Synthesis-Based Design

De novo molecular design offers a promising avenue to accelerated molecular design by facilitating a more efficient and expansive exploration of chemical space than traditional screening techniques permit. However, the practical application of these algorithms, along with experimental validations of their outputs, remains limited. A primary hindrance is the often-overlooked factor of synthesizability. In contrast to screening libraries, where molecules are typically accessible, de novo strategies are designed to explore broader chemical space by assembling molecules atom-by-atom or fragment-by-fragment in silico, which makes the incorporation of synthetic feasibility challenging. This chapter discusses "synthesis-based design", a subset of de novo design approaches that restricts exploration to synthesizable chemical space. We first introduce the problem formulation and the main motivation behind this approach, followed by a historical overview and description of recent strategies augmented by machine learning. As a case study, we provide an in-depth examination of a synthesis-based design method: SynNet. We conclude by discussing existing challenges and future directions in the field.