| 초록 |
Artificial protein materials with high strength and toughness can overcome the shortcomings of natural proteins and have important applications in biomedical and protective fields. However, the design of protein-based biomaterials so far still relies on manual work and extensive field experience. With protein data grows and machine learning develops, generative models such as generative adversarial networks (GAN) have been widely studied and successfully designed protein sequences with specific functions. In this work, based on the extensive collection of high strength and high toughness protein sequence data sets, it uses generative adversarial network to generate and optimize protein sequences, so that they can be folded into high strength and high toughness materials. The similarity between the generated sequence and the natural sequence is analyzed theoretically. Finally, the generated sequence was expressed experimentally and its mechanical properties were measured after purification. Protein sequence design based on data-driven and generative adversarial networks provides a new method for designing protein materials with high strength and toughness. |
