Tight integration of computer-assisted molecular design with practical realization by medicinal chemistry will be essential for finding next-generation drugs that are optimized for multiple pharmaceutically relevant properties. ETH Zürich has established an interdisciplinary research group devoted to exploring the potential of this scientific approach by combining expertise from pharmaceutical chemistry and computer sciences. In this article, some of the group's activities and projects are presented. A current focus is on machine-learning applications aiming at hit and lead structure identification by virtual screening and de novo design. The central concept of 'adaptive fitness landscapes' is highlighted along with practical examples from drug discovery projects.