While interest in the use of enzymes as biocatalysts for chemical applications is increasing, the performance of natural enzymes is rarely adequate for a commercially viable process.  Despite advances in protein engineering methods, there is a continuing need for more efficient and effective methods to improve multiple biochemical characteristics of enzymes simultaneously. We have developed a hybrid strategy for directed evolution that promises to enhance enzyme properties such as activity, specificity and stability. The key to the approach is to incorporate machine learning into each round of experimental evolution by assessing the value of mutations through use of the protein sequence activity relationship (ProSAR) algorithm.  ProSAR uses data from a set of multiply mutated variants to estimate the contributions of each mutation to the property of interest and thereby guide selection of mutations that should be carried over to the next round of evolution.  Used in concert with in vitro genetic recombination, this statistical approach exploits additional information embedded in sequence-activity data to enable a mutation-oriented approach to enzyme optimization that is substantially more efficient and less blind than traditional hit-oriented approaches.  The approach is now used routinely at Codexis to rapidly optimize biocatalysts to enable commercially viable “green chemistry” based manufacturing routes.