Molecular dynamics (MD) simulations are a powerful tool for understanding complex molecular behavior, but exhaustively exploring the large space of input parameters can be computationally prohibitive, especially when the outcomes are noisy and/or expensive to evaluate. In this work, we introduce MD-Bayesian algorithm execution (BAX), a general-purpose, automated design framework that builds on BAX acquisition strategy to efficiently guide simulation campaigns toward learning meaningful features of the system. Unlike optimization-centric Bayesian optimization approaches, MD-BAX seeks to identify broader system properties (e.g., phase transition boundaries, level sets, and threshold crossings) by strategically selecting input/parameter settings based on uncertainty. To accurately represent the variability in simulation outcomes, MD-BAX incorporates a Gaussian process surrogate model with input-dependent noise, estimated directly from MD trajectory statistics at each simulation setting. This enables construction of reliable uncertainty estimates for guiding the next simulation. We demonstrate the approach on a case study involving coil-to-globule transitions in amphiphilic block copolymers, highlighting that explicitly including trajectory-derived noise improves uncertainty calibration and enables our framework to more efficiently map the relationship between polymer structure, solvent quality, and conformational behavior. MD-BAX represents a domain-informed specialization of the BAX framework for MD and is broadly applicable to molecular modeling problems where the goal is to infer key system behaviors from stochastic, trajectory-based simulation outputs rather than to locate a single optimal condition.

Designing molecules that must satisfy multiple, often conflicting, objectives is a central challenge in molecular discovery. The enormous size of the chemical space and the cost of high-fidelity simulations have driven the development of machine learning-guided strategies for accelerating design with limited data. Among these, Bayesian optimization (BO) offers a principled framework for sample-efficient search, while generative models provide a mechanism to propose novel, diverse candidates beyond fixed libraries. However, existing methods that couple the two often rely on continuous latent spaces, which introduce both architectural entanglement and scalability challenges. This work introduces an alternative, modular "generate-then-optimize" framework for de novo multiobjective molecular design/discovery. At each iteration, a generative model is used to construct a large, diverse pool of candidate molecules, after which a novel acquisition function, qPMHI (multipoint Probability of Maximum Hypervolume Improvement), is used to optimally select a batch of candidates most likely to induce the largest Pareto front expansion. The key insight is that qPMHI decomposes additively, enabling exact, scalable batch selection via only a simple ranking of probabilities that can be easily estimated with Monte Carlo sampling. We benchmark the framework against state-of-the-art latent-space and discrete molecular optimization methods, demonstrating significant improvements across synthetic benchmarks and application-driven tasks. Specifically, in a case study related to sustainable energy storage, we show that our approach quickly uncovers novel, diverse, and high-performing organic (quinone-based) cathode materials for aqueous redox flow battery applications.

Microplastic (MP) pollution, which is present in the ecosystem in vast quantities, adversely affects human health and the environment, making it imperative to develop methods for its mitigation. The challenge of detecting or capturing MPs could potentially be addressed using plastic-binding peptides (PBPs). The ideal PBP for MP remediation would not only bind strongly to plastic, but also have other properties such as high solubility in water or great binding specificity to a certain plastic. However, the scarcity or absence of known PBPs for common plastics along with the lack of methods that can discover PBPs with all of the desired properties precludes the development of peptide-based MP remediation strategies. In this study, we discovered short linear PBPs with high predicted water solubility and binding specificity by employing an in-silico discovery pipeline that combines deep learning and biophysical modeling. First, a long short-term memory (LSTM) network was trained on biophysical modeling data to predict peptide affinity to plastic. High affinity peptides were generated by pairing the trained LSTM with a Monte Carlo tree search (MCTS) algorithm. Molecular dynamics (MD) simulations showed that the PBPs discovered for polyethylene, the most common plastic, had 15% lower binding free energy than PBPs obtained using biophysical modeling alone. PBPs with both high affinity and high predicted solubility in water were found by including the CamSol solubility score in the MCTS peptide scoring function, increasing the average solubility score from 0.2 to 0.9, while only minimally decreasing affinity for polyethylene. The framework also discovered peptides with high binding specificity between polystyrene and polyethylene, two major constituents of MP pollution, using a competitive MCTS approach that optimized the difference in affinity between the two plastics. MD simulations showed that competitive MCTS increased the binding specificity of PBPs for polystyrene and identified peptides with relatively great preference for either of the two plastics. The framework can readily be applied to design PBPs for other types of plastic. Overall, the high-affinity PBPs with desirable properties discovered by marrying artificial intelligence and biophysics can be valuable for remediating MP pollution and protecting the health of humans and the environment.

Soil organic carbon (SOC) plays a vital role in global carbon cycling and sequestration, underpinning the need for a comprehensive understanding of its distribution and controls. This study explores the importance of various covariates on SOC spatial distribution at both local (up to 1.25 km) and continental (USA) scales using a deep learning approach. Our findings highlight the significant role of terrain attributes in predicting SOC concentration distribution with terrain, contributing approximately one-third of the overall prediction at the local scale. At the continental scale, climate is only 1.2 times more important than terrain in predicting SOC distribution, whereas at the local scale, the structural pattern of terrain is 14 and 2 times more important than climate and vegetation, respectively. We underscore that terrain attributes, while being integral to the SOC distribution at all scales, are stronger predictors at the local scale with explicit spatial arrangement information. While this observational study does not assess causal mechanisms, our analysis nonetheless presents a nuanced perspective about SOC spatial distribution, which suggests disparate predictors of SOC at local and continental scales. The insights gained from this study have implications for improved SOC mapping, decision support tools, and land management strategies, aiding in the development of effective carbon sequestration initiatives and enhancing climate mitigation efforts.

Polyelectrolyte complex based micelles have attracted significant attention due to their potential regarding bio-applications. Although the morphology and functions have been studied extensively, dynamic properties, particularly component exchange with other surrounding molecules, have remained elusive to date. Here, we show how micelles based on metal-ligand coordination complex coacervate-core micelles (M-C3Ms) respond to addition of extra ligand and metal ions. The micelles are prepared from a polycationic-neutral diblock copolymer and an anionic coordination polyelectrolyte, which is obtained by coordination between metal ions (lanthanides Ln 3+ and Zn 2+ ) and a bis-ligand (LEO) containing two dipicolinic acid (DPA) groups connected by a tetra-ethylene oxide spacer (4EO). Our findings show that the bis-ligand LEO is essential for the growth of coordination polymers and consequently the formation of micelles, leading to equilibrium structures with the same micellar composition and structure independent of the order of mixing. In other words, adding single DPA has no effect on the formed M-C3Ms. As for metal exchange, we find that added Zn 2+ can replace some of the Ln 3+ from Ln-C3Ms, leading to a hybrid coordination structure with both Ln 3+ and Zn 2+ . We find that component exchange occurs in these coordination polyelectrolyte micelles, but it is more favorable in the direction of replacing the weak binding components with strong ones. Hence, the designed M-C3Ms based on the strong binding components, such as Ln-C3Ms, shall be relatively stable in biological surroundings, paving the way for the application of such particles as bio-imaging probes.