Optimization of chemical synthesis with heuristic algorithms

Optimizing reaction conditions to improve the yield is fundamental for chemical synthesis and industrial processes. Experiments can only be performed under a small portion of reaction conditions for a system, so a strategy of experimental design is required. Bayesian optimization, a global optimization algorithm, was found to outperform human decision-making in reaction optimization. Similarly, heuristic algorithms also have the potential to solve optimization problems. In this work, we optimize these reaction conditions for Buchwald-Hartwig and Suzuki systems by predicting reaction yields with three heuristic algorithms and three encoding methods. Our results demonstrate that particle swarm optimization with numerical encoding is better than the genetic algorithm or simulated annealing. Moreover, its performance is comparable to Bayesian optimization without the computational costs of descriptors. Particle swarm optimization is simple and easy to perform, and it can be implemented into laboratory practice to promote chemical synthesis.