Fragment-based drug design is an established strategy of finding new drugs. Instead of doing mass screening of chemical libraries containing compounds that are already lead-like, starting from a small fragment can be be a more efficient strategy, because fragment space can actually represent a much larger chemical space than an equally sized library of lead-like compounds. This strategy has led to the discovery of binders for targets where high-throughput screening has previously failed. However, because fragments are small they bind with low affinity, and must be optimized into high affinity and lead-like compounds, and fragment to lead optimization is challenging. This thesis describes the development and validation of an automated fragment-to-lead optimization pipeline. The pipeline can be seen as a focused virtual screening of the chemical space surrounding a given fragment. The pipeline does the virtual screening iteratively to harnesses information about the chemotypes and features of the hits. It does this by determining the most likely to bind analogues through complementary structure-based methods, and then using the best hits to determine the chemical space for the next round of virtual screening. It has been developed to be scalable to screen the continuously growing available libraries. Furthermore, it performs structure-based scaffold hopping, to explore as much chemical space as possible. It's iterative nature makes it possible to seamlessly integrate it into drug discovery pipelines, and also makes it possible to control the ligand efficiency, i.e. maintaining only the most important part of the molecules, necessary for molecular recognition. Our results show that the platform is capable of finding active ligands for known targets such as BRD4, HSP90, DYRK1A, but also unknown targets e.g. NUDT21. Furthermore, it's been shown to be at least as equally successful as other virtual screening methods.