Abstract
AlphaFold2, RoseTTAFold and ProteinGAN/ProtGPT2 are currently three widely used artificial intelligence tools for protein structure prediction and design. In this article, we make a comparative analysis of these three tools based on literature research and exemplary simulation. AlphaFold2 and RoseTTAFold use deep neural networks to predict three-dimensional structure of proteins from amino acid sequence. AlphaFold2 almost reaches the accuracy of experiment in single protein prediction, while RoseTTAFold uses three track architecture to model protein-protein complexes, and its accuracy is slightly lower. ProteinGAN and ProtGPT2 use Generative Adversarial Networks (GANs) and language models to generate new amino acid sequences. ProteinGAN and ProtGPT2 expand protein sequence and sometimes can fold into stable sequence. Their refers to the research results of protein generation tools, which have opened up new directions for the application of artificial intelligence in the field of protein engineering. In this article, we summarize the principle, performance characteristic and typical application of each method. The research results in this paper can provide a reference for selecting the appropriate tool in different situations. AlphaFold2 or RoseTTAFold can be used for accurate structural prediction in bioinformatics and structural biology, and ProtGPT2/ProteinGAN can be used for new design of protein in enzyme engineering and synthetic biology. By mastering the advantages of each tool, researchers can better use computational protein modeling to accelerate scientific innovation.