Describing biological phenomena using physics is a difficult problem. Biological systems such as cells and tissues exhibit emergent behavior which is driven by genes, proteins, and the interplay and feedback loops between them. To capture biological behavior in a physical theory, one has to sort through this complexity, often by hand, and determine how to account for these sub-cellular interactions. This physical theory must then be connected to experimental reality, posing an even greater challenge. To reconcile the laws of physics with the complexity of nature, one must sift through large experimental datasets in order to find the critical details which enable connections between these two pictures. This thesis presents machine learning as a tool to streamline this process. Our method of data-driven biophysical modeling combines physical theory, biological insight, and machine learning to characterize and understand diverse phenomena. Using experimental case studies on protein dynamics, cell mechanics, and fruit fly embryo development, we show how this approach can not only predict the future of complex systems, but also help uncover interpretable rules governing their behavior.