Significant advancements have been made in the accuracy of bioinformatic predictive modeling over recent years. As these predictive models become increasingly utilized in biomedical research, it is important to study and quantify their predictive accuracy in relation to real-world mutations. These technologies provide quick, efficient, and cost-effective techniques to discover and investigate potentially detrimental mutations in human proteins. However, the clinical value of these predictive models has been mostly unproven. We sought to design an experiment that would allow us to quantify the accuracy of these_ in silico_ predictive models in relation to _in vitro_ assays. To determine the validity of these predictive models in future research and clinical studies, we conducted _in vitro_ assays that quantified the effects of numerous mutations on the human enzyme GALT. All three variants tested (A81T, H47D, E58K) showed decreased enzymatic activity relative to native protein. When compared to the results of numerous predictive models (Molecular dynamics RMSD comparison, PredictSNP, EVE, ConSurf, SIFT), our results indicated extensive discrepancies between the predicted impact of mutations versus their actual effects on GALT function _in vitro_. We were able to quantifiably demonstrate discrepancies between predictive models and _in vitro_ studies across several mutations within the GALT protein. These results indicate possible issues with relying on predictive modeling to determine the effects of mutations of protein function. Further studies to elucidate the clinical value of these predictive models are warranted, not only for GALT but also for any other proteins of human clinical significance.