(154d) Predicting the Transition Temperature of Multi-Responsive Poly(N-isopropylacrylamide)-Based Microgels Using a Cluster-Based Partial Least Squares Modelling Approach

A variety of applications in diverse fields has been made possible by smart microgels that can swell and deswell reversibly upon exposure to external stimuli, in particular temperature (by exploiting the volume phase transition temperature (VPTT) behaviour of microgels based on poly(N-isopropylacrylamide) or other thermoresponsive polymers) and/or pH (by incorporating ionizable functional groups into the microgel network). However, microgels with a targeted VPTT value are difficult to design because various physical and chemical factors affect both the VPTT and the ionization profile of microgels; dual pH/temperature-responsive microgels present even more complexities given that both stimuli contribute to the overall response of the microgel to either variable. To overcome these challenges, a data-driven model was designed and implemented in this work to predict the swelling profile and, ultimately, the VPTT of dual pH/temperature-responsive microgels based on copolymers of N-isopropylacrylamide and a variety of carboxylic acid-containing comonomers with different copolymerization kinetics and pK_a values. In the proposed clustering-based adaptation of Partial Least Squares (PLS) modelling, the data is grouped (either in the recipe space or the swelling profile space) and individual clusters are fitted to a PLS-based model. In the subsequent step, the method is applied to a data library of 33 microgels prepared with 8 recipe variables (i.e. the amount of co-monomers, crosslinkers, and surfactants) used as inputs to the model and 12 swelling profile variables measured at 6 temperatures at pH 4 (fully protonated regime) and pH 10 (fully ionized regime) as outputs to the model. To maximize the information extracted from the existing dataset, five different re-arrangements are also suggested under all three kinds of clustering policies (no clustering, recipe-based clustering, and swelling profile-based clustering). This approach could predict the microgels' transition temperatures within 0.8℃ at pH=4 and 2.4℃ at pH=10, an accuracy in line with the error bars associated with the measured transition temperatures (0.6℃ and 2.2℃ at pH=4 and pH=10, respectively). As such, the data-driven approach can accurately predict swelling transitions in dual-responsive microgels and, in tandem, predict microgel recipes that offers potential to achieve specific swelling/deswelling targets.