Pathologists must rapidly interpret microscopic images of a tissue specimen to provide a diagnosis. To diagnose, a pathologist may additionally consider or interpret patient history, radiology, specialized histological stains, genetic sequencing, etc. Integrating into an accurate predictive model each patient’s diverse information is one of many challenges in computational pathology. This thesis’ scope is computational pathology methods that seek to empower pathologists by automating tedious work, providing new capabilities, or finding similar patient cases. First, we video-recorded pathologists diagnosing at a microscope, and found deep learning could accurately predict where pathologist gaze would dwell. Such areas may be priority regions of interest for diagnosis. Second, we applied deep learning to prostate cancer whole slide images to predict if a cancerous tumor has a SPOP gene mutation. Such methods may expand the capabilities of hospitals where genomic sequencing is unavailable. Third, we integrated histological imaging and clinical information into a multimodal method to find similar patient cases across social media and notify the pathologists having these cases. Such a method brings to low-resource hospitals the subspecialty expertise of pathologists worldwide. Taken together, we conclude it is practical for such computational methods to empower pathologists on an international scale. In doing so, we were the first to predict a somatic mutation from histology images of prostate cancer, we uncovered the utility of pathology data on social media, and we devised novel methods to interpret deep learning.