Here’s an interesting application of generated database chemistry to identify relatively simple unexplored scaffolds in the CNS area. The authors from Bern have done a lot of work in the past on generating large virtual libraries of compounds starting from simple rules regarding ring numbers and sizes and physicochemical properties. The question always is whether these generated databases contain novel chemical territory. The current study seems to suggest that there could be plenty of “white space” to mine in these databases, unexplored space that could yield novel molecules.

In this particular case, the authors used their GDB-4c database that has more than 900,000 carbocylic systems. They applied some very simple rules - rings equal to or less than 2, less than or equal to two exocylic or endocylic nitrogens. That yielded 1169 compounds, and surprisingly 60% of these are not found in PubChem (my feeling is that the paucity of your standard flat-ring, amino-pyrimidine type structures accounts significantly for this novelty).

The authors then used the "Polypharmacology Browser 2 (PPB2)", a web-based target prediction tool, that predicted monoamine transporters (MATs) and σ-1-receptor as likely targets. Assays validated the activity of the in-silico compound, an N-benzylated bicyclic azepane, at NET and DAT with IC₅₀s of <100 nM and at the σ-1-receptor with IC₅₀ ≈ 110 nM. The lead showed good ADME/PK in mice (good metabolic stability, brain exposure compatible with CNS action).

As a chemist I have to say I find the scaffold pretty - targeted functionality packaged in a small space. It's particularly interesting to see two basic nitrogens in a brain-penetrant molecule. These could also provide a handle for tuning selectivity through appropriate stereoelectronic and pKa tuning. But the study is generally interesting because it’s a good data point in favor of these virtual ultra-large generated databases. There might be more hiding in there than we skeptics might think.