Memetic Genetic Particle Swarm Optimization for Druglike Molecule Discovery

Given the vast and complex chemical search space, developing new techniques for identifying promising ligands that satisfy multiple objectives is highly desirable to reduce the costs and times required for effective drug discovery. Neural networks are frequently employed for this task, but they tend to generate molecules that are invalid both chemically and syntactically. As an alternative, metaheuristics have emerged as promising approaches, delivering notable results with reasonable computational costs. However, they often suffer from information loss during the process, leading to poor quality generations. In this work, we introduce a novel memetic algorithm that hybridizes Particle Swarm Optimization with Simulated Annealing. This approach aims to improve the balance between exploration and exploitation in the de-novo drug discovery process, ensuring that promising molecules are not overlooked during generation steps. We compare our approach against six state-of-the-art algorithms, and the results demonstrate that our algorithm enhances molecule generation quality, showing an increased diversity and improved chemical properties of the resulting ligands.