Transfer learning with graph neural networks for optoelectronic properties of conjugated oligomers

Abstract

Despite the remarkable progress of machine learning (ML) techniques in chemistry, modeling the optoelectronic properties of long conjugated oligomers and polymers with ML remains challenging due to the difficulty in obtaining sufficient training data. Here we use transfer learning to address the data scarcity issue by pretraining graph neural networks using data from short oligomers. With only a few hundred training data, we are able to achieve an average error of about 0.1 eV for excited state energy of oligothiophenes against TDDFT calculations. We show that the success of our transfer learning approach relies on the relative locality of low-lying electronic excitations in long conjugated oligomers. Finally, we demonstrate the transferability of our approach by modeling the lowest-lying excited-state energies of poly(3-hexylthiopnene) (P3HT) in its single-crystal and solution phases using the transfer learning models trained with data of gas-phase oligothiophenes. The transfer learning predicted excited-state energy distributions agree quantitatively with TDDFT calculations and capture some important qualitative features observed in experimental absorption spectra.

Quantum Deep Descriptor: Physically Informed Transfer Learning from Small Molecules to Polymers

“…The principal future goal should not be the achievement of SoTA performance on a single benchmark database. It is more important (and will yield more practical applications) to consider other materials [including additional polymers, biologically active materials (e.g., drugs), and catalysts] . ,, The properties of these materials should be predicted by transferring the physically informed ML models, such as our QDF, electron density–based ML, , molecular orbital–based ML (MOB-ML), − and differential Kohn–Sham ML models and , training them on various quantum-chemical property databases, such as PubChemQC, tmQM, Alchemy, and others . ,, The key questions are as follows: (1) On what data (e.g., excited states) do we train a physically informed ML model? (2) What data (e.g., catalysis) do we transfer and predict with the trained ML model?…”

“…The final layer, denoted as in Figure a, provides a fixed N -dimensional vector for any atomistic system, regardless of the size and structure of the molecule, and this vector can be reused as an input molecular descriptor for other tasks in materials informatics (MI). We refer to as the QDD and reuse it for transfer learning − of much larger molecules, such as the polymer in Figure b. We show that the QDD, which was pre-trained using the atomization energy, highest occupied molecular orbital (HOMO), and lowest unoccupied molecular orbital (LUMO) of small molecules simultaneously, can predict not only the atomization energy and band gap but also different properties (e.g., dielectric constants) of polymers.…”

New Opportunity: Machine Learning for Polymer Materials Design and Discovery

“…Under the circumstance of dealing with the limited training set, transfer learning is required to obtain the pretrained model of the features and relative properties, which could produce a more outstanding performance in the prediction accuracy with the comparison of the direct ML model with the small data set. [63][64][65] Lee et al [66] employed the GNNs with transfer learning to predict the lowest excited state energy of poly(3hexylthiophene) in single crystal and solution phases. DFT calcu-lation data from short oligomers of different lengths were used to pre-train a GNNs model data from short oligomers of various lengths whose repeating unit is the same as that of the target long oligomers.…”

Orbital Mixer: Using Atomic Orbital Features for Basis-Dependent Prediction of Molecular Wavefunctions

“…An explosion of interest has surrounded applying machine learning (ML) methods to quantum chemistry with a plethora of interesting application areas such as learning interatomic potentials, − constructing density functionals, − and predicting spectroscopic properties, , optoelectronic properties, − activation energies, , and a variety of physical properties throughout chemical compound space. − Quantum chemistry workflows can obtain such chemical and physical information by modeling the electronic Schrödinger equation in a chosen basis set of localized atomic orbitals that is then used to derive the ground-state molecular wavefunction. Using ML, we can rather directly predict the molecular electronic structure, which then provides access to a plethora of these derived properties without needing to train specialized models for properties of interest.…”