Protein design is crucial for the treatment of human diseases, but traditional protein design methods have some limitations. Site-directed mutagenesis is highly dependent on the physiological properties and structure of parental protein. Directed evolution explores only the sequence space regions around natural proteins. Protein de novo design software based on physical methods relies on energy functions to predict the stability and folding behavior, but current energy functions may not be sufficient to accurately simulate the complex folding and interactions of proteins in nature, this has led to instances of design failures.

To solve the problems, a research team led by Dr. Weiwei Xue at Chongqing University published their new research in Frontiers of Computer Science co-published by Higher Education Press and Springer Nature.

The team used the advanced deep learning-based framework ProteinMPNN for the first time to expand the sequence space of synthetic binding proteins (SBPs). Synthetic binding proteins designed by ProteinMPNN show superior performance compared to classical protein engineering methods.

Comprehensive bioinformatics analysis revealed that the novel protein sequences generated by ProteinMPNN exhibit enhanced solubility and stability compared to the original SBPs. Interestingly, it was observed that sequences derived from monomer structures outperform those from complex structures in terms of solubility and stability, whereas sequences designed based on complex structures demonstrate superior calculated binding energy. Through rigorous screening, eight scaffolds with significantly improved solubility and stability were identified, encompassing Neocarzinostatin-based binder, Diabody, CI2-based binder, scFv, Repebody, Fab, Affilin, and Evibody.

DOI: 10.1007/s11704-024-31060-3