The foundation of protein engineering lies in introducing new mutations for the purpose of modifying protein function. The grand challenge is in developing accurate models to predict the amino acid sequence to function relationships. Protein function is often indirectly measured by a correlated proxy fitness function, and predicting the change in fitness from wild type of a given mutation is always the limiting factor when introducing changes to a given amino acid sequence \cite{eriksson,romero,maynard}.
This work is an analysis of how one can use unsupervised deep learning to discover features in proteins, and how these features can assist in protein engineering and biohybrid materials design.