Evan R. Antoniuk

Although generative models hold promise for discovering molecules with optimized desired properties, they often fail to suggest synthesizable molecules that improve upon the properties of the structures represented in the training distribution. We find that this limitation arises not only from the molecule generation process itself, but also from the poor generalization capabilities of molecular property predictors. We address this challenge by creating a closed-loop molecule generation pipeline with iterative retraining on new quantum chemical simulation data. Compared against static, single-pass generative modeling approaches, only our closed-loop iterative workflow generates molecules with properties extending beyond the training distribution (up to 0.44 standard deviations beyond the original range) and achieves a 79% improvement in out-of-distribution molecule classification accuracy. Furthermore, by conditioning molecular generation on thermodynamic stability data obtained during the iterative loop, the proportion of stable and hence potentially synthesizable molecules generated is 3.5x higher than the next-best model.

Although generative models hold promise for discovering molecules with optimized desired properties, they often fail to suggest synthesizable molecules that improve upon the known molecules seen in training. We find that a key limitation is not in the molecule generation process itself, but in the poor generalization capabilities of molecular property predictors. We tackle this challenge by creating an active-learning, closed-loop molecule generation pipeline, whereby molecular generative models are iteratively refined on feedback from quantum chemical simulations to improve generalization to new chemical space. Compared against other generative model approaches, only our active learning approach generates molecules with properties that extrapolate beyond the training data (reaching up to 0.44 standard deviations beyond the training data range) and out-of-distribution molecule classification accuracy is improved by 79%. By conditioning molecular generation on thermodynamic stability data from the active-learning loop, the proportion of stable molecules generated is 3.5x higher than the next-best model.

Polymer processing, purification, and self-assembly have significant roles in the design of polymeric materials. Understanding how polymers behave in solution (e.g., their solubility, chemical properties, etc.) can improve our control over material properties via their processing-structure–property relationships. For many decades the polymer science community has relied on thermodynamic and physics-based models to aid in this endeavor, but all rely on disparate data sets and use-case scenarios. Hence, there are still significant challenges to predict a priori the solubility of a polymer, whether it is for selecting sustainable solvents, obtaining thermodynamic parameters for phase separation, or navigating the coexistence curve. This perspective aims to discuss the different approaches of applying computational tools to predict polymer solubility, with a significant focus on machine learning techniques to capture the rapid progress in that space. We examine challenges and opportunities that remain for creating a comprehensive solubility toolset that can accelerate the design of a broad range of applications including films, membranes, and pharmaceuticals.

The work function is the key surface property that determines the energy required to extract an electron from the surface of a material. This property is crucial for thermionic energy conversion, band alignment in heterostructures, and electron emission devices. This work presents a high‐throughput workflow using density functional theory (DFT) to calculate the work function and cleavage energy of 33,631 slabs (58,332 work functions) that are created from 3,716 bulk materials. The number of calculated surface properties surpasses the previously largest database by a factor of ≈27. Several surfaces with an ultra‐low (7 eV) work function are identified. Specifically, the (100)‐Ba‐O surface of BaMoO₃ and the (001)‐F surface of Ag₂F have record‐low (1.25 eV) and record‐high (9.06 eV) steady‐state work functions. Based on this database a physics‐based approach to featurize surfaces is utilized to predict the work function. The random forest model achieves a test mean absolute error (MAE) of 0.09 eV, comparable to the accuracy of DFT. This surrogate model enables rapid predictions of the work function (≈ 10⁵ faster than DFT) across a vast chemical space and facilitates the discovery of material surfaces with extreme work functions for energy conversion and electronic device applications.

Reliably identifying synthesizable inorganic crystalline materials is an unsolved challenge required for realizing autonomous materials discovery. In this work, we develop a deep learning synthesizability model (SynthNN) that leverages the entire space of synthesized inorganic chemical compositions. By reformulating material discovery as a synthesizability classification task, SynthNN identifies synthesizable materials with 7× higher precision than with DFT-calculated formation energies. In a head-to-head material discovery comparison against 20 expert material scientists, SynthNN outperforms all experts, achieves 1.5× higher precision and completes the task five orders of magnitude faster than the best human expert. Remarkably, without any prior chemical knowledge, our experiments indicate that SynthNN learns the chemical principles of charge-balancing, chemical family relationships and ionicity, and utilizes these principles to generate synthesizability predictions. The development of SynthNN will allow for synthesizability constraints to be seamlessly integrated into computational material screening workflows to increase their reliability for identifying synthetically accessible materials.

Reliably identifying synthesizable inorganic crystalline materials is an unsolved challenge required for realizing autonomous materials discovery. In this work, we develop a deep learning synthesizability model (SynthNN) that leverages the entire corpus of synthesized inorganic chemical compositions. By reformulating material discovery as a synthesizability classification task, SynthNN identifies synthesizable materials with 7x higher precision than with DFT-calculated formation energies. In a head-to-head material discovery comparison against 20 expert material scientists, SynthNN outperforms all experts, achieves 3.6x higher precision and completes the task five orders of magnitude faster than the average human expert. Remarkably, without any prior chemical knowledge, our experiments indicate that SynthNN learns the chemical principles of charge-balancing, chemical family relationships and ionicity, and utilizes these principles to generate synthesizability predictions. The development of SynthNN will allow for synthesizability constraints to be seamlessly integrated into computational material screening workflows to increase their reliability for identifying synthetically accessible materials.

Accurately predicting new polymers' properties with machine learning models apriori to synthesis has potential to significantly accelerate new polymers' discovery and development. However, accurately and efficiently capturing polymers' complex, periodic structures in machine learning models remains a grand challenge for the polymer cheminformatics community. Specifically, there has yet to be an ideal solution for the problems of how to capture the periodicity of polymers, as well as how to optimally develop polymer descriptors without requiring human-based feature design. In this work, we tackle these problems by utilizing a periodic polymer graph representation that accounts for polymers' periodicity and coupling it with a message-passing neural network that leverages the power of graph deep learning to automatically learn chemically relevant polymer descriptors. Remarkably, this approach achieves state-of-the-art performance on 8 out of 10 distinct polymer property prediction tasks. These results highlight the advancement in predictive capability that is possible through learning descriptors that are specifically optimized for capturing the unique chemical structure of polymers.

The high brightness, low emittance electron beams achieved in modern X‐ray free‐electron lasers (XFELs) have enabled powerful X‐ray imaging tools, allowing molecular systems to be imaged at picosecond time scales and sub‐nanometer length scales. One of the most promising directions for increasing the brightness of XFELs is through the development of novel photocathode materials. Whereas past efforts aimed at discovering photocathode materials have typically employed trial‐and‐error‐based iterative approaches, this work represents the first data‐driven screening for high brightness photocathode materials. Through screening over 74 000 semiconducting materials, a vast photocathode dataset is generated, resulting in statistically meaningful insights into the nature of high brightness photocathode materials. This screening results in a diverse list of photocathode materials that exhibit intrinsic emittances that are up to 4x lower than currently used photocathodes. In a second effort, multiobjective screening is employed to identify the family of M₂O (M = Na, K, Rb) that exhibits photoemission properties that are comparable to the current state‐of‐the‐art photocathode materials, but with superior air stability. This family represents perhaps the first intrinsically bright, visible light photocathode materials that are resistant to reactions with oxygen, allowing for their transport and storage in dry air environments.