In the Harms lab, we are interested in the interplay between the biophysical properties of proteins and their evolution. I will discuss two ongoing projects. In the first, we are investigating how the map between protein genotype and phenotype shapes evolutionary outcomes. We have uncovered extensive multi-way interactions between mutations, meaning that the effect of a given mutation depends strongly on the presence of two or more other mutations. The magnitudes of these interactions are small, but they have an out-size effect on the accessibility of evolutionary trajectories. We have further found that we can produce these multi-way interactions using a simple, statistical thermodynamic model of proteins. This work reveals that the effect of mutations will be different if they occur early or late in evolution, and that knowing the effects of a mutation early in a trajectory is insufficient to predict future evolution. In the second project, we are experimentally investigating the evolution of new peptide binding specificity in several members of the S100 protein family. These proteins bind to extremely diverse, short regions of target proteins. Despite this apparent lack of specificity, we find that the same peptide binding profile has been conserved for different protein family members for the last 300 million years, suggesting this profile has been maintained by natural selection. To understand the origins of the specificity of these proteins, we next constructed a predictive binding model using a high-throughput peptide interaction assays coupled to supervised machine learning. We find that these proteins discriminate peptides based largely on packing and shape criteria rather than specific polar contacts. Further, by repeating these analyses on reconstructed ancestral proteins, we were able to reveal that protein specificity increased on one lineage while decreasing along the other. This reveals that extremely “sloppy” low-specificity proteins exhibit evolutionary patterns similar to those of well-studied high-specificity proteins.