Enzyme engineering plays a central role in developing efficient biocatalysts for biotechnology, biomedicine, and life sciences. Apart from classical rational design and directed evolution approaches, machine learning methods have been increasingly applied to find patterns in data that help predict protein structures, improve enzyme stability, solubility, and function, predict substrate specificity, and guide rational protein design. In this Perspective, we analyze the state of the art in databases and methods used for training and validating predictors in enzyme engineering. We discuss current limitations and challenges which the community is facing and recent advancements in experimental and theoretical methods that have the potential to address those challenges. We also present our view on possible future directions for developing the applications to the design of efficient biocatalysts.