Full Modification Control over Retrosynthetic Routes for Guided Optimization of Lead Structures
Uschi Dolfus1, Hans Briem2 and Matthias Rarey1
1Universität Hamburg, ZBH – Center for Bioinformatics, Bundesstraße 43, 20146 Hamburg
2Bayer AG, Research & Development, Pharmaceuticals, Computational Molecular Design, Berlin, Building S110, 711, 13342 Berlin
In modern computer-aided drug design processes, synthesizability is still one of the major
bottlenecks. For example, both classical approaches such as fragment based molecular design and modern methods using deep learning techniques often disregard the question of synthesis during the modeling process. This can result in molecules that are perfectly suited for their designed task but cannot be readily synthesized. 
Computer-aided synthesis planning (CASP) methods try to tackle the complexity of the problem. However, they are usually added only after the design process is complete. New machine learning approaches can produce impressive results in finding synthesis routes for novel molecules. Yet, these methods give the synthetic chemist little to no opportunity to contribute their expertise. In addition, there is no guarantee that these models will find a synthetic route or that the proposed routes are suitable for individual needs. Therefore, we decided to shift the perspective and allow the synthetic chemist to propose suitable retrosynthetic routes for hit structures. We use these routes as a pathway to guide structural modifications that introduce desired structural changes to the target molecule, without compromising the synthesizability of the structure. This idea led to our recent publication ‘Synthesis-Aware Generation of Structural Analogs’ . The method is based on a simple approach: exchange a single reactant structure in the tree, verify the integrity of the route
and generate the modified target structure with the structural changes introduced by the new reactant. However, this left the user only with limited control over the modifications introduced to the retrosynthetic route. Since then, we have gone a substantial step further enabling modifications of all components of the given route. In this way, the synthetic chemist is able to further integrate and apply their expertise into the design process.
The algorithmic approaches ended up in our software tool Synthesia making them readily accessible to the medical chemist. Users can either specify exactly where their retrosynthetic route should be modified and are presented with suitable alternatives, or they specify only the substructure of the target molecule to be modified and let the method automatically determine the responsible subtree, proposing modification options. Furthermore, users can exchange or skip reactions, exchange multiple reactant structures simultaneously, and create a target function that defines wanted or unwanted substructures in the target molecule.
Overall, Synthesia allows a new way of thinking during lead structure modification putting the synthetic route into the focus of the compound development. We present several examples of application scenarios to provide an overview of the utility and benefits of our method.
 Gao, W. and Coley, C.W., 2020. The synthesizability of molecules proposed by generative
models. Journal of chemical information and modeling, 60(12), pp.5714-5723.
 Dolfus, U., Briem, H. and Rarey, M., 2022. Synthesis-Aware Generation of Structural
Analogues. Journal of Chemical Information and Modeling, 62(15), pp.3565-3576.