Transfer learning across different chemical domains: virtual screening of organic materials with deep learning models pretrained on small molecule and chemical reaction data

Machine learning is becoming a preferred method for the virtual screening of organic materials due to its cost-effectiveness over traditional computationally demanding techniques. However, the scarcity of labeled data for organic materials poses a significant challenge for training advanced machine learning models. This study showcases the potential of utilizing databases of drug-like small molecules and chemical reactions to pretrain the BERT model, enhancing its performance in the virtual screening of organic materials. By fine-tuning the BERT models with data from five virtual screening tasks, the version pretrained with the USPTO-SMILES dataset achieved R2 scores exceeding 0.94 for three tasks and over 0.81 for two others. This performance surpasses that of models pretrained on the small molecule or organic materials databases and outperforms three traditional machine learning models trained directly on virtual screening data. The success of the USPTO-SMILES pretrained BERT model can be attributed to the diverse array of organic building blocks in the USPTO database, offering a broader exploration of the chemical space. The study further suggests that accessing a reaction database with a wider range of reactions than the USPTO could further enhance model performance. Overall, this research validates the feasibility of applying transfer learning across different chemical domains for the efficient virtual screening of organic materials.Scientific contributionThis study verifies the feasibility of applying transfer learning to large language models in different chemical fields to help organic materials perform virtual screening. Through the comparison of transfer learning from different chemical fields to a variety of organic material molecules, the high precision virtual screening of organic materials is realized.

Deep transfer learning for predicting frontier orbital energies of organic materials using small data and its application to porphyrin photocatalysts

A deep transfer learning approach is used to predict HOMO/LUMO energies of organic materials with a small amount of training data.

Exploring Deep Learning for Metalloporphyrins: Databases, Molecular Representations, and Model Architectures

Metalloporphyrins have been studied as biomimetic catalysts for more than 120 years and have accumulated a large amount of data, which provides a solid foundation for deep learning to discover chemical trends and structure–function relationships. In this study, key components of deep learning of metalloporphyrins, including databases, molecular representations, and model architectures, were systematically investigated. A protocol to construct canonical SMILES for metalloporphyrins was proposed, which was then used to represent the two-dimensional structures of over 10,000 metalloporphyrins in an existing computational database. Subsequently, several state-of-the-art chemical deep learning models, including graph neural network-based models and natural language processing-based models, were employed to predict the energy gaps of metalloporphyrins. Two models showed satisfactory predictive performance (R2 0.94) with canonical SMILES as the only source of structural information. In addition, an unsupervised visualization algorithm was used to interpret the molecular features learned by the deep learning models.

High-throughput virtual screening for organic electronics: a comparative study of alternative strategies

We present a review of the field of high-throughput virtual screening for organic electronics materials focusing on the sequence of methodological choices that determine each virtual screening protocol. These choices are present in all high-throughput virtual screenings and addressing them systematically will lead to optimised workflows and improve their applicability. We consider the range of properties that can be computed and illustrate how their accuracy can be determined depending on the quality and size of the experimental datasets. The approaches to generate candidates for virtual screening are also extremely varied and their relative strengths and weaknesses are discussed. The analysis of high-throughput virtual screening is almost never limited to the identification of top candidates and often new patterns and structure-property relations are the most interesting findings of such searches. The review reveals a very dynamic field constantly adapting to match an evolving landscape of applications, methodologies and datasets.

Do HOMO-LUMO Energy Levels and Band Gaps Provide Sufficient Understanding of Dye-Sensitizer Activity Trends for Water Purification?

A dye-sensitized solar cell assembly can be used to harvest solar energy, while suitable dye sensitizers can be used to purify water. Here, we characterized the activity trends of four dye sensitizers, namely, PORPC-1, PORPC-2, PORPC-3, and PORPC-4, for water purification applications using density functional theory (DFT) with the Perdew-Burke-Ernzerhof (PBE), B3LYP, and PBE0 functionals, ΔSCF, time-dependent DFT (TD-DFT), and quasiparticle Green's function (GW) methods. The energy levels of the highest occupied molecular orbitals (HOMOs) and lowest unoccupied molecular orbitals (LUMOs) were calculated using gas-phase and aqueous-phase methods in order to understand charge-injection abilities and the dye regeneration processes. PBE, B3LYP, PBE0, and TD-DFT methods failed to predict PORPC-4 to be the best sensitizer, while PORPC-2 and PORPC-4 were predicted to be the best sensitizers using ΔSCF coupled with the implicit solvation method, and HOMO-LUMO energies were corrected for the aqueous environment in the GW calculations. However, none of these methods accurately predicted the performance trend of all four dye sensitizers. Consequently, we used the aggregation assembly patterns of the dye molecules in an aqueous environment to further probe the activity trends and found that PORPC-3 and PORPC-4 prefer J-aggregated assembly patterns, whereas PROPC-1 and PORPC-2 prefer to be H-aggregated. Therefore, the performance of these dye molecules can be determined by combining HOMO-LUMO energy levels with aggregate-assembly patterns, with the activity trend predicted to be PORPC-4 > PORPC-2 > PORPC-3 > PORPC-1, which is in good agreement with experimental findings.

A Protocol for Fast Prediction of Electronic and Optical Properties of Donor-Acceptor Polymers Using Density Functional Theory and the Tight-Binding Method

The ability of donor-acceptor (D-A) type polymers to control the positions of the highest occupied (HOMO) and lowest unoccupied (LUMO) molecular orbitals makes them a popular choice for organic solar cell applications. The alternating D-A pattern in a monomer leads to a weak electronic coupling between the constituent monomers within the polymer chain. Exploiting the weak electronic coupling characteristics, we developed a method to efficiently calculate (1) the electronic properties and (2) the optical gap of such polymer chains. The electronic properties (HOMO and LUMO energies, ionization potential, electron affinity, and quasiparticle gap of an oligomer of any length up to an infinitely long polymer) of the D-A polymers are predicted by combining density functional theory calculation results and a tight-binding model. The weak electronic coupling implies that the optical gap of the polymer is size-independent, and thus, it can be calculated using a monomer. We validated the methods using a set of 104 polymers by checking the consistency where the electronic gap of a polymer is larger than the optical gap. Furthermore, we establish relationships between the results obtained from more accurate, yet slower methods (i.e., B3LYP functional, singlet-ΔSCF) with those obtained from the faster counterparts (i.e., BLYP functional, triplet-ΔSCF). Leveraging the found relationships, we propose a way in which the electronic and optical properties of the polymers can be calculated efficiently while retaining high accuracy. The use of the tight-binding model combined with the approach to estimate more accurate results based on less expensive simulations is crucial in the applications where a large volume of computations needs to be carried out efficiently with sufficiently high accuracy, such as high-throughput computational screening or training a machine-learning model.

Theoretical investigation of the structural and spectroscopic properties of expanded metalloporphyrin complexes

The frontier molecular orbitals, UV-Vis absorption spectra, charge transfer (CT) and triplet excited states of 12 expanded D-A porphyrin/benzoporphyrin complexes were investigated using the density functional theory (DFT) method and time-dependent DFT in this work. The results showed that thiophene was an effective fragment for absorption of 'long wavelength', while the benzoporphyrin worked on the 'short wavelength', which was derived from its saddle-shaped structure; this expanded D-A conjugated system had a mild CT process with anthraquinone/isoindigo as acceptors and a strong CT process with naphtoquinone as acceptor. In addition, based on the simulation of the triplet state, the theoretical phosphorescence wavelength range of this series of derivatives was between 1000 and 1200 nm. This study is expected to assist the design of conjugated porphyrin for the field of porphyrin chemistry.

Theoretical design of porphyrin sensitizers with different acceptors for application in dye-sensitized solar cells

Using density functional theory (DFT) and time-dependent density functional theory (TDDFT) methods, three porphyrin dyes with different acceptors, such as carboxylic acid, cyanoacrylic acid, and 2-cyano-N-hydroxyacrylamide, have been designed. Compared to the best sensitizer (YD2-o-C8) so far, these designed dyes have small highest occupied orbital to lowest unoccupied orbital (HOMO–LUMO) band gaps, and wide absorptions with large oscillator strength at porphyrin Q bands. And the designed Dye1 is similar to YD2-o-C8 in electronic coupling with TiO₂, while improved Dye2 and Dye3 are better than YD2-o-C8, thus, Dye2 and Dye3 will be much faster for electron injection in dye-sensitized solar cell systems based on their long-term stable and efficient anchor groups. All these features show that our designed dyes, especially Dye2 and Dye3, have better absorption performance and faster electron injection. In addition, our results point out that 2-cyano-N-hydroxyacrylamide is a new promising acceptor. This study is expected to assist the molecular design of new efficient dyes for the advancement of dye-sensitized solar cells.

Using density functional theory (DFT) and time-dependent density functional theory (TDDFT) methods, three porphyrin dyes with different acceptors, such as carboxylic acid, cyanoacrylic acid, and 2-cyano-N-hydroxyacrylamide, have been designed.

Accurate description of the electronic structure of organic semiconductors by GW methods

Electronic properties associated with charged excitations, such as the ionization potential (IP), the electron affinity (EA), and the energy level alignment at interfaces, are critical parameters for the performance of organic electronic devices. To computationally design organic semiconductors and functional interfaces with tailored properties for target applications it is necessary to accurately predict these properties from first principles. Many-body perturbation theory is often used for this purpose within the GW approximation, where G is the one particle Green's function and W is the dynamically screened Coulomb interaction. Here, the formalism of GW methods at different levels of self-consistency is briefly introduced and some recent applications to organic semiconductors and interfaces are reviewed.

Theoretical design of triphenylamine-based derivatives with asymmetric D-D-π-A configuration for dye-sensitized solar cells

The use of theoretical techniques in the structural development of dye-sensitized solar cells helps in the efficient screening of the dyes. To properly rationalize the dye's design process, benchmark calculations were conducted using long-range corrected exchange-correlation (xc) functionals with varying separation parameters to be able to predict the excited-state energies of triphenylamine-based dyes, namely: PPS, PSP, and PSS, wherein they differ at the π-conjugated bridge using thiophene and/or phenyl moieties. The results show that LC-ωPBE xc functional with an optimized parameter provided better correlation with the experimental results compared to the other functionals. The relative shifts of the absorption spectra, light harvesting efficiency, normal dipole moments, as well as the ionization potentials and electron affinities of the dyes were well-correlated with the experimental data. A new set of dyes was designed in an effort to increase its solar cell efficiency that was patterned after PSS with an additional donor moiety such as fluorene, cyclopentaindole, and pyrene attached asymmetrically at the triphenylamine ring. Among the newly designed dyes, analogs that contain 4-phenyl-1,2,3,4-tetrahydrocyclopenta[b]indole (I) and pyrido[2,3,4-5-imn]phenanthridine-5,10(4H,9H)-dione (P2) as the additional donor moiety produced the best photophysical properties and charge-transfer characteristics for a promising dye for solar cell applications.

Design of two-photon molecular tandem architectures for solar cells by ab initio theory

We present new two-photon molecular architectures for photovoltaics where atomic precision can be obtained by synthetic chemistry.

An extensive database of spectroscopic properties of molecules from ab initio calculations is used to design molecular complexes for use in tandem solar cells that convert two photons into a single electron–hole pair, thereby increasing the output voltage while covering a wider spectral range. Three different architectures are considered: the first two involve a complex consisting of two dye molecules with appropriately matched frontier orbitals, connected by a molecular diode. Optimized combinations of dye molecules are determined by taking advantage of our computational database of the structural and energetic properties of several thousand porphyrin dyes. The third design is a molecular analogy of the intermediate band solar cell, and involves a single dye molecule with strong intersystem crossing to ensure a long lifetime of the intermediate state. Based on the calculated energy levels and molecular orbitals, energy diagrams are presented for the individual steps in the operation of such tandem solar cells. We find that theoretical open circuit voltages of up to 1.8 V can be achieved using these tandem designs. Questions about the practical implementation of prototypical devices, such as the synthesis of the tandem molecules and potential loss mechanisms, are addressed.

Can elongation of the π-system in triarylamine derived sensitizers with either benzothiadiazole and/or ortho-fluorophenyl moieties enrich their light harvesting efficiency? – a theoretical study

This study established that elongation of the π-system in a series of hypothetical triphenylamine dyes by the judicious placement either a fluorophenyl or benzothiadiazole group, or a combination of both groups, results in improved light harvesting efficiency in DSSCs.

Theoretical study on the light harvesting efficiency of zinc porphyrin sensitizers for DSSCs

DFT and TDDFT calculations have been carried out to investigate the effect of donor and acceptor groups on the electronic properties of zinc-porphyrin sensitizers. The calculated results show that increasing the electron releasing strength of a meso-donor group opposite to a meso substituted acceptor group increases the light harvesting efficiency and short circuit current density.

Optimizing porphyrins for dye sensitized solar cells using large-scale ab initio calculations

We present a systematic study of the level alignment of 5145 porphyrin based dyes for dye sensitized solar cells.