Data-driven finding of organic anode active materials for lithium-ion batteries from natural flower scent products using capacity predictors

Abstract

Organic electrode-active materials are important for the development of metal-free, high-performance energy storage devices without resource consumption. Further exploration at an accelerated pace is necessary to find new compounds that exhibit high performances in an infinite number of organic molecules. Exploration based on professional experience, research expertise and intuition has its limitations. In the present work, potential compounds for organic anode active materials are efficiently found in the natural products of flower scents. A total of 65 potential compounds with conjugated moieties are extracted from the original data by the initial screening. Prior to the experiments, eight compounds are selected as candidates using the capacity predictor. Two compounds, 1,4-dichlorobenzene and 6-methyl-2-pyridinecarboxyaldehyde, actually exhibit higher specific capacities of 532 and 293 mA h g⁻¹ with subtraction of the capacity originating from conductive carbon at the current density of 100 mA g⁻¹. Polymerizable structural analogues are found from further exploration. The polymer of pyrrole-2-carboxyaldehye as an analogue exhibits a specific capacity of 934 mA h g⁻¹ at 100 mA g⁻¹. Thus, a couple of new potential anode active materials have been successfully found by efficient exploration using the capacity predictor in natural products.

---

1. Introduction

Organic energy storage contributes to saving resources amid the rapidly growing demand for secondary batteries.^1–6 As an alternative to inorganic electrode-active materials, organic materials can achieve superior performance, such as capacity, through molecular design.^7–14 The capacity has been enhanced by molecular design based on professional expertise and research experience (Fig. 1a and Table S1 in the ESI†).^15–37 In recent years, already reported redox-active moieties have been introduced into crystalline frameworks, such as metal–organic and covalent organic frameworks (MOFs and COFs).^38–40 The discovery of new unknown compounds is required to further enhance electrochemical performances. However, manual exploration among the infinite number of organic molecules is not a realistic approach due to the time and effort consumption. If electrochemical performances are predicted, experiments can be prioritized to achieve the accelerated discovery of new compounds. In the present work, new anode active materials are identified in the natural products using capacity predictors (Fig. 1).

Fig. 1 Summary of recent and current works on organic anode active materials. (a) Relationship between current density and specific capacity of low-molecular-weight compounds (circles) and polymers (squares) used as anode active materials in the previous studies.^{15–37,60,68} The letters in the circles and squares correspond to literature studies listed in Table S1 in the ESI.†^{15–37,60,68} (b) Relationship between year and number of papers related to the organic anode.⁴⁸ (c) Exploration of new anode active materials among 62 compounds of flower scent. (d and e) Extraction of eight potential compounds using capacity predictors. (f) Additional exploration of structural analogues. (g) Polymerization of the discovered compound.

Organic anode active materials have been studied since the discovery of conductive polymers, such as polyacetylene.^41–45 However, these compounds were not widely used because of structural instability and redox potentials. In the 2000s, enhanced performance was achieved by new reaction schemes, including lithium alkoxylation of the carbonyl groups and the superlithiation of the conjugated carbons.^46,47 Based on these new schemes, various conjugated molecules and/or carbonyl compounds exhibited high specific capacities (circles in Fig. 1a and Table S1 in the ESI†).^15–37 Polymerized structures provided enhanced performance (squares in Fig. 1a). However, superior capacity is not achieved for all conjugated and/or carbonyl compounds. Various parameters, such as reaction potential, specific capacity, and cycle stability, are significant for the active materials of batteries. The reaction potential of organic anode active materials is not clearly defined but broadened within the range of 0–1.5 V vs. Li/Li⁺ because multiple Li⁺ ions are reacted within the potential range. The specific capacity is an important factor for the performance. Nevertheless, the specific capacity is not easily predicted based only on the molecular structures determined from our research experience. In addition, the number of published papers indicates that the discovery of new potential compounds has slowed down since 2020 (Fig. 1b).⁴⁸ Therefore, widening the search space is required to find new compounds. Researchers have found new functional and useful compounds in natural products. For example, itaconic acid and lignin were found to be active materials of lithium-ion batteries.^23,49,50 In the present work, we have explored new potential anode active materials in natural products of flower scents (Fig. 1c). New active materials are explored in a different field. Low molecular-weight and simple compounds can be found in flower scents because of the volatility, whereas other natural products have high molecular weight and are complex. If potential compounds are found, the stable active materials can be prepared by polymerization. This exploration was accelerated using the capacity prediction models constructed in our previous works (Fig. 1d).^51,52 Two new potential compounds, such as 1,4-dichlorobenzene and 6-methyl-2-pyridinecarboxyaldehyde, were found from a limited number of experiments (Fig. 1e). Moreover, the polymerization of a structural analogue, pyrrole-2-carboxyaldehye, exhibited enhanced capacity compared with that in the reported compounds (Fig. 1f, g and S1 in the ESI†). In previous works, conjugated carbonyls and thiophene derivatives were the main compounds in anode active materials (Table S1 in the ESI†). The present work has found dichlorobenzene and pyrrole derivative as new anode active materials.

Data-driven approaches have been used for the exploration of materials.^53–59 A variety of machine-learning (ML) algorithms have been studied to predict the structures and properties of new molecules and materials under sufficient data sets. In recent years, high-throughput and robotic experiments have been used to increase the data size and accelerate the number of experiments.^60–62 In addition to the data collecting and mining steps, the efficient exploration of new molecules and materials is a significant step for accelerated discovery in infinite chemical space. By comparison, manual explorations based on professional experiences and intuition are limited. Although numerous potential compounds can be proposed by the in silico automatic generation of molecules, both unprecedented and rational molecules are not easily derived using such tools without professional expertise. Selecting an appropriate search space is the key to succeed in this exploration. Our group has constructed various prediction models based on small data for process optimization and materials exploration.^63–67 However, the search space was limited in the conditions and materials proposed based on our experience. Widening the search space remains a challenge for our method to find new compounds. Here, we focused on molecules in a different field to find unprecedented, rational, and realistic ones. Anode active materials were explored in different compounds of flower scents. Whereas a typical method is exploration of derivatives based on the reported high-performance materials, this crossover exploration has the potential for finding new unprecedented materials.

2. Results and discussion

2.1. Extraction of the potential organic anode active materials

In ca. 2000 compounds of flower scents, such as lilac, marigold, and hyacinth, 62 molecules (F01–F62) with conjugated moieties were selected by an initial screening (Fig. 1c and Scheme S1 in the ESI†). As the molecules having only aliphatic alkyl chains are not redox active, such molecules were not selected to save the calculation cost for the capacity prediction. In our previous works,^51,52,63 the prediction model generations 1–3 (G1–G3) were constructed to estimate the specific capacity as an anode active material based on the molecular structures. The specific capacity as an anode active material was collected based on our own charge–discharge measurement of organic compounds, such as conjugated carbonyls, under the same conditions. The specific capacities of 16, 25, and 36 compounds were used as objective variables (y) for the construction of models G1, G2, and G3, respectively. About 20 physicochemical parameters, such as the energy levels of molecular orbitals, melting point, dipole moment, and solubility parameter, were set as explanatory variables. As the dataset was small for conventional ML algorithms, sparse modeling for small data (SpM-S) was used to construct a linear regression model. The descriptors were selected by exhaustive search with linear regression (ES-LiR) combined with our chemical insights to supplement the data size. The validity of the used descriptors was discussed with chemical insights and studied in a data-scientific manner.^51,52 The prediction accuracy was improved for the later models G2 and G3 with increasing data size.^51,52 Prior to the charge–discharge measurement, the predicted capacity (C_pred/mA h g⁻¹) of F01–F62 was calculated using the models G2 and G3 (eqn (1) and (2), respectively).^51,52

y = 20.4x₄ − 307.6x₁₆ + 303.2x₂₂ − 9.13x₂₃ + 12.4x₂₅ + 40.3x₃₅ + 218.9 (1)

y = 164.6x₂ + 58.0x₂₀ + 116.8x₂₅ + 98.5x₂₈ − 280.1x₃₃ + 296.9x₃₆ + 229.9 (2)

In these models, the objective variable (y) is C_pred. The explanatory variables x_n with discrete numbers (n = 2, 4, 16, 20, 22, 23, 25, 28, 33, 35, and 36) are the following descriptors: one and three energy levels (E) higher than that of the lowest unoccupied molecular orbital (LUMO, E_LUMO) (x₂, x₄), molecular weight (x₁₆), the number of carbonyl groups (x₂₀), the number of the occupied orbitals (N_orb) lower than the work function of lithium (Φ_Li) and energy level (E) E = 0 (x₂₂: N_orb, E_LUMO0 ≤ E < Φ_Li, x₂₃: N_orb, E_LUMO0 ≤ E < 0), Hansen solubility parameter (HSP) distance between the target compound and electrolyte solution (x₂₅), HSP hydrogen-bonding (δ_H) term (x₃₃), the number of oxygen in the heteroaromatic rings (x₃₅), and the ratio of the number of heteroatoms (N, S, O) to the total number of carbons (R_N,S,O/C, x₃₆). In eqn (1) and (2), the coefficients are converted into the normalized frequency distribution, namely mean 0 and standard deviation 1, to quantify the weight of each x_n to y.

The C_pred of F01–F62 was calculated from each x_n using eqn (1) and (2) (Fig. 2a and Table S2 in the ESI†). Fig. 2a summarizes the distribution of C_pred. The model G3 exhibited higher C_pred than the model G2. The model G3 is preferred for finding potential new compounds with a high specific capacity. In contrast, the model G2 is not preferred because the predicted capacity is limited within the range lower than 500 mA h g⁻¹. However, both models were used for exploration in a completely new search space in the present work. The targeted compounds were selected in terms of the high predicted capacity, bulk-scale commercial availability, and stability in the solid state at room temperature. The selected compounds were F21 (Rank 13, C_pred = 337 mA h g⁻¹), F27 (Rank 11, C_pred = 374 mA h g⁻¹), and F50 (Rank 19, C_pred = 311 mA h g⁻¹) using the model G2 and F5 (Rank 3, C_pred = 820 mA h g⁻¹), F52 (Rank 6, C_pred = 609 mA h g⁻¹), and F58 (Rank 8, C_pred = 592 mA h g⁻¹) using the model G3 (Fig. 2b and c). In addition, F12 and F39 were extracted from both predicted results using models G2 and G3 (Fig. 2d).

Fig. 2 Capacity prediction and extracted potential compounds for the experiment. (a) Histogram of C_pred using models G2 (orange) and G3 (yellow) for compounds F01–F62. (b) Potential compounds and their C_pred extracted using models G2 (b), G3 (c), and both (d). All compounds and C_pred are listed in Scheme S1 and Table S2 in the ESI,† respectively.

2.2. Electrochemical properties of the potential organic anode active materials

The specific capacity was measured for these eight commercially available compounds (Fig. 2b–d, 3 and S2 in the ESI†). The detailed method is described in the ESI.† Solid powders of these compounds, acetylene black (AB) as a conductive carbon, and polytetrafluoroethylene (PTFE) as a binder were mixed by the weight ratio of 3 : 6 : 1 in a mortar without the addition of a dispersant. The resultant mixture was pasted onto a copper mesh as a current collector, and then pressed to prepare the working electrode. The specific capacity was measured in a beaker cell filled with an electrolyte solution of ethylene carbonate (EC) and diethyl carbonate (DEC) containing lithium perchlorate (LiClO₄). Metallic lithium (Li) was used as the counter and reference electrodes. The capacity originating from the AB conductive carbon was subtracted to estimate the actual corrected specific capacity of the active materials (Fig. S2 in the ESI†). The actual discharge capacities at the second cycle were 532 mA h g⁻¹ for F5, 293 mA h g⁻¹ for F12, 79.1 mA h g⁻¹ for F27, 61.5 mA h g⁻¹ for F50, and 93.1 mA h g⁻¹ for F52 at 100 mA g⁻¹ (Fig. 3a). The capacity was measured for three different samples (the sample number N = 3, Fig. S2 and Table S3 in the ESI†). As these five samples show the actual specific capacity and reproducibility, the highest values are displayed here. In contrast, the others (F21, F39, F58) exhibited low specific capacity (Fig. S2 and Table S3 in the ESI†). The capacity at the first cycle was not used because a large irreversible capacity was included with the formation of a solid electrolyte interphase (SEI) (Fig. S3 in the ESI†). F5, F12, F27, F50, and F52 retained their specific capacities for 10 cycles (Fig. 3b and Table S3 in the ESI†).

Fig. 3 Electrochemical properties of the eight potential compounds. (a) Representative charge–discharge curves of F5, F12, F21, F27, F39, F50, F52, and F58 at 100 mA g⁻¹ and of the reference conductive carbon (AB) at 50 mA g⁻¹ at the second cycle. (b) Relationship between the cycle number and actual (corrected) specific capacity for F5, F12, F27, F50, and F52. Detailed data are summarized in Fig. S2 and Table S3 in the ESI.†

Here, we focus on the top two compounds, F5 and F12. Based on the specific capacity, F5 and F12 were reacted with 2.9 and 1.3 Li⁺ per molecule, respectively. The SEI formation at the first cycle and stable redox reactions in the following cycles were observed by cyclic voltammetry (CV) within a potential range lower than 1.5 V vs. Li/Li⁺ (Fig. S4 in the ESI†). In this manner, the two potential compounds F5 and F12 were efficiently found within a limited number of experiments.

In the screening step, the cycle performance was measured for 10 cycles to ensure capacity data under temporal stable charge–discharge reactions without rapid dissolution into the electrolyte. F5 and F12 retained around 90% capacity after 10 cycles (Table S3 in the ESI†). The decrease of ca. 10% can be caused by the dissolution in the electrolyte solution. However, long-term stability is not expected for these low-molecular-weight compounds. If polymerizable compounds are found in the additional exploration of the derivatives, the polymerization can improve the cycle stability by inhibiting the dissolution.

2.3. F5 and its structural analogues

The reaction behavior with charge and discharge was analyzed by ex situ Fourier-transform infrared (FT-IR) spectroscopy (Fig. 4a). As SEI is formed on the surface of the anode active materials, FT-IR is a preferred method to analyze the reaction behavior compared with X-ray photoelectron spectroscopy.^{15,17,25,47,51} The electrode sample was prepared by mixing the active material, conductive carbon, and PTFE binder with the weight ratio of 8 : 1 : 1, and the mixture was pasted on a copper mesh. After the electrochemical reactions, the electrode was measured by attenuated total reflection (ATR) method after withdrawal from the beaker cell, and rinsing with EC and propylene carbonate (PC). The spectroscopic changes were analyzed after the first discharge and subsequent charge cycles at the current density of 1 mA g⁻¹ (Fig. 4a). The spectra of the active material itself, PTFE, and AB without the reactions were measured as the references. In addition, the charge–discharge measurement was carried out for only AB as a reference experiment (Fig. S5 in the ESI†). After the first discharge and subsequent charge cycles, the stretching vibration of C–Cl in F5 was observed at around 790 cm⁻¹ (band A in Fig. 4a). This fact indicates that the low molecular-weight F5 was not dissolved in the electrolyte with the redox reactions. Absorption of the C=C stretching vibration in the benzene ring around 1500 cm⁻¹ decreased with discharging and increased with charging compared with that of the C–Cl vibration (band D in Fig. 4a). The spectroscopic change indicates the lithiation of the benzene ring and its recovery. In addition, the strong absorption bands corresponding to the O–Li, C–O–Li, and C=O peaks were observed at around 1070, 1400, and 1770 cm⁻¹, respectively (bands B, C, and E in Fig. 4a).⁶⁸ The reference AB showed three similar absorption peaks (Fig. S5 in the ESI†). These absorptions originate from SEI containing carbonates and Li⁺ with the decomposition of the electrolytes. The ex situ FT-IR analysis indicates that F5 is redox-active with the uptake of Li⁺ in the benzene ring and formation of SEI. As the capacity of F5 indicates the introduction of 2.9 Li⁺ per molecule, the high specific capacity is achieved by superlithiation in the benzene ring. If the study is carried out without the use of the predictors, F5 as a simple new active material cannot be found based only on research experience and insight.

Fig. 4 Electrochemical properties of F5 and its structural analogues. (a) Ex situ FT-IR spectra of the electrode containing F5 in the discharged and charged states and its references, namely the original F5, PTFE, and AB. (b) Molecular structures of F5-A–F. (c) Charge–discharge curves of F5 and its structural analogues at 100 mA g⁻¹ and of the reference AB at 50 mA g⁻¹ during the second cycle. (d) Relationship between the cycle number and actual (corrected) specific capacity for F5 and F5-A–F at 100 mA g⁻¹.

Additional exploration was carried out for the structural analogues of F5 (Fig. 4b and S6 in the ESI†). Here, the six commercially available structural analogues of F5, F5-A–F5-F were extracted to find new additional compounds with enhanced performances and to study the redox-active moieties (Fig. 4b and c). The ortho- and meta-dichlorobenzenes were not used as active materials because these are in the liquid state at room temperature. Whereas F5 had the specific capacity of 532 mA h g⁻¹, these analogues exhibited the actual specific capacity of 82.3 mA h g⁻¹ for F5-A, 115 mA h g⁻¹ for F5-B, 338 mA h g⁻¹ for F5-C, 97.7 mA h g⁻¹ for F5-D, 113 mA h g⁻¹ for F5-E, and 26.7 mA h g⁻¹ for F5-F (Fig. 4c, Fig. S6 and Table S4 in the ESI,†N = 3). Stable redox reactions were observed for ten cycles, except F5-F (Fig. 4d and Table S4 in the ESI†).

The specific capacity decreased with the extension of the π-conjugated moiety for F5-A and introduction of the benzoquinone moiety for F5-B. Conversely, an enhanced capacity of 338 mA h g⁻¹ was measured for F5-C, which corresponds to the reaction with 4.4 Li⁺ per molecule, with the extension of the dichlorobenzene unit. The differences in the reactivity of these molecules were studied by density functional theory (DFT) calculation (Fig. S7 in the ESI†). The negativity in the benzene ring was increased by the chlorine substituents for F5, F5-A, F5-B, and F5-C. The charge density was delocalized for F5 and F5-C, whereas the charge was localized at the specific sites for F5-A and F5-B (Fig. S7 in the ESI†). This fact indicates that multiple Li⁺ ions can be reacted with the delocalized negative benzene ring. The capacity was not enhanced by substitution of two Cl groups with other halogens (F5-D and F5-E) and increasing Cl groups (F5-F). The bulky groups on the benzene ring decreased the reactivity with Li⁺. Although further analyses are required to understand the redox mechanisms, 1,4-dichlorobenzene as a new type of anode active material has been found in the present work. Such simple redox-active moiety was not reported in the previous works (Table S1 in the ESI†). Moreover, the enhanced specific capacity can be achieved by further molecular design of the F5 analogues.

2.4. F12 and its structural analogues

Ex situ FT-IR analysis of F12 showed spectroscopic changes similar to those of F5 (Fig. 5a). Strong absorptions corresponding to C–H bending vibrations were observed in the range of 700–800 cm⁻¹ after the discharge and charge cycles (band A in Fig. 5a), indicating the stability of F12 during the redox reactions. Absorption of the C=C stretching vibration in the pyridine ring around 1500 cm⁻¹ decreased with discharging and increased with charging, compared with that of C–H (band D in Fig. 5a). The absorption bands B, C, and E indicate the formation of SEI. The ex situ FT-IR analysis of F12 indicates the redox reactions of the pyridine ring and formation of SEI.

Fig. 5 Electrochemical properties of F12 and its structural analogues. (a) Ex situ FT-IR spectra of the electrode containing F12 in the discharged and charged states and its references, namely the original F12, PTFE, and AB. (b) Molecular structures of F12-A–E. (c) Charge–discharge curves of F12 and its structural analogues at 100 mA g⁻¹ and reference AB at 50 mA g⁻¹ during the second cycle. (d) Relationship between the cycle number and actual specific capacity for F12 and F12-A, C–E at 100 mA g⁻¹.

The structural analogues of F12 were explored to find the polymerizable molecules (Fig. 5b). The polymerization has the potential to inhibit dissolution of the low-molecular-weight F12, and improve the conductivity because of the extension of the conjugation length. In our previous work, the specific capacity was improved by the polymerization of the thiophene derivatives.^63,69 Five commercially available analogues of F12 (F12-A–F12-E) were selected based on their structural characteristics, i.e., heteroaromatic ring containing a nitrogen atom and formyl group (Fig. 5b). These compounds exhibited the specific discharge capacities of 171 mA h g⁻¹ for F12-A, 170 mA h g⁻¹ for F12-C, 92.7 mA h g⁻¹ for F12-D, and 181 mA h g⁻¹ for F12-E at the second cycle (Fig. 5c, S8 and Table S5 in the ESI,†N = 3). In contrast, F12-B had low specific capacity. The stable charge–discharge reactions were observed for ten cycles (Fig. 5d and Table S5 in the ESI†). Except for F12-B, the analogues had adequate specific capacities as an anode active material, even though the capacity was not higher than that of the original F12.

2.5. Prediction accuracy of the models G2 and G3

Both models G2 and G3 showed a low prediction accuracy for F5, F12, and their derivatives (Table S6 in the ESI†). The models G2 and G3 were constructed on the training data mainly containing thiophene derivatives and conjugated carbonyls.^51,52 As dichlorobenzene, pyrrole, and pyridine derivatives are not trained, the capacity of these compounds is not accurately predicted. The general applicability of the models G2 and G3 is not sufficient for unknown new compounds. This fact indicates that the significant descriptors are not extracted and used in these models. In our previous works,^51,52,63 the models G1–G3 were improved with additional data. The specific capacity of F5, F12, and their derivatives can be used as additional training data to update the prediction model to improve the accuracy and generalizability.

2.6. Polymerization of F12-D and F12-E

Oxidative polymerization was performed in an aqueous solution containing these F12 analogues, iron chloride (FeCl₃), and ammonium peroxodisulfate (APS). The detailed method is described in the ESI.† The polymerized black precipitates were obtained from the pyrrole derivatives F12-D and F12-E. In contrast, such polymerized precipitates were not collected from the pyridine derivatives F12-A–F12-C after filtration of the reaction solutions. The reactivity of pyridine is lower than that of pyrrole because of the stability of the resonance structure. The polymerization of F12-D and F12-E was briefly analyzed using thermogravimetry (TG) prior to the detailed characterization (Fig. 6a). In the TG curves, the monomeric F12-D and F12-E showed weight losses at around 200 °C with combustion under an air atmosphere. After the polymerization, the weight-loss temperature was shifted to a higher temperature range, comparable to that of a commercial polypyrrole (PPy). The TG analysis indicates the formation of the polymerized F12-D (pF12-D) and F12-E (pF12-E).

Fig. 6 Structural analyses and electrochemical performances of pF12-D and pF12-E. (a) TG curves of F12-D, pF12-D, F12-E, pF12-E, and PPy under an air atmosphere. (b) Charge–discharge curves of pF12-D, pF12-E, and the reference AB at 100 mA g⁻¹ during the second cycle. (c) Relationship between the cycle number and specific capacity for pF12-D and pF12-E at 100 mA g⁻¹. (d) FT-IR spectra of F12-D, pF12-D, and PPy. (e) UV-Vis-NIR spectra of F12-D, pF12-D, and PPy. (f) Estimated composition (upper) and partial structure (lower) of pF12-D.

The specific capacity was 934 mA h g⁻¹ for pF12-D and 285 mA h g⁻¹ for pF12-E (Fig. 6b and S9 and Table S7 in the ESI,†N = 3). The capacity was improved by the polymerization, particularly for pF12-D (Fig. 5c, d, 6b and c). Stable redox reactions were observed for 10 cycles (Fig. 6c and Table S7 in the ESI†). Although the specific capacity slightly decreased with increasing current density from 20 to 5000 mA g⁻¹, the specific capacity did not decrease in the following 100 cycles at 100 mA g⁻¹ (Fig. S9 in the ESI†). The cycle stability of pF12-D was better than that of F12-D. Whereas thiophene derivatives exhibited high performances as organic anodes in previous works (Fig. S1 and Table S1 in the ESI†),^{15,21,28,29,35,63,69} a lower capacity was reported for the pyrrole derivative.⁶⁹ Interestingly, pF12-D, a simple pyrrole derivative, exhibited superior capacity in the present work.

2.7. Amorphous conjugated polymer network of pF12-D

The detailed analysis indicates that pF12-D is not a linear polymer. Instead, pF12-D is an amorphous conjugated polymer network (CPN) (Fig. 6d–f and S10 in the ESI†) with a random networking structure of multiple conjugated monomers, as reported in our recent works.^70,71 The following absorptions were observed in the FT-IR spectra (Fig. 6d).^72,73 The F12-D monomer exhibited absorption peaks corresponding to the C=O stretching vibration of the formyl group in the monomeric state around 1700 cm⁻¹ (band B) and hydrogen-bonded state around 1640 cm⁻¹ (band C). After polymerization, the relative intensity of the band B increased. In addition, a weak absorption corresponding to the C=O stretching vibration of the monomeric carboxy group was observed around 1750 cm⁻¹ (band A). The peaks were broadened for pF12-D compared with those of the F12-D monomer. Whereas the pyrrole ring with the formyl group is arranged in the crystalline state of the F12-D monomer, the arrangement becomes disordered through the formation of the amorphous polymer. A portion of the formyl groups in pF12-D are oxidized into carboxyl groups during oxidative polymerization. The solid-state F12-D monomer showed peaks in the X-ray diffraction (XRD) pattern (Fig. S10 in the ESI†). The sharp peaks disappeared and a broadened halo was observed after polymerization. These TG, FT-IR, and XRD analyses indicate the formation of an amorphous polymer based on F12-D.

In the FT-IR spectra, pF12-D and commercial polypyrrole (PPy) exhibited the following similar absorptions, indicating the formation of the conjugated framework (Fig. 6d):^72,74 stretching vibration of C=C and C–C in the pyrrole ring in the range of 1600–1510 cm⁻¹ (band D), ring breathing vibration of the pyrrole ring containing C=C, C–C and C–N in the ranges of 1510–1380 and 1280–1100 cm⁻¹ (bands E and G), C–H in-plane bending vibration in the ranges of 1380–1280 and 1080–1000 cm⁻¹ (bands F and H), C–H in-plane bending vibration in the range of 930–880 cm⁻¹ (band I), ring bending and deformation vibrations of the pyrrole ring in the range of 870–800 cm⁻¹ (band J). The Raman spectrum of pF12-D showed C–H in-plane bending, C–N stretching, and C=C stretching vibrational features, as observed for that of PPy (Fig. S9 in the ESI†).^72,74 These FT-IR and Raman spectra indicate that pF12-D forms a conjugated structure similar to that of PPy. Although pF12-D shows absorption in the UV-Vis-NIR region (Fig. 6e), the absorption in the NIR region is weak compared with that of PPy. This fact indicates that the effective conjugation length of pF12-D is shorter than that for linear PPy. The CHN elemental analysis of pF12-D was not consistent with the calculated value of the linearly linked F12-D polymer (Table S8 in the ESI†). As reported for the porphyrin synthesis,^75,76 the condensation reaction of the formyl group and pyrrole can provide the specific dimer and branched structures (Fig. 6f). In addition, the formyl group is partially eliminated with decarboxylation under acidic conditions, followed by the oxidation of F12-D into carboxylic acid.⁷⁷ If these reactions proceed with the oxidative polymerization of the pyrrole rings, the three different structural units (namely, monomeric (M), dimerized (D), and trimerized (T)) with the functional group R=H, CHO, or COOH can be randomly linked in the network structure (Fig. 6f). The inclusion of these monomer units in pF12-D is supported by ¹³C solid-state nuclear magnetic resonance (NMR) and X-ray photoelectron spectroscopy (XPS) (Fig. S10 in the ESI†).

The weight ratios of C, H, N, and others (O and Cl) in pF12-D were measured to be 52.48, 3.82, 13.61 and 27.79 using CHN elemental analysis, respectively. The proportion of Cl was measured to be 2.81 wt% using a halogen quantification analysis with combustion. Based on these analyses, the amorphous CPN of pF12-D contained the following units and substituents in mole percentage (Fig. 6f, total 100.1 mol%). The mole percentages of the M, D, and T units were estimated to be 21.4 : 1.14 : 1.01, respectively, with 15.0 H₂O as hydrated water. The mole percentages of the substituent (R =) H, CHO, and COOH were 17.0 : 4.76 : 2.81, respectively. The terminal H of the Py rings was 36.9 mol%. The chloride anion as the dopant was 0.08 mol% in the polymer. The dopant proportion was 7.8 mol% to the pyrrole rings. Based on the estimated structure and composition of pF12-D, the weight ratios of C, H, N, the others (O and Cl), and Cl were calculated to be 53.06, 4.40, 14.18, 28.36, and 2.31, respectively (Table S8 in the ESI†). The calculated values are consistent with the measured ones within 0.6 wt%.

These results indicate that pF12-D forms the amorphous CPN based on the F12-D monomer (Fig. 6f). Our group has proposed the structural advantages of amorphous CPNs in recent years.^67,70,71 The flexible nature originating from the amorphous and random networking structure contributes to enhancement of the electrochemical performances.^67,70,71 In the previous works, amorphous CPNs were synthesized from multiple different monomers. In the present work, a similar structure is derived from a single monomer by the oxidative polymerization with a couple of side reactions. This synthetic approach can be regarded as a new route to obtain amorphous CPNs.

3. Conclusions

New organic anode active materials were found through the exploration of compounds from flower scents using capacity predictors. Prior to the experiments, 62 conjugated molecules were extracted from the original data containing ca. 2000 compounds. Then, eight commercially available candidates were selected for the experiments using the capacity prediction models G2 and G3. The discovered compounds F5 and F12 exhibited specific capacities of 532 and 293 mA h g⁻¹ at 100 mA g⁻¹ with stable redox reactions for ten cycles, respectively. Conjugated carbonyls and thiophene derivatives have been extensively studied in previous works. Conversely, the discovered active materials in the present work, dichlorobenzene and pyrrole derivatives, can be regarded as new types of anode active materials. A polymerizable structural analogue F12-D was found via this additional exploration. The resultant amorphous CPN of pF12-D exhibited a specific capacity of 934 mA h g⁻¹ at 100 mA g⁻¹. These active materials can be used for organic energy storage. The use of prediction models is crucial for facilitating the efficient discovery of such new compounds, as professional research experience alone is insufficient. In this manner, a few new anode active materials could be rapidly and efficiently found with the assistance of performance predictors within a wide search space. This predictor-assisted exploration can be applied to finding new potential compounds for other energy-related applications toward a sustainable society.

Data availability

The data supporting this article have been included as part of the ESI.†

Conflicts of interest

There are no conflicts to declare.

Acknowledgements

The authors thank Dr Yasumasa Kuwahara (Professor Emeritus, Kyoto University and former Professor, Kyoto Gakuen University) for providing the compound list of flower scents in a personal communication. This work was supported by JSPS-KAKENHI (JP22K19071, Y. O.), the New Energy and Industrial Technology Development Organization (NEDO) (JPNP14004), Toray Science Foundation (Y. O.), and Mitsubishi Foundation (Y. O., 202410007).

Notes and references

1. M. Armand and J. M. Tarascon, Nature, 2008, 451, 652.
2. H. Nishide and K. Oyaizu, Science, 2008, 319, 737.
3. Z. Song and H. Zhou, Energy Environ. Sci., 2013, 6, 2280.
4. P. Poizot, J. Gaubicher, S. Renault, L. Dubois, Y. Liang and Y. Yao, Chem. Rev., 2020, 120, 6490.
5. Y. Hong, Z. Ma, K. Li, J. Li, S. Tang, Z. Xu, D. Yu, D. Chen, L. Qin, J. Xie and O. He, J. Mater. Chem. A, 2023, 11, 7898.
6. K. Hatakeyama-Sato and K. Oyaizu, Chem. Rev., 2023, 123, 11336.
7. J. Kim, J. H. Kim and K. Ariga, Joule, 2017, 1, 739.
8. C. Friebe, A. Lex-Balducci and U. S. Schubert, ChemSusChem, 2019, 12, 4093.
9. S. Lee, J. Hong and K. Kang, Adv. Energy Mater., 2020, 10, 2001445.
10. J. J. Shea and C. Luo, ACS Appl. Mater. Interfaces, 2020, 12, 5361.
11. Y. Chen and C. Wang, Acc. Chem. Res., 2020, 53, 2636.
12. H. Yu, J. Xiong, C. Wu, M. Hu, J. Liu and J. Yang, J. Mater. Chem. A, 2024, 12, 8100.
13. S. Gu, J. Chen, I. Hussain, Z. Wang, X. Chen, M. Ahmad, S. P. Feng, Z. Lu and K. Zhang, Adv. Mater., 2024, 36, 2306491.
14. F. Baskoro, S. U. Sharma, A. L. Lubis and H. J. Yen, J. Mater. Chem. A, 2025, 13, 1552.
15. H. H. Lee, Y. Park, K.-H. Shin, K. T. Lee and S. Y. Hong, ACS Appl. Mater. Interfaces, 2014, 6, 19118.
16. W. Walker, S. Grugeon, O. Mentre, S. Laruelle, J.-M. Tarascon and F. Wudl, J. Am. Chem. Soc., 2010, 132, 6517.
17. S. Wang, L. Wang, K. Zhang, Z. Zhu, Z. Tao and J. Chen, Nano Lett., 2013, 13, 4404.
18. V. A. Mihali, S. Renault, L. Nyholm and D. Brandell, RSC Adv., 2014, 4, 38004.
19. L. Fedele, F. Sauvage, J. Bois, J.-M. Tarascon and M. Becuwe, J. Electrochem. Soc., 2014, 161, A46.
20. P. Yang, L. Ma, S. Bi, T. Huang, R. Liu, Y. Su and D. Wu, Chem. Eng. J., 2020, 394, 124924.
21. W. Hu, N. Chen, D. Chen and B. Tong, ChemElectroChem, 2022, 9, e202200026.
22. F.-M. Wang, K. W. Guji, A. Ramar, L. Merinda and W.-C. Chien, ACS Sustainable Chem. Eng., 2021, 9, 12286.
23. Y. Wang, W. Liu, R. Guo, Q. Qu, H. Zheng, J. Zhang and Y. Huang, J. Mater. Chem. A, 2019, 7, 22621.
24. S. Li, J. Lin, Y. Zhang, S. Zhang, T. Jiang, Z. Hu, J. Liu, D.-Y. Wu, L. Zhang and Z. Tian, Adv. Energy Mater., 2022, 12, 2201347.
25. J. Wu, X. Rui, G. Long, W. Chen, Q. Yan and Q. Zhang, Angew. Chem., Int. Ed., 2015, 54, 7354.
26. Y. Wang, Z. Liu, H. Liu, H. Liu, B. Li and S. Guan, Small, 2018, 14, 1704094.
27. C. Zhang, Y. He, P. Mu, X. Wang, Q. He, Y. Chen, J. Zeng, F. Wang, Y. Xu and J.-X. Jiang, Adv. Funct. Mater., 2018, 28, 1705432.
28. S.-B. Ren, W. Ma, C. Zhang, L. Chen, K. Wang, R.-R. Li, M. Shen, D.-M. Han, Y. Chen and J.-X. Jiang, ChemSusChem, 2020, 13, 2295.
29. X. Wang, C. Zhang, Y. Xu, Q. He, P. Mu, Y. Chen, J. Zeng, F. Wang and J.-X. Jiang, Macromol. Chem. Phys., 2018, 219, 1700524.
30. E. R. Wolfson, N. Xiao, L. Schkeryantz, W. K. Haug, Y. Wu and P. L. McGrier, Mol. Syst. Des. Eng., 2020, 5, 97.
31. Y. Tong, Z. Sun, J. Wang, W. Huang and Q. Zhang, SmartMat, 2022, 3, 685.
32. L. Shi, W. Li, Y. Wu, F. Wei, T. Zhang, J. Fu, C. Jing, J. Cheng and S. Liu, Macromol. Rapid Commun., 2022, 43, 2100897.
33. J. Zou, K. Fan, X. Wang, Y. Chen, Y. Cao, H. Dai, C. Zhang, M. Fu, Y. Gao, H. Liu and C. Wang, Chem. Eng. J., 2023, 460, 141703.
34. S. Zhang, K. Zhu, Y. Gao, T. Bao, H. Wu and D. Cao, Chem. - Asian J., 2023, 18, e202300439.
35. S. Hou, X. Zhang, P. Zhou, S. Chen, H. Lin, J. Zhou, S. Zhuo and Y. Liu, RSC Adv., 2021, 11, 35020.
36. M. Wu, Y. Zhao, H. Zhang, J. Zhu, Y. Ma, C. Li, Y. Zhang and Y. Chen, Nano Res., 2022, 11, 9779.
37. H. Liu, Y. Wang, H. Liu, Y. Sun and S. Guan, J. Electroanal. Chem., 2019, 848, 113289.
38. S. B. Alahakoon, C. M. Thompson, G. Occhialini and R. A. Smaldone, ChemSusChem, 2017, 10, 2116.
39. D. Xiong, X. Deng, Z. Cao, S. Ato, Z. Song, X. Xiao, W. Deng, H. Hou, G. Zou and X. Ji, Energy Environ. Sci., 2023, 6, e12521.
40. Z. Wang, J. Hu and Z. Lu, Batteries Supercaps, 2023, 6, e202200545.
41. P. Novák, K. Müller, K. S. V. Santhanam and O. Hass, Chem. Rev., 1997, 97, 207.
42. J. Ĉaja, R. B. Kaner and A. G. MacDiarmid, J. Electrochem. Soc., 1984, 131, 2744.
43. A. Mahammadi, O. Inganäs and I. Lundström, J. Electrochem. Soc., 1986, 133, 947.
44. A. N. Chowdhury, Y. Harima, K. Kunugi and K. Yamashita, Electrochim. Acta, 1996, 41, 1993.
45. S. Yata, K. Sakurai, T. Osaki, Y. Inoue, K. Yamaguchi, K. Tanaka and T. Yamabe, Synth. Met., 1990, 38, 185.
46. M. Armand, S. Grugeon, H. Vezin, S. Laruelle, P. Ribiére, P. Poizot and J. M. Tarascon, Nat. Mater., 2009, 8, 120.
47. X. Han, G. Qing, J. Sun and T. Sun, Angew. Chem., Int. Ed., 2012, 51, 5147.
48. Searched using keyword “organic anode” using Web of Science (February 2025).
49. M. Baloch and J. Labidi, RSC Adv., 2021, 11, 23644.
50. M. Miroshnikov, K. Mahankali, N. K. Thangavel, S. Satapathy, L. M. R. Arava, P. M. Ajayan and G. John, ChemSusChem, 2020, 13, 2186.
51. T. Komura, K. Sakano, Y. Igarashi, H. Numazawa, H. Imai and Y. Oaki, ACS Appl. Energy Mater., 2022, 5, 8990.
52. H. Tobita, Y. Namiuchi, T. Komura, H. Imai, K. Obinata, M. Okada, Y. Igarashi and Y. Oaki, Energy Adv., 2023, 2, 1014.
53. S. Curtarolo, G. L. W. Hart, M. B. Nardelli, N. Mingo, S. Sanvito and O. Levy, Nat. Mater., 2013, 12, 191.
54. B. Sanchez-Lengeling and A. Aspuru-Guzik, Science, 2018, 361, 360.
55. Y. Miyake and A. Saeki, J. Phys. Chem. Lett., 2021, 12, 12391.
56. Y. Oaki and Y. Igarashi, Bull. Chem. Soc. Jpn., 2021, 94, 2410.
57. K. Terayama, M. Sumita, R. Tamura and K. Tsuda, Acc. Chem. Res., 2021, 54, 1334.
58. K. Hatakeyama-Sato, Polym. J., 2023, 55, 117.
59. S. Jha, M. Yen, Y. S. Salinas, E. Palmer, J. Villafurete and H. Liang, J. Mater. Chem. A, 2023, 11, 3904.
60. S. Lo, S. G. Baird, J. Schrier, B. Blaiszik, N. Carson, I. Foster, A. Aguilar-Granda, S. V. Kalinin, B. Maruyama, M. Politi, H. Tran, T. D. Sparks and A. Aspuru-Guzik, Digital Discovery, 2024, 3, 842.
61. J. M. Granda, L. Donina, V. Dragone, D. L. Long and L. Cronin, Nature, 2018, 559, 377.
62. K. Nishio, A. Aiba, K. Takihara, Y. Suzuki, R. Nakayama, S. Kobayashi, A. Abe, H. Baba, S. Katagiri, K. Omoto, K. Ito, R. Shimizu and T. Hitosugi, Digital Discovery, 2025, 4 10.1039/D4DD00326H.
63. H. Numazawa, Y. Igarashi, K. Sato, H. Imai and Y. Oaki, Adv. Theory Simul., 2019, 2, 1900130.
64. K. Noda, Y. Igarashi, H. Imai and Y. Oaki, Adv. Theory Simul., 2020, 3, 2000084.
65. R. Mizuguchi, Y. Igarashi, H. Imai and Y. Oaki, Nanoscale, 2021, 13, 3853.
66. Y. Haraguchi, Y. Igarashi, H. Imai and Y. Oaki, Digital Discovery, 2022, 1, 26.
67. W. Hamada, M. Hishida, R. Sugiura, H. Tobita, H. Imai, Y. Igarashi and Y. Oaki, J. Mater. Chem. A, 2024, 12, 3294.
68. H. Gao, Q. Zhu, A. R. Neale, M. Bahri, X. Wang, H. F. Yang, L. J. Liu, R. Clowes, N. D. Browning, R. S. Sprick, M. A. Little, L. J. Hardwick and A. I. Cooper, Adv. Energy Mater., 2021, 11, 2101880.
69. H. Numazawa, K. Sato, H. Imai and Y. Oaki, NPG Asia Mater., 2018, 10, 397.
70. Y. Oaki and K. Sato, RSC Appl. Polym., 2025, 3, 78.
71. J. Suzuki, A. Ishizone, K. Sato, H. Imai, Y. J. Tseng, C. H. Peng and Y. Oaki, Chem. Sci., 2020, 11, 7003.
72. R. Kostić, D. Rković, S. A. Stepanyan, I. E. Davidova and L. A. Gribov, J. Chem. Phys., 1995, 22, 3104.
73. R. D. Davidson and T. G. Turner, Synth. Met., 1994, 72, 121.
74. T. Hagiwara, M. Hiasaka, K. Sato and M. Yamaura, Synth. Met., 1990, 26, 241.
75. P. Rothemund, J. Am. Chem. Soc., 1936, 58, 625.
76. J. S. Lindsey, H. C. Hsu and I. C. Schreiman, Tetrahedron Lett., 1986, 27, 4969.
77. A. A. Vandersteen, S. O. C. Mundle, G. Lacrampe-Couloume, B. S. Lollar and R. Kluger, J. Org. Chem., 2013, 78, 12176.

---

Footnote

† Electronic supplementary information (ESI) available: Experimental methods, molecular structures, charge–discharge curves, CV curves, FT-IR spectra, structural analyses. See DOI: https://doi.org/10.1039/d5ta03476k

---