Supporting Experts in Early Drug Discovery: Algorithmic and Visualization Approaches for Improved Synthetic Accessibility

Abstract

Since early drug development largely takes place on the computer, it must be ensured that the virtually developed structures are not only theoretically effective but also practically producible. To ensure this, the synthesizability of candidates must be included in the design process as early as possible. However, synthesizability is a complex issue that requires a deep understanding of chemistry, practical experience, and often creativity; all characteristics that are difficult to automate. For these reasons, existing methods and approaches address the problem from different directions and attempt to solve different aspects of it. In this work, algorithms were developed to promote the integration of synthesizability into early drug design. The aim is to support synthetic chemists during the design phase of new structures. The modification and optimization of lead structures is a fundamental concept for modern drug development. This includes the systematic exploration and refinement of the chemical structure to improve its pharmacological properties such as efficacy, selectivity and metabolic stability. In the first approach, a method for generating synthetically accessible structural analogues was developed, starting with a lead structure, to enable the efficient synthesis of structural analogues during the design-make-test-analysis cycle. For the developed approach, it was decided explicitly not to create new, artificially constructed synthetic pathways, but to leave the selection of a suitable synthetic route to the synthetic chemist and only generate structural analogues that can in theory be produced via the selected pathway. With synthetic pathways that have already been tested in own laboratories and are well established, structural analogues can be generated, that not only have the desired physicochemical properties but can be tested quickly and efficiently in experiments. Desired reactants can be individually selected to favor one’s own library of building blocks or to allow only commercially available ones. The ability of the method not only to create synthetically accessible, property-specific chemical spaces of structural analogues, but also to be used for the analysis of the synthetic effort of molecule series is demonstrated. The concept of synthesizability is based on chemical reactions. Generic reaction patterns are commonly chosen formats to represent chemical reactions so that a computer can read, analyze, and apply them. The correct creation and human understanding of these patterns is essential for the computer to receive the correct instructions. However, these string-based representations are often difficult to read and interpret, even for trained chemists or developers. To support the use and distribution of these representations of chemical reactions and to provide an easy way to understand them, an algorithm for the visualization of chemical reaction patterns has been developed. The simplicity of interpreting reaction patterns using the chosen visualization strategy is explained using various examples. In addition, two common reaction data sets are provided fully visualized. Synthetic routes are typically constructed and adapted by chemists using their extensive knowledge of common synthesis methods and practical experience. This includes careful selection of starting materials, consideration of well-known chemical reactions, and avoidance of structures that are notoriously difficult to synthesize. In a third approach, a method has been developed to modify synthetic routes and adapt them to individual needs and circumstances. Functionality is provided to exchange all structures as well as all reactions in a synthetic pathway. The physicochemical properties of all structures can be influenced. The individual customization options are explained using examples. In addition, a further use case is presented in which the method is used to analyze which scaffold structures from a given set are synthetically accessible for scaffold hopping with a specific target structure.