Ada Sedova

Scientific workflows on large computing clusters increasingly combine massively parallel simulation engines, machine-learned models, and theory development with the technical aspects of installation, testing and debugging, batch scheduling, and designing architecture-specific GPU-accelerated software. On leadership-class systems, these workflows can be difficult to deploy successfully for production science due to compounding problems: compilation, installation, and package compatibility; accelerator runtime issues; scheduler behavior; and many others. We report early experiences using OpenAI's Codex and Anthropic's Claude Code as AI assistants for molecular simulation workflow development on OLCF resources. Our results show several important capabilities. With not much more than the user guide for the Frontier supercomputer, both assistants were able to help resolve complex build and deployment paths on this nonstandard HPC system. The overall speedup for completion of these types of difficult technical tasks, over the time it would take an unassisted human computational researcher, can approach several orders of magnitude. However, the amount of work that must be verified, due to problems with AI reliability, faithfulness and reproducibility, also grows accordingly, which in the case of HPC production workflows, can become unmanageable. This result supports the recently recognized shift from a production bottleneck to a verification bottleneck for AI in academic pursuits, including science and mathematics, and helps us to map out requirements for agentic AI for HPC-based computational science in this new shifting paradigm. Under expert supervision and with file-based provenance, AI coding systems could reduce the translation cost between scientific intent and machine-specific execution, as long as scientific truth can be codified and agentic workflows can be verified with auditable artifacts.

The accurate description of non-covalent interactions is critical for understanding the structure, dynamics, and eventual function of biomolecules. The adenine dimer serves as a benchmark system for computational methods due to its role in nucleic acid structures and its rich conformational landscape. In this study, we employ benchmark diffusion quantum Monte Carlo (DMC) methods to investigate the relative energies and role of electron correlation on a set of adenine dimer conformations generated via a search of the potential energy landscape using the global optimizer algorithm. Relative DMC energies are compared against a wide range of density functional theory (DFT) approximation results. We find that although most of the DFT functionals perform well for low-energy structures, their accuracy varies significantly for higher-energy conformations, including stacked and T-shaped structures. A large fraction of the variation is due to the treatment of the van der Waals interaction. BLYP, B3LYP, and PBE0 significantly improve with added D4 dispersion, while the recent r2SCAN-D4 and ωB97M-V functionals show the least scatter and closest agreement with the DMC. These findings highlight the delicate nature of these interactions in biomolecular systems and provide guidance for simulations of their structure and dynamics and for the development of machine learned interatomic potentials.

Beryllium fluoride is a key component of molten salt mixtures for next-generation nuclear reactors. Despite its importance, reliable experimental data on its thermophysical and thermochemical properties remain scarce, and predictive modeling is complicated by the slow structural relaxation of its highly viscous melt. Here, we investigate the structural stability and thermophysical behavior of BeF ₂ across solid and liquid phases using a combination of density functional theory (DFT) methods and two architectures of machine learning interatomic potentials (MLIPs). For crystalline BeF ₂ , we compare polymorph stabilities predicted by density functional theory and MLIPs, highlighting the critical role of dispersion corrections in stabilizing the α-quartz phase at 0 K and the ability of MLIPs trained exclusively on molten state configurations to accurately reproduce the relative stability of various polymorphs. MLIP-based simulations are then extended to molten BeF ₂ , whose high viscosity and network-forming character require trajectories on the order of tens of nanoseconds to obtain converged properties. Employing the hybrid PBE0-D3 functional, compared to the semilocal PBE-D3 functional, systematically improves agreement with experimental density, ionic conductivity, and heat capacity for molten BeF ₂ . Overall, this work demonstrates the potential of machine learning-driven simulations for predictive modeling of BeF ₂ across phases and provides a useful reference point for future studies of complex multicomponent fluoride systems.

Aliphatic amino alcohols such as 6-amino-1-hexanol are potential platform chemicals for a variety of advanced materials. Selective enzymatic synthesis from biomass-derived diols could simplify production while minimizing hazards.

Enzymes have evolved to rapidly and selectively hydrolyze diverse natural and anthropogenic polymers, but only a limited group of related enzymes have been shown to hydrolyze synthetic polyamides. In this work, we synthesized and characterized a panel of 95 diverse enzymes from the N-terminal nucleophile hydrolase superfamily with 30-50% pairwise amino acid identity. We found that nearly 40% of the enzymes had substantial nylon hydrolase activity, in many cases comparable to that of the best-characterized nylon hydrolase, NylC. There was no relationship between phylogeny and activity, nor any evidence of prior selection for nylon hydrolase activity. Several newly-identified hydrolases showed significant substrate selectivity, generating up to 20-fold higher product titers with Nylon 6,6 versus Nylon 6. Finally, we determined the crystal structure and oligomerization state of a Nylon 6,6-selective hydrolase to elucidate structural factors that could affect activity and selectivity. These new enzymes provide insights into the widespread potential for nylon hydrolase evolution and opportunities for analysis and engineering of improved hydrolases. Significance: Nylons are common industrial polyamides with few recycling options. As an alternative to mechanical or chemical recycling, enzymes may provide a selective and energy-efficient route to deconstruct nylons from mixed waste. Several nylon hydrolases have been identified, most notably NylC, but these enzymes are all closely related and demonstrated similar activity and substrate range. In this work, we investigated a diverse set of enzymes and showed that nylon hydrolase activity is common, providing new insights into the evolution of microbial nylon hydrolysis. Unlike NylC, several enzymes demonstrated unprecedented substrate selectivity, preferentially hydrolyzing Nylon 6,6 compared to Nylon 6. These enzymes can be used to understand substrate selectivity in nylon hydrolysis and to engineer enzymes for nylon recycling from mixed waste.

This work describes an analytical procedure, single particle-inductively coupled plasma-time-of-flight-mass spectrometry (SP-ICP-TOF-MS), that was developed to determine the platinum binding efficiency of protein-coated magnetic microparticles. SP-ICP-TOF-MS is advantageous due to its ability to quasi-simultaneously detect all nuclides (⁷Li–²⁴²Pu), allowing for both platinum and iron (composition of magnetic microparticles) to be measured concurrently. This method subsequently allows for the differentiation between bound and unbound platinum. The 1 μm magnetic microparticles were fully characterized for their iron concentration, particle concentration, and trace element composition by bulk digestion-ICP-MS and SP-ICP-TOF-MS. The results of both approaches agreed with the certificate values. Using the single particle methodology the platinum loading was quantified to be to 0.18 ± 0.02 fg per particle and 0.32 ± 0.02 fg per particle, for the streptavidin-coated and azurin-coated microparticles, respectively. Both streptavidin-coated and the azurin-coated microparticles had a particle–platinum association of >65%. Platinum bound samples were also analyzed via bulk digestion-based ICP-MS. The bulk ICP-MS results overestimated platinum loading due to free platinum in the samples. This highlights the importance of single particle analysis for a closer inspection of platinum binding performance. The SP-ICP-TOF-MS approach offers advantages over typical bulk digestion methods by eliminating laborious sample preparation, enabling differentiation between bound/unbound platinum in a solution, and quantification of platinum on a particle-by-particle basis. The procedure presented here enables quantification of metal content per particle, which could be broadly implemented for other single particle applications.

In this work, we expand on a dataset recently introduced for protein interface prediction (PIP), the Database of Interacting Protein Structures (DIPS), to present DIPS-Plus, an enhanced, feature-rich dataset of 42,112 complexes for machine learning of protein interfaces. While the original DIPS dataset contains only the Cartesian coordinates for atoms contained in the protein complex along with their types, DIPS-Plus contains multiple residue-level features including surface proximities, half-sphere amino acid compositions, and new profile hidden Markov model (HMM)-based sequence features for each amino acid, providing researchers a curated feature bank for training protein interface prediction methods. We demonstrate through rigorous benchmarks that training an existing state-of-the-art (SOTA) model for PIP on DIPS-Plus yields new SOTA results, surpassing the performance of some of the latest models trained on residue-level and atom-level encodings of protein complexes to date.

Sphagnum-dominated peatlands store a substantial amount of terrestrial carbon. The genus is undersampled and under-studied. No experimental crystal structure from any Sphagnum species exists in the Protein Data Bank and fewer than 200 Sphagnum-related genes have structural models available in the AlphaFold Protein Structure Database. Tools and resources are needed to help bridge these gaps, and to enable the analysis of other structural proteomes now made possible by accurate structure prediction. We present the predicted structural proteome (25 134 primary transcripts) of Sphagnum divinum computed using AlphaFold, structural alignment results of all high-confidence models against an annotated nonredundant crystallographic database of over 90,000 structures, a structure-based classification of putative Enzyme Commission (EC) numbers across this proteome, and the computational method to perform this proteome-scale structure-based annotation. All data and code are available in public repositories, detailed at https://github.com/BSDExabio/SAFA. The structural models of the S. divinum proteome have been deposited in the ModelArchive repository at https://modelarchive.org/doi/10.5452/ma-ornl-sphdiv.

Summary: Easy-to-use, open-source, general-purpose programs for modeling a protein structure from inter-atomic distances are needed for modeling from experimental data and refinement of predicted protein structures. OpenMDlr is an open-source Python package for modeling protein structures from pairwise distances between any atoms, and optionally, dihedral angles. We provide a user-friendly input format for harnessing modern biomolecular force fields in an easy-to-install package that can efficiently make use of multiple compute cores.Availability and Implementation: OpenMDlr is available at https://github.com/BSDExabio/OpenMDlr-amber. The package is written in Python (versions 3.x). All dependencies are open-source and can be installed with the Conda package management system.Supplementary information: Supplementary data are available at Bioinformatics online.