Recombinant protein production is a cornerstone of a variety of molecular life science research studies, as well as a key enabling technology for biotechnological applications. Eukaryotic cell-based protein secretion is a preferred mode of recombinant protein production. However, protein secretion is a complex process of cellular machinery interacting with the nascent protein, and its complexity makes that prediction of (fragment) secretability of a given protein is in general still impossible. In this project, we will use a new technology that was recently developed in our laboratory, which generates experimental secretability data on hundreds of thousands of protein fragments, in yeast. We have already demonstrated that these datasets, in combination with modern machine learning, reveal unexpected protein features that are important for protein secretion, and that predictive models can be built which can predict whether a particular protein fragment is likely to be secretable or not. This methodology allows to tackle longstanding research questions about the secretory system, such as how the nature of the secretory signal influences protein secretion and which protein types require which chaperones for productive folding. We expect that the generated knowledge will be immensely useful for the fundamental understanding of the eukaryotic secretory processes, as well as for overcoming practical issues in heterologous protein secretion.