Premium
Design of a Genetic Algorithm for the Simulated Evolution of a Library of Asymmetric Transfer Hydrogenation Catalysts
Author(s) -
Vriamont Nicolas,
Govaerts Bernadette,
Grenouillet Pierre,
de Bellefon Claude,
Riant Olivier
Publication year - 2009
Publication title -
chemistry – a european journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.687
H-Index - 242
eISSN - 1521-3765
pISSN - 0947-6539
DOI - 10.1002/chem.200802192
Subject(s) - acetophenone , selection (genetic algorithm) , catalysis , computer science , throughput , transfer (computing) , algorithm , genetic algorithm , chemistry , parallel computing , artificial intelligence , organic chemistry , machine learning , telecommunications , wireless
Breeding new catalysts : A library of 1980 catalysts was designed for asymmetric hydrogen transfer to acetophenone. The library was submitted to evaluation and further simulated evolution experiments, based on a genetic algorithm (see scheme). We demonstrated that it was easily possible to get 5–6 of the ten best catalysts, while investigating only 10% of the library.A library of catalysts was designed for asymmetric‐hydrogen transfer to acetophenone. At first, the whole library was submitted to evaluation using high‐throughput experiments (HTE). The catalysts were listed in ascending order, with respect to their performance, and best catalysts were identified. In the second step, various simulated evolution experiments, based on a genetic algorithm, were applied to this library. A small part of the library, called the mother generation (G0), thus evolved from generation to generation. The goal was to use our collection of HTE data to adjust the parameters of the genetic algorithm, in order to obtain a maximum of the best catalysts within a minimal number of generations. It was namely found that simulated evolution's results depended on the selection of G0 and that a random G0 should be preferred. We also demonstrated that it was possible to get 5 to 6 of the ten best catalysts while investigating only 10 % of the library. Moreover, we developed a double algorithm making this result still achievable if the evolution started with one of the worst G0.