Dataset Information

Limitations of machine learning models when predicting compounds with completely new chemistries: possible improvements applied to the discovery of new non-fullerene acceptors.

ABSTRACT: We try to determine if machine learning (ML) methods, applied to the discovery of new materials on the basis of existing data sets, have the power to predict completely new classes of compounds (extrapolating) or perform well only when interpolating between known materials. We introduce the leave-one-group-out cross-validation, in which the ML model is trained to explicitly perform extrapolations of unseen chemical families. This approach can be used across materials science and chemistry problems to improve the added value of ML predictions, instead of using extrapolative ML models that were trained with a regular cross-validation. We consider as a case study the problem of the discovery of non-fullerene acceptors because novel classes of acceptors are naturally classified into distinct chemical families. We show that conventional ML methods are not useful in practice when attempting to predict the efficiency of a completely novel class of materials. The approach proposed in this work increases the accuracy of the predictions to enable at least the categorization of materials with a performance above and below the median value.

SUBMITTER: Zhao ZW

PROVIDER: S-EPMC9189862 | biostudies-literature | 2022 Jun

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Limitations of machine learning models when predicting compounds with completely new chemistries: possible improvements applied to the discovery of new non-fullerene acceptors.

Zhao Zhi-Wen ZW Del Cueto Marcos M Troisi Alessandro A

Digital discovery 20220325 3

We try to determine if machine learning (ML) methods, applied to the discovery of new materials on the basis of existing data sets, have the power to predict completely new classes of compounds (extrapolating) or perform well only when interpolating between known materials. We introduce the leave-one-group-out cross-validation, in which the ML model is trained to explicitly perform extrapolations of unseen chemical families. This approach can be used across materials science and chemistry proble ...[more]

PMID: 35769202

Similar Datasets

Project description:The active layer in a solution processed organic photovoltaic device comprises a light absorbing electron donor semiconductor, typically a polymer, and an electron accepting fullerene acceptor. Although there has been huge effort targeted to optimize the absorbing, energetic, and transport properties of the donor material, fullerenes remain as the exclusive electron acceptor in all high performance devices. Very recently, some new non-fullerene acceptors have been demonstrated to outperform fullerenes in comparative devices. This Account describes this progress, discussing molecular design considerations and the structure-property relationships that are emerging. The motivation to replace fullerene acceptors stems from their synthetic inflexibility, leading to constraints in manipulating frontier energy levels, as well as poor absorption in the solar spectrum range, and an inherent tendency to undergo postfabrication crystallization, resulting in device instability. New acceptors have to address these limitations, providing tunable absorption with high extinction coefficients, thus contributing to device photocurrent. The ability to vary and optimize the lowest unoccupied molecular orbital (LUMO) energy level for a specific donor polymer is also an important requirement, ensuring minimal energy loss on electron transfer and as high an internal voltage as possible. Initially perylene diimide acceptors were evaluated as promising acceptor materials. These electron deficient aromatic molecules can exhibit good electron transport, facilitated by close packed herringbone crystal motifs, and their energy levels can be synthetically tuned. The principal drawback of this class of materials, their tendency to crystallize on too large a length scale for an optimal heterojunction nanostructure, has been shown to be overcome through introduction of conformation twisting through steric effects. This has been primarily achieved by coupling two units together, forming dimers with a large intramolecular twist, which suppresses both nucleation and crystal growth. The generic design concept of rotationally symmetrical aromatic small molecules with extended π orbital delocalization, including polyaromatic hydrocarbons, phthalocyanines, etc., has also provided some excellent small molecule acceptors. In most cases, additional electron withdrawing functionality, such as imide or ester groups, can be incorporated to stabilize the LUMO and improve properties. New calamitic acceptors have been developed, where molecular orbital hybridization of electron rich and poor segments can be judiciously employed to precisely control energy levels. Conformation and intermolecular associations can be controlled by peripheral functionalization leading to optimization of crystallization length scales. In particular, the use of rhodanine end groups, coupled electronically through short bridged aromatic chains, has been a successful strategy, with promising device efficiencies attributed to high lying LUMO energy levels and subsequently large open circuit voltages.

Project description:This work was inspired by a previous report [Janjua, M. R. S. A. Inorg. Chem. 2012, 51, 11306-11314] in which the optoelectronic properties were improved with an acceptor bearing heteroaromatic rings. Herein, we have designed four novel Y-series non-fullerene acceptors (NFAs) by end-capped acceptor modifications of a recently synthesized 15% efficient Y21 molecule for better optoelectronic properties and their potential use in solar cell applications. Density functional theory (DFT) along with time-dependent density functional theory (TDDFT) at the B3LYP/6-31G(d,p) level of theory is used to calculate the band gap, exciton binding energy along with transition density matrix (TDM) analysis, reorganizational energy of electrons and holes, and absorption maxima and open-circuit voltage of investigated molecules. In addition, the PM6:YA1 complex is also studied to understand the charge shifting from the donor polymer PM6 to the NFA blend. Results of all parameters suggest that the DA'D electron-deficient core and effective end-capped acceptors in YA1-YA4 molecules form a perfect combination for effective tuning of optoelectronic properties by lowering frontier molecular orbital (FMO) energy levels, reorganization energy, and binding energy and increasing the absorption maximum and open-circuit voltage values in selected molecules (YA1-YA4). The combination of extended conjugation and excellent electron-withdrawing capability of the end-capped acceptor moiety in YA1 makes YA1 an excellent organic solar cell (OSC) candidate owing to promising photovoltaic properties including the lowest energy gap (1.924 eV), smallest electron mobility (λe = 0.0073 eV) and hole mobility (λh = 0.0083 eV), highest λmax values (783.36 nm (in gas) and 715.20 nm (in chloroform) with lowest transition energy values (E x) of 1.58 and 1.73 eV, respectively), and fine open-circuit voltage (V oc = 1.17 V) with respect to HOMOPM6-LUMOacceptor. Moreover, selected molecules are observed to have better photovoltaic properties than Y21, thus paving the way for experimentalists to look for future developments of Y-series-based highly efficient solar cells.

Dataset Information

Limitations of machine learning models when predicting compounds with completely new chemistries: possible improvements applied to the discovery of new non-fullerene acceptors.

Publications

Limitations of machine learning models when predicting compounds with completely new chemistries: possible improvements applied to the discovery of new non-fullerene acceptors.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets