Análise comparativa de desempenho da geração de conjuntos de moléculas utilizando redes generativas adversárias

Abstract

The generation of new molecules and molecular compounds is a task with numerous applications in various scientific fields, currently playing a crucial role in the development of new drugs, medicines, and chemical compounds that may benefit human life or serve various chemical purposes. To achieve this, it is essential to advance computational methods to optimize this generation process, ensuring that machine learning algorithms effectively learn the patterns of chemical and molecular compounds to generate valid molecules from a physicochemical perspective. Among the molecular generation methods, generative adversarial networks (GANs) stand out as recent algorithms capable of learning complex patterns from large datasets and generating new examples based on the original data. This study aims to evaluate the performance of GANs in the context of molecular generation by comparing the necessary evaluation metrics with other generative AI methods and different datasets related to pharmaceuticals and molecular compounds. This analysis contributes to identifying the strengths and weaknesses of GANs compared to other molecular generation methods, such as VAEs and GNNs, across various contexts and datasets.