Agnieszka Gajewicz-Skretna is an associate professor at the Faculty of Chemistry, Gdańsk University of Technology, Poland. She has over 15 years of experience in the development and application of in silico approaches to predict the toxicity, environmental behaviour and physicochemical properties of chemicals, nanomaterials and advanced materials. She completed a one-year postdoctoral fellowship at the National Institute for Environmental Studies in Tsukuba, Japan, and has also undertaken research stays in the USA and Germany. Her research has resulted in over 80 scientific publications in the areas of computational toxicology, chemoinformatics, nanotoxicology, environmental chemistry and materials safety. She has participated in numerous national and international research projects in the areas of computational toxicology, chemical and nanomaterial safety, environmental risk assessment, and safe-and-sustainable-by-design. Her broader research interests include improving the reliability, interpretability, and transferability of in silico models, as well as supporting their use in next-generation hazard and safety assessment.
OpenTox Virtual Conference 2026
From Material Properties to Biological Context: Rethinking Predictive Toxicology
Predictive toxicology is undergoing a fundamental transition from conventional structure–activity modeling toward more integrated approaches that reflect the complexity of real biological systems. In the case of nanomaterials, advanced materials, and micro- and nanoplastics, this transition is particularly important because toxicity cannot be understood solely from intrinsic physicochemical properties. Biological responses are shaped by the interplay between material characteristics, exposure conditions, time, and the biological context in which effects are measured. Capturing this complexity is becoming essential for the development of computational New Approach Methodologies (NAMs) that are not only predictive, but also biologically meaningful and transferable across experimental settings. This presentation will illustrate how machine-learning-based QSAR approaches can evolve from static, single-endpoint models toward dynamic, multi-output, and context-aware frameworks. Dynamic modeling introduces exposure-related variables, such as dose and time, into the predictive space, enabling toxicity to be interpreted as a response that changes under different exposure conditions rather than as a fixed property of the material. Multi-output modeling extends this concept by considering several biologically relevant responses simultaneously, providing a more integrated representation of toxicological effects while addressing the incompleteness that frequently characterizes experimental datasets. The presentation will also address a broader question that is becoming increasingly important in predictive toxicology: how much biological context is required to generate meaningful and transferable predictions? Using examples from nanomaterials, advanced materials, and micro- and nanoplastics, the discussion will consider the extent to which variables such as species, cell type, experimental system, exposure conditions, and material characteristics contribute to predictive performance and model interpretation. Overall, the presentation will highlight a shift from asking whether a material is intrinsically toxic toward understanding when, under which conditions, and in which biological context toxicity emerges. Such a perspective may help make predictive toxicology more mechanistically informative, biologically relevant, and useful for next-generation safety assessment of chemicals and advanced materials.