The use of computational tools has become an increasingly popular tool for engineering protein function. While there are numerous examples of computational tools enabling the design of novel protein functions, there remains room for improvement in both prediction accuracy and success. To improve algorithms for functional and stability predictions, we have initiated the development of a data set designed to be used for training new computational algorithms for enzyme design. To date our dataset is composed of over 129 mutants with associated expression levels, kinetic data, and thermal stability for the enzyme glucosidase B (BglB) from Paenibacillus polymyxa. In this study, we introduced three new variants (M319C, T431I, and K337D) to our existing dataset with the goal of cultivating a larger dataset to train new design algorithms and more broadly explore structure-function relationships in BglB.