Angew.Chem.Int.Ed.http://doi.org/f2qh85(2014)

To find an exotic ingredient for a cooking project you could search in a huge supermarket, but a visit to a specialized food store might be quicker. Researchers are confronted with a similar problem when using combinatorial chemistry for drug discovery: a huge and diverse chemical library might harbour lead compounds for a specific target. But searching a small specialized library is more economical. Consequently, the question of how to design specialized chemical libraries is a matter of intense research. Now, Gisbert Schneider and co-workers from ETH Zürich have described a computational strategy for building such libraries.

The core of their method is an algorithm inspired by the way ants search for food: they fan out randomly but those that find food leave a pheromone trail. Other ants follow the trail and strengthen the connection when more food is found. But the pheromones evaporate if routes remain unused, which makes the algorithm adaptive. The algorithm works even if more than one optimization criterion has to be considered, a feature that Schneider and co-workers use to hunt a drug with activity against two targets involved in neuropsychiatric disorders — the sigma-1 and dopamine D4 receptors.

The starting point is an exhaustive list of amines and aldehydes/ketones. Randomly combining these in reductive amination yields 20 million possible products. But the ant algorithm allows them to home in on the useful ones. Equipped with bioactivities from a large database, the virtual ants assign scores to the product combinations they encounter. The scores reflect a compound's affinity and selectivity for the targets. At the end of the simulation, the final connection strength points the authors towards promising reactant combinations — islands of perfectly specialized chemical libraries in a sea of unusable structures. Analysis of these smaller libraries yields a few highly potent structures for which the bioactivity is experimentally verified. With an overall success rate of 90%, the ant-inspired algorithm holds great promise for computational drug discovery.