The application of neural network approaches for the recognition of compounds that are drug-like or suited for crop protection and their discrimination from basic chemicals were discussed. These methods succeed for both application areas with an average correctness of classification of 70-80%. They will be extended in the future to similar criteria like toxicity or bioavailability.

In a second approach, a genetic algorithm was used to optimize simultaneously three criteria - crop protection suitability, diversity, and cost - for a small sub-library out of a huge virtual combinatorial library. It could be

^ max. diverse





10k random

^-.min. diverse

Figure 8. Diversity index vs. percentage of suitable compounds (score >0.3): 10 000 randomly drawn 15 x 15 libraries, the best library after optimization, the maximal and the minimal diverse libraries.

shown that in cases where a systematic approach is impossible (1082 possibilities) the GA is by far superior to a stochastic solution.

0 0

Post a comment