Presentation: 2025 ND EPSCoR Annual conference
October 21, 2025, NDSU Memorial Union, Fargo, North Dakota
Application of AI/ML methods to Develop QSAR Models for Multiple Fouling-Release Coating Systems.
Estefania
Ascencio
Doctoral Student
North Dakota State University
Co-authors: Estefania Ascencio1,2,3, Gerardo M. Casanola-Martin1,2, Amirreza Daghighi 1,4, Achiya Khanam1, Shan He 1,2,3 Sonia Arrasate2, Humberto González-Díaz 2,5, and Bakhtiyor Rasulev1,4 1Department of Coatings and Polymeric Materials, North Dakota State University, Fargo, ND 58102, USA 2Department of Organic and Inorganic Chemistry, Faculty of Science and Technology, University of The Basque Country (UPV/EHU), P.O. Box 644, 48080, 3IKERDATA S.L., ZITEK, University of The Basque Country (UPVEHU), Rectorate Building, 48940 Leioa, Spain. 4Biomedical Engineering Program, North Dakota State University, Fargo, ND 58105, USA. 5IKERBASQUE, Basque Foundation for Science, 48011 Bilbao, Biscay, Spain
Session
Poster number: 64
Ballroom
Abstract Biofouling represents an environmental and economic problem caused by the accumulation of microorganisms, such as Cellulophaga lytica and Navicula incerta, on submerged surfaces, increasing hydrodynamic resistance and CO₂ emissions. Fouling-release (FR) coatings offer a non-toxic alternative by facilitating the removal of these organisms through hydrodynamic forces. This study covers development of artificial intelligence (AI) and machine learning (ML)-based Quantitative Structure-Activity Relationship (QSAR) predictive models to estimate the fouling-release efficiency related to specific microorganisms, such as N. incerta and C. lytica. The datasets were collected that consist of 75 and 87 data fouling release coating systems. For this study, to describe complex coating system a mixture-descriptor approach was applied, treating each coating as a mixture system. For example, the key components analyzed in this study include such polymeric blocks as PDMS, PMHS, PEG, and SBMA. A set of AI/ML alogrithms were aplied in this study to develop the QSAR model and Gradient Boosting (GB) models demonstrated the best performance, with R² values of 0.96 and 0.90 for training and test sets for N. incerta as an endpoint, and 0.85 and 0.76 for C. lytica. These results highlight the importance of several descriptors that are responsible for electronegativity, topological charge and number of valence electrons. The developed models and involved combinatorial descriptors underscore the relevance of complex relationships between variables and the synergistic effects of the components on fouling-release performance. In future, the online AI/ML-based web-applet will be generated as well, to allow experimental scientists to predict fouling-release properties for the coating systems