C.-C. Zeng, L.-M. Hu, R.
D. Little, Org. Lett. 2015, 17, 986-989; b) T. Siu, A. K. Yudin,
J. Am. Chem. Soc. 2002, 124, 530-531; c) T. Siu, C. J.
Picard, A. K. Yudin, J. Org. Chem. 2005, 70, 932-937; d) J.
B. Sweeney, Chem. Soc. Rev. 2002, 31, 247-258; e) L.
Degennaro, P. Trinchera, R. Luisi, Chem. Rev. 2014, 114,
7881-7929.
M. Miyazaki, K. Uoto, Y. Sugimoto, H. Naito, K. Yoshida, T.
Okayama, H. Kawato, M. Miyazaki, M. Kitagawa, T. Seki, S.
Fukutake, M. Aonuma, T. Soga, Bioinorg. Med. Chem.
2015, 23, 2360-2367.
I. C. Stewart, C. C. Lee, R. G. Bergman, T. F Dean, J. Am.
Chem. Soc. 2005, 127, 17616-17617.
a) J. E. Nutting, M. Rafiee, S. S. Stahl, Chem. Rev. 2018,
118, 4834-4885; b) E. Steckhan, Angew. Chem. Int. Ed.
1986, 25, 683-701; c) E. Steckhan, Top. Curr. Chem. 1987,
142, 1-69.
Y. Wu, H. Yi, A. Lei, ACS Catal. 2018, 8, 1192-1196.
a) H. Gao, T. J. Struble, C. W. Coley, Y. Wang, W. H. Green,
K. F. Jensen, ACS Cent. Sci. 2018, 4, 1465-1476; b) R.
Gómez-Bombarelli, J. N. Wei, D. Duvenaud, J. M.
Hernández-Lobato, B. Sánchez-Lengeling, D. Sheberla, J.
Aguilera-Iparraguirre, T. D. Hirzel, R. P. Adams, A. Aspuru-
Guzik, ACS Cent. Sci. 2018, 4, 268-276; c) B. Liu, B.
Ramsundar, P. Kawthekar, J. Shi, J. Gomes, Q. Luu
Nguyen, S. Ho, J. Sloane, P. Wender, V. Pande, ACS Cent.
Sci. 2017, 3, 1103-1113; d) B. Sanchez-Lengeling, A.
Aspuru-Guzik, Science 2018, 361, 360; e) M. H. S. Segler,
M. Preuss, M. P. Waller, Nature 2018, 555, 604-610; f) R.
N. Straker, Q. Peng, A. Mekareeya, R. S. Paton, E. A.
Anderson, Nat. Commun. 2016, 7, 10109; g) S. Szymkuć,
E. P. Gajewska, T. Klucznik, K. Molga, P. Dittwald, M.
Startek, M. Bajczyk, B. A. Grzybowski, Angew. Chem. Int.
Ed. 2016, 55, 5904-5937; h) M. H. Todd, Chem. Soc. Rev.
2005, 34, 247-266; i) J. N. Wei, D. Duvenaud, A. Aspuru-
Guzik, ACS Cent. Sci. 2016, 2, 725-732; j) A. F. Zahrt, J. J.
Henle, B. T. Rose, Y. Wang, W. T. Darrow, S. E. Denmark,
Science 2019, 363, eaau5631.
[4]
[5]
[19]
[20]
[6]
[7]
[21]
[22]
[8]
[9]
[10]
[11]
[12]
[13]
[23]
a) D. T. Ahneman, J. G. Estrada, S. Lin, S. D. Dreher, A. G.
Doyle, Science 2018, 360, 186; b) F. Sandfort, F. Strieth-
Kalthoff, M. Kühnemund, C. Beecks, F. Glorius, Chem 2020,
6, 1379-1390.
M. W. Mahoney, P. Drineas, Proc. Nat. Acad. Sci. 2009,
106, 697.
a) B. Moore, IEEE Trans. Automat. Control 1981, 26, 17-
32; b) K. Pearson, The London, Edinburgh, and Dublin
Philosophical Magazine and Journal of Science 1901, 2,
559-572; c) H. Zou, T. Hastie, R. Tibshirani, Journal of
Computational and Graphical Statistics 2006, 15, 265-286.
[24]
[25]
[14]
[15]
a) J. P. Reid, R. S. J. Proctor, M. S. Sigman, R. J. Phipps,
J. Am. Chem. Soc. 2019, 141, 19178-19185; b) J. P. Reid,
M. S. Sigman, Nature 2019, 571, 343-348; c) A. V.
Brethomé, S. P. Fletcher, R. S. Paton, ACS Catal. 2019, 9,
2313-2323; d) S. G. Robinson, M. S. Sigman, Acc. Chem.
Res. 2020, 53, 289-299.
a) G. M. Kosolapoff, J. Am. Chem. Soc. 1947, 69, 2112-
2113; b) E. K. Fields, J. Am. Chem. Soc. 1952, 74, 1528-
1531; c) R. Zhang, Y. Qin, L. Zhang, S. Luo, J. Org. Chem.
2019, 84, 2542-2555; d) J. Dhineshkumar, M. Lamani, K.
Alagiri, K. R. Prabhu, Org. Lett. 2013, 15, 1092-1095; e) B.
Lin, S. Shi, R. Lin, Y. Cui, M. Fang, G. Tang, Y. Zhao, J.
Org. Chem. 2018, 83, 6754-6761; f) S. L. McDonald, Q.
Wang, Angew. Chem. Int. Ed. 2014, 53, 1867-1871; g) M.
Rueping, S. Zhu, R. M. Koenigs, Chem. Comm. 2011, 47,
8679-8681; h) M. Huang, J. Dai, X. Cheng, M. Ding, Org.
Lett. 2019, 21, 7759-7762.
[16]
a) A. J. J. Lennox, J. E. Nutting, S. S. Stahl, Chem. Sci.
2018, 9, 356-361; b) M. Rafiee, F. Wang, D. P.
Hruszkewycz, S. S. Stahl, J. Am. Chem. Soc. 2018, 140,
22-25; c) F. Wang, M. Rafiee, S. S. Stahl, Angew. Chem.
Int. Ed. 2018, 57, 6686-6690; d) J. Li, W. Huang, J. Chen,
L. He, X. Cheng, G. Li, Angew. Chem. Int. Ed. 2018, 57,
5695-5698.
9
This article is protected by copyright. All rights reserved.

10.1002/anie.202014072
Angewandte Chemie International Edition
RESEARCH ARTICLE
Entry for the Table of Contents
Insert graphic for Table of Contents here.
Insert text for Table of Contents here
For typical electro-organic synthetic systems, three electro-descriptors, onset potential (Eonset), Tafel slope and effective voltage
(Eeffective) are identified from CV diagram to characterize thermodynamic, electro-kinetic and empirical aspects of a given electro-organic
reaction, providing an effective and powerful tool for machine learning or direct visual prediction of the reaction yield.
10
This article is protected by copyright. All rights reserved.