Artificial Intelligence Driven Drug Affinity Optimization -- 2

Over the last years, there have been several developments in the rapid synthesis and screening of multi-component reaction compounds, which should allow for more efficient drug discovery. Through acoustic dispensing small volumes can be rapidly and precisely screened. However, the large amount of possible combinations of building blocks still poses a challenge. [1] In this report two aspects of this challenge are discussed. The first of these is drug affinity optimization driven by a genetic algorithm (GA) to find high scoring compounds. A new software package with a genetic algorithm, called compound evolver, is presented that uses a fitness function that is primarily based on the potency of compounds, which is represented by the score of various docking programs. The software package delivers a web interface for the interaction with the genetic algorithm that is easy to use for chemists. In addition to the methods employed for building the software package, the genetic algorithm results are also analysed, and parameter tuning is carried out. Various aspects of the genetic algorithm are demonstrated with two protein targets and accompanying multi-component reactions. In the experiments that were carried out it is shown that the genetic algorithm can converge to an optimum on which a large set of GA runs can agree.


Should improve the existing code.