Abstract: A method includes receiving one input feature and one output feature of importance of polymers used to stabilize a protein. A set of polymers from a library are identified based on the input feature and the output feature of importance that are applied to a machine learning model. Data for each polymer in the library includes features for each polymer and reagents for stabilizing the protein. Each polymer in the identified set is used to stabilize samples of the protein in well plates in a well plate array based on the reagents from the library data in the identified set. A score for each sample of the protein is assigned by comparing the measured output feature from the well plates corresponding to the identified set to the output feature of importance. The samples of the protein are identified having scores higher than a predefined threshold.