Chemical Space Travel

Modern medicine critically depends on the discovery of new drugs. In this context a detailed knowledge of the ensemble of all possible organic molecules would be extremely useful to identify new structural types. This so-called chemical space is estimated at 10-10 structures in the typical drug range of MW 500 Da, which is far too large for an exhaustive listing. On the other hand known drugs define regions of chemical space that might be particularly favourable for discovering useful compounds. Herein we report a “spaceship” program which travels from a starting molecule A to a target molecule B through a continuum of structural mutations, and thereby charts unexplored chemical space. The compounds encountered along the way provide valuable starting points for virtual screening, as exemplified for ligands of the AMPA receptor. The principle of chemical space travel presented here is different from previously reported molecular structure evolution programs that combine fragments of different molecules, which did not follow the structural continuum and were not shown to reach a set target molecule. Chemical space is often visualized as a property space whose dimensions represent numerical properties of molecules, such as physicochemical descriptor values, pharmacophore descriptors, or similarity measures to reference compounds. One can define nearest neighbours in such property space as compounds with the most similar numerical property values, a concept which has been previously used for data mining by classification of existing libraries and databases. However, because one cannot derive a structure from its descriptor values, it is not possible to move between nearest neighbours in property space unless the structure of the nearest neighbours and hence all compounds under consideration are known in advance. To enable movement in an unexplored chemical space and the discovery of new structures, we describe chemical space as a structural continuum. Rather than referring to proximity in property space, we define nearest neighbours as molecules related through a single structural mutation, for example an atom-type exchange, the addition or retrieval of a bond or atom, or a skeletal rearrangement (Table 1). This description or-