Biological function often arises from the interactions of proteins with other biomolecules, and these interactions are grounded in the three-dimensional structures of the complexes. My lab develops computational tools to study the atomic structure of protein interfaces to explain not only how biological and disease processes work but also how one might alter these processes at the molecular level. Traditionally, creating computational tools first requires solving problems of basic science including (1) how to sample the myriad conformations available to proteins and (2) how to accurately calculate the energy of each conformation. In this talk, I first will share our biophysical approaches to protein docking and the difficult challenge of capturing conformational change upon binding. We have made progress through novel sampling algorithms using libraries of protein conformations and through rapid, course-grained scoring schemes using large data sets. I will finish with recent work recent work exploiting deep learning approaches to predict antibody CDR H3 loop structure, the key feature for antibody-antigen binding, extensions to predict the entire Fv structure, and implications for design. I will discuss the philosophy of emerging deep learning approaches, the performance on a practical but general antibody problem, the capability to learn general features of protein structure, and the outlook for these approaches for molecular engineering in general.