Fragment-based drug discovery (FBDD) has established itself as a powerful tool for developing probe and drug candidates by rationally elaborating small chemical fragment hits into larger, optimised lead compounds. The use of X-ray crystallography as a medium-throughput screening tool for FBDD results in a wealth of structural data on low molecular weight molecules in complex with a protein target. Interpreting this data and distilling it into prioritised suggestions for the elaboration of fragment hits into leads with increased potency and selectivity for the target protein is currently a significant challenge to using the technique. Thus, computational methods and pipelines designed to streamline and automate the process, providing follow-up hypotheses in an objective and high-throughput way, are in high demand. Fragment hotspot mapping is a computational method that highlights specific interactions within a protein's binding site that drive the binding of small molecule fragments. As crystallographic FBDD campaigns result in an ensemble of structures of the same protein, a method to combine fragment hotspot maps information for these structures into an "ensemble map" for the protein target was developed. A workflow for comparing ensemble maps between a target and a related off-target protein was implemented and extended to allow comparisons across a protein family. This workflow was applied to examples from the well-researched human bromodomain and kinase families, and was able to identify selectivity-determining regions that have been exploited in past drug discovery campaigns. Dynamic undocking, a steered molecular dynamics method for estimating the structural stability of protein-ligand complexes, was then investigated as a way of characterising specific binding site interactions. To facilitate integration into computational workflows, an open-source implementation of the method was benchmarked and shown to perform comparably. A workflow combining the extended fragment hotspot maps method, dynamic undocking, docking and a chemistry recommendation engine was developed and used to suggest follow-up compounds in three ongoing medicinal chemistry projects. The compounds showed detectable binding affinity, a significant improvement from the starting fragment hits, demonstrating the workflow's utility in the initial round of compound elaboration.