Efficient 3D Molecular Design with an E(3) Invariant Transformer VAE.

This work introduces a three-dimensional (3D) invariant graph-to-string transformer variational autoencoders (VAE) (Vagrant) for generating molecules with accurate density functional theory (DFT)-level properties. Vagrant learns to model the joint probability distribution of a 3D molecular structure and its properties by encoding molecular structures into a 3D-aware latent space. Directed navigation through this latent space implicitly optimizes the 3D structure of a molecule, and the latent embedding can be used to condition a generative transformer to predict the candidate structure as a one-dimensional (1D) sequence. Additionally, we introduce two novel sampling methods that exploit the latent characteristics of a VAE to improve performance. We show that our method outperforms comparable 3D autoregressive and diffusion methods for predicting quantum chemical property values of novel molecules in terms of both sample quality and computational efficiency.

Boron removal from synthetic brines and oilfield produced waters using aluminum electrocoagulation.

High boron (B) levels in oil and gas produced waters prevent its beneficial reuse as irrigation water without proper treatment. Aluminum (Al) electrocoagulation (EC) is a promising technology for B removal, but further research and development is needed to optimize EC for use in removing B from produced waters. To this end, B removal by adsorption onto insoluble aluminum hydroxide solids, generated by EC in simulated brines (up to 50,000 mg/L NaCl) and real oilfield produced waters, was studied. B removal during EC was greater than when aluminum hydroxide solids formed by EC were subsequently exposed to B containing solutions. Working parameters affecting B removal during the EC process, including current, total dissolved solid (TDS), temperature, pH, scale-forming cations and organic matter, were investigated to explore ways to achieve higher B removal. Boron removal increased with increased current loading and time, and with the concomitant increased Al solids concentration. However, too high a current loading limited B removal because of a change in the structure of the aluminum hydroxide solids. Higher TDS decreased B removal slightly, but lower TDS concentrations limited the use of higher current loadings. Temperature increased during EC treatment, particularly at higher current loadings, and this inhibited B removal due to an accelerated aggregation of amorphous Al solids into larger, denser, and presumably more crystalline particles. The best B removal occurred at pH 8, corresponding to a slightly positive zeta potential for aluminum hydroxide and a small but significant fraction of negatively charged B species. Scale-forming cations such as Ba2+ and Sr2+ had no obvious effect on the EC process. The presence of high concentrations of Mg2+ and Ca2+ resulted in low bulk pH values during the EC process and greater formation of solid products, but B removal did not decrease during a pH-controlled (pH = 8) EC process with these divalent cations present. Two produced water samples collected from oilfields in Kansas, US were treated using EC for 1 h, resulting in up to ~70 % B removal from solution with a current loading of 6.67 A/L, and up to 78 % with 13.33 A/L.

Predictive Theoretical Framework for Dynamic Control of Bioinspired Hybrid Nanoparticle Self-Assembly.

At-will tailoring of the formation and reconfiguration of hierarchical structures is a key goal of modern nanomaterial design. Bioinspired systems comprising biomacromolecules and inorganic nanoparticles have potential for new functional material structures. Yet, consequential challenges remain because we lack a detailed understanding of the temporal and spatial interplay between participants when it is mediated by fundamental physicochemical interactions over a wide range of scales. Motivated by a system in which silica nanoparticles are reversibly and repeatedly assembled using a homobifunctional solid-binding protein and single-unit pH changes under near-neutral solution conditions, we develop a theoretical framework where interactions at the molecular and macroscopic scales are rigorously coupled based on colloidal theory and atomistic molecular dynamics simulations. We integrate these interactions into a predictive coarse-grained model that captures the pH-dependent reversibility and accurately matches small-angle X-ray scattering experiments at collective scales. The framework lays a foundation to connect microscopic details with the macroscopic behavior of complex bioinspired material systems and to control their behavior through an understanding of both equilibrium and nonequilibrium characteristics.

Attention-based generative models for de novo molecular design.

Attention mechanisms have led to many breakthroughs in sequential data modeling but have yet to be incorporated into any generative algorithms for molecular design. Here we explore the impact of adding self-attention layers to generative ß-VAE models and show that those with attention are able to learn a complex "molecular grammar" while improving performance on downstream tasks such as accurately sampling from the latent space ("model memory") or exploring novel chemistries not present in the training data. There is a notable relationship between a model's architecture, the structure of its latent memory and its performance during inference. We demonstrate that there is an unavoidable tradeoff between model exploration and validity that is a function of the complexity of the latent memory. However, novel sampling schemes may be used that optimize this tradeoff. We anticipate that attention will play an important role in future molecular design algorithms that can make efficient use of the detailed molecular substructures learned by the transformer.

Quantifying the Dynamics of Protein Self-Organization Using Deep Learning Analysis of Atomic Force Microscopy Data.

The dynamics of protein self-assembly on the inorganic surface and the resultant geometric patterns are visualized using high-speed atomic force microscopy. The time dynamics of the classical macroscopic descriptors such as 2D fast Fourier transforms, correlation, and pair distribution functions are explored using the unsupervised linear unmixing, demonstrating the presence of static ordered and dynamic disordered phases and establishing their time dynamics. The deep learning (DL)-based workflow is developed to analyze detailed particle dynamics and explore the evolution of local geometries. Finally, we use a combination of DL feature extraction and mixture modeling to define particle neighborhoods free of physics constraints, allowing for a separation of possible classes of particle behavior and identification of the associated transitions. Overall, this work establishes the workflow for the analysis of the self-organization processes in complex systems from observational data and provides insight into the fundamental mechanisms.

Boron removal by electrocoagulation: Removal mechanism, adsorption models and factors influencing removal.

Boron (B), normally present in ground water and sea water, is a vital micronutrient for plants, but is also toxic in excessive amounts. Under typical conditions, aqueous boron is present as boric acid (H3BO3), which is uncharged, making B particularly challenging to remove by mechanisms commonly applicable to removal of trace constituents. Adsorption of B onto aluminum hydroxide solids (Al(OH)3(s)) generated using aluminum-based electrocoagulation (EC) is a promising strategy for B removal. Infrared spectroscopy analysis indicated complexation of B(OH)3 with aluminum hydroxide solids via surface hydroxyl groups, while X-ray and infrared spectroscopy results indicated that the structure of the Al(OH)3(s) was influenced both by EC operating conditions and by water quality. A linear adsorption model predicted B removal well when initial concentrations were lower than 50 mg/L, but fit the experimental data poorly at higher initial B concentrations. The Langmuir adsorption model provided a good fit for a broader range of initial B concentrations (5-1000 mg/L). Factors affecting B adsorption during the EC process, including current intensity, Al dissolution rate, boron concentration, pH, and total dissolved solid (TDS), were investigated. Increasing current intensity initially led to a higher Al dissolution rate, and therefore higher B adsorption, but there was a limit, as further increases in current intensity caused rapid formation of Al(OH)3(s) having a large particle size and a low capacity to complex B. Boron removal decreased as its concentration increased. The best removal of B occurred at pH 8, corresponding to a slightly positive zeta potential for aluminum hydroxide and a small but significant fraction of negatively charged B species. Higher TDS concentrations facilitated the use of higher current intensities, i.e., the limit on the effective Al dissolution rate increased with increasing TDS. Two real water samples (river water and oilfield produced water) spiked with B were treated using EC, resulting in up to 50% B removal from river water (C0 = 10 mg/L, current = 0.2 A) in 2 h, and 80% B removal from produced water (C0 = 50 mg/L, current = 1.0 A) in 2 h.