Seminars Archive

Predictive modeling of nanomaterials and new opportunities for micro-CT data

Abstract
Modern materials research increasingly depends not only on acquiring high-quality experimental data, but also on extracting from them quantitative and predictive knowledge. In the field of nanomaterials, this challenge is particularly pronounced: material behavior often depends on many interrelated physicochemical characteristics and environmental conditions, while experimental testing remains time-consuming, costly, and difficult to generalize. Over the past years, these limitations have motivated the development of nano-QSAR, nanoinformatics, and machine-learning-based strategies aimed at linking material descriptors with measurable properties, transformations, and biological effects.
In this talk, I will present the research direction I work on at the University of Gdansk, developed in the broader context of the work of Tomasz Puzyn's group and collaborators. Our approach combines data curation, descriptor engineering, and predictive modeling to study complex nanomaterial endpoints, including dissolution processes, colloidal properties, and nano-bio interactions. Rather than focusing on machine learning as a black box, I will show how interpretable descriptor-based models can be used to identify key variables, quantify relationships, and support understanding of material behavior under different conditions.
I will discuss in particular our work on modeling the dissolution of engineered nanomaterials, where material composition, structure, and environmental parameters are integrated into predictive frameworks. These studies illustrate how relatively heterogeneous experimental data can be transformed into robust modeling workflows, and how such workflows can support screening, hypothesis generation, and the design of new experiments. Additionally, I would like to discusse the question of the reliability of such models and their applicability within the context of current European standards warrants deliberation.
Finally, I will address how this way of thinking may be transferred to micro-CT research. Although micro-CT and nanoinformatics operate on different types of data, both face a similar challenge: how to move from rich but complex measurements to reproducible descriptors, interpretable models, and useful predictions. I will outline possible ways in which QSAR-inspired and machine-learning-based methodologies could be adapted to micro-CT datasets, with the aim of linking image-derived features and experimental metadata to material properties, sample classification, or process-relevant outcomes. The broader goal of the seminar is to open a discussion on how descriptor-based predictive modeling may complement micro-CT research at Elettra