(59ar) Designing pH-Temperature Responsive Microgels with Targeted Transition Temperature Using a Novel Partial Least Squares (PLS) Model Inversion Technique

A wide range of applications in a wide range of fields has been made possible by smart microgels that can swell and deswell reversibly in response to external stimuli. In particular, microgels that respond reversibly to temperature (i.e. microgels with a volume phase transition temperature or VPTT, typical of microgels based on poly(N-isopropylacrylamide) or other thermoresponsive polymers) and/or pH (typically formed by copolymerizing ionizable functional groups into a microgel network) have attracted application interest. However, designing temperature-responsive microgels with a specific VPTT value is challenging given that a variety of physical and chemical factors influence the VPTT; designing dual pH/temperature-responsive microgels is even more challenging because both stimuli affect the response of a given microgel to either variable. In our prior work, a PLS data-driven model coupled with a clustering approach was developed to predict each microgel's final swelling profile based on recipes of microgels identify to be contained in the same cluster, allowing for good predictions of VPTT values in both the protonated (pH 4) and ionized (pH 10) states. In this work, we leverage these predictions to a priori design a dual pH/temperature-responsive microgel with targeted VPTT values. Taking into account the three different data blocks available (i.e. the microgel recipe, the measured swelling profile, and the VPTT value), a systematic strategy was introduced to facilitate for the recognition of the best possible targeted VPTT value based on both the synthetic feasibility of the proposed recipe as well as the similarity of this targeted VPTT value to the corresponding training data set (assuming that more similar VPTT values are more likely to result in an accurate prediction). Following, a new PLS model inversion technique was developed to identify potential recipes matching the desired VPTT values. In this technique, a second PLS model was developed in which the input of the second PLS model is the output scores (U) of the first PLS model and the output of the second PLS model is the input score (T) of the first PLS model. To demonstrate the superior performance of our introduced model inversion technique versus the conventional PLS model inversion technique, the same VPTT targets were applied to both techniques and the resulting suggested recipes were synthesized for experimental validation. Our new model inversion technique suggested recipes yielding colloidally stable microgels with VPTT values close to the target values; in contrast, the recipes suggested by the conventional model inversion technique exhibited bulk aggregation. By identifying the best goal within the models' validity space and the best potential recipe to achieve the target properties, the approach introduced has the potential to accelerate the dual-responsive microgel design process as well as other materials optimization processes.