De NovoDesign of Molecules with Low Hole Reorganization Energy Based on a Quarter-Million Molecule DFT Screen: Part 2

Abstract

Organic semiconductors have many desirable properties including improved manufacturing and flexible mechanical properties. Due to the vastness of chemical space, it is essential to efficiently explore chemical space when designing new materials, including through the use of generative techniques. New generative machine learning methods for molecular design continue to be published in the literature at a significant rate but successfully adapting methods to new chemistry and problem domains remains difficult. These challenges necessitate continual method evaluation to probe method viability for use in alternative applications not covered in the original works. In continuation of our previous work, we evaluate four additional machine-learning-based de novo methods for generating molecules with high predicted hole mobility for use in semiconductor applications. The four generative methods evaluated here are (1) Molecule Deep Q-Networks (MolDQN), which utilizes Deep-Q learning to directly optimize molecular structure graphs for desired properties instead of generating SMILES, (2) Graph-based Genetic Algorithm (GraphGA), which uses a genetic algorithm for optimization where crossovers and mutations are defined in terms of RDKit’s reaction SMILES, (3) Generative Tensorial Reinforcement Learning (GENTRL), which is a variational autoencoder (VAE) with a learned prior distribution and optimized using reinforcement learning, and (4) Monte Carlo tree search exploration of chemical space in conjunction with a recurrent neural network (RNN) decoder (ChemTS). The generated molecules were evaluated using density functional theory (DFT) and we discovered better performing molecules with the GraphGA method compared to the other approaches.

Augmented Memory: Sample-Efficient Generative Molecular Design with Reinforcement Learning

“…Such workflows have been applied to design optoelectronics 89,90 and semiconducting materials. 91,92 By contrast, Li et al 90 used REINVENT 30,31 to design optoelectronics by explicitly optimizing for xTB proposed Selective Memory Purge heuristic. We benchmark Augmented Memory on the PMO benchmark 29 and achieve a new state-of-the-art performance, outperforming the previous state-of-the-art on 14/23 tasks (statistically significant at the 95% confidence level) and by a total sum of 0.986 AUC Top-10.…”

“…The trade-off to avoiding the costly simulation is that the generative design is constrained to the surrogate model’s domain of applicability, i.e., if the generative model proposes a molecule too dissimilar to the surrogate’s training data, its prediction will more likely be inaccurate. Such workflows have been applied to design optoelectronics , and semiconducting materials. , By contrast, Li et al . used REINVENT , to design optoelectronics by explicitly optimizing for xTB and DFT-computed properties, thus mitigating surrogate out-of-domain concerns.…”

“…Such workflows have been applied to design optoelectronics 89 , 90 and semiconducting materials. 91 , 92 By contrast, Li et al. 90 used REINVENT 30 , 31 to design optoelectronics by explicitly optimizing for xTB 80 and DFT-computed properties, thus mitigating surrogate out-of-domain concerns.…”

Hetero-Polycyclic Aromatic Systems: A Data-Driven Investigation of Structure-Property Relationships

“…Compared to the parent polycyclic aromatic hydrocarbons (PAHs), PASs containing heterocycles offer greater structural diversity as well as a much broader range of optoelectronic properties. Such molecules have been used in diverse settings, functioning as organic field effect transistors, [10][11][12] light-emitting diodes, [13][14][15] organic semiconductors, 16,17 organic photovoltaics, 1,[18][19][20][21][22] photocatalysts, 23 and biological agents for tracking or inhibition, 24,25 and have also been incorporated into larger structures such as nano-hoops, in order to tune and expand their functionality. 26 Herein, we perform an in-depth analysis of the data contained within COMPAS-2, aiming to elucidate the effects of electron count, geometry, atomic composition, and aromatic nature on the molecular properties of PASs.…”