arXiv subjects

Ulrich S. Schubert

Publications and source records attributed to Ulrich S. Schubert.

4 recordsLinked to original sources

arXiv2026

It's All Connected: Topology-Aware Structural Graph Encoding Improves Performance on Polymer Prediction

Graph Neural Networks (GNNs) have achieved strong results in molecular property prediction, but polymers present distinct challenges: labeled datasets are scarce and small (typically in the order of hundreds of polymers) due to the need for expensive experimentation, and complex polymer chain distributions influence polymer properties. Established practice in polymer prediction represents polymers solely by graphs of their repeat units, discarding the chain-scale morphology that governs key properties such as the glass transition temperature ($T_g$). In this work, we propose a principled graph construction that addresses this gap. Given a polymer's molecular mass distribution (MMD), we sample representative chains from the Schulz-Zimm distribution and construct representative sets of large graphs encoding chain-scale topology directly, with atoms and bonds featurized using rich chemical descriptors. We further pretrain GNN encoders via masked graph modeling on 100,000 unlabeled PSMILES strings before fine-tuning on labeled data. On a dataset of 381 polymers (180 homopolymers and 201 copolymers), we show that graph construction and self-supervised pretraining are jointly necessary: without pretraining, the large graph method matches the repeat-unit baseline (28.40 K vs. 28.36 K RMSE); with pretraining, it achieves 24.76 K +/- 3.30 K, a 5.1% reduction in mean error over the pretrained repeat-unit baseline (26.08 K +/- 4.20 K, p < 0.001, 30 runs). An ablation removing chemical features degrades performance to 36.65 K, confirming both components are essential. Results are architecture-agnostic, holding for both GINE and GATv2 encoders.

cs.LG↗

arXiv2024

Are large language models superhuman chemists?

Large language models (LLMs) have gained widespread interest due to their ability to process human language and perform tasks on which they have not been explicitly trained. However, we possess only a limited systematic understanding of the chemical capabilities of LLMs, which would be required to improve models and mitigate potential harm. Here, we introduce "ChemBench," an automated framework for evaluating the chemical knowledge and reasoning abilities of state-of-the-art LLMs against the expertise of chemists. We curated more than 2,700 question-answer pairs, evaluated leading open- and closed-source LLMs, and found that the best models outperformed the best human chemists in our study on average. However, the models struggle with some basic tasks and provide overconfident predictions. These findings reveal LLMs' impressive chemical capabilities while emphasizing the need for further research to improve their safety and usefulness. They also suggest adapting chemistry education and show the value of benchmarking frameworks for evaluating LLMs in specific domains.

cs.LG↗

arXiv2024

Thermodynamic compatibility of actives encapsulated into PEG-PLA nanoparticles: In Silico predictions and experimental verification

Achieving optimal solubility of active substances in polymeric carriers is of fundamental importance for a number of industrial applications, including targeted drug delivery within the growing field of nanomedicine. However, its experimental optimization using a trial-and-error approach is cumbersome and time-consuming. Here, an approach based on molecular dynamics (MD) simulations and the Flory-Huggins theory is proposed for rapid prediction of thermodynamic compatibility between active species and copolymers comprising hydrophilic and hydrophobic segments. In contrast to similar methods, our approach offers high computational efficiency by employing MD simulations that avoid explicit consideration of the actual copolymer chains. The accuracy of the method is demonstrated for compatibility predictions between pyrene and nile red as model dyes as well as indomethacin as model drug and copolymers containing blocks of poly(ethylene glycol) and poly(lactic acid) in different ratios. The results of the simulations are directly verified by comparison with the observed encapsulation efficiency of nanoparticles prepared by nanoprecipitation.

cond-mat.soft↗

arXiv2023

Self-Assembly of Core-Shell Hybrid Nanoparticles by Directional Crystallization of Grafted Polymers

Nanoparticle self-assembly is an efficient bottom-up strategy for the creation of nanostructures. In the standard approach, ligands are grafted on the surfaces of nanoparticles to keep them separated and control interparticle interactions. Ligands then remain secondary and usually are not expected to order significantly during superstructure formation. Here, we investigate how ligands can play a more primary role in the formation of inorganic-organic hybrid materials. We graft poly(2-iso-propyl-2-oxazoline) (PiPrOx) as a crystallizable shell onto SiO$_2$ nanoparticles. By varying the PiPrOx grafting density, solution stability, and nanoparticle aggregation behavior can be controlled. Upon prolonged heating, anisotropic nanostructures form in conjunction with the crystallization of the ligands. Self-assembly of hybrid PiPrOx@SiO$_2$ (shell@core) nanoparticles proceeds in two steps: First, rapid formation of amorphous aggregates via gelation, mediated by the interaction between nanoparticles through grafted polymers; second, slow radial growth of fibers via directional crystallization, governed by the incorporation of crystalline ribbons formed from unbound polymers coupling to the grafted polymer shell. Our work reveals how crystallization-driven self-assembly of ligands can create intricate hybrid nanostructures.

cond-mat.soft↗