摘要
Synthetic binding proteins(SBPs)with small size,marked solubility and stability,and high affinity are important for protein-based research,treatment,and diagnostics.Over the last several decades,site-directed mutagenesis and directed evolution of privileged protein scaffold make up the great majority of SBPs.The groundbreaking advancement of deep learning(DL)in recent years has revolutionized the problem of protein structure prediction and design.Here,for the first time,the cutting-edge DL framework ProteinMPNN was applied to fulfill the de novo design of 7,245 new synthetic proteins covering 55 different scaffolds based on the original SBPs collected in our SYNBIP database.Comprehensive bioinformatics analysis indicated that,in addition to the excellent performance of sequence recovery,the designed synthetic proteins have a significant improvement in solubility and thermal stability compared to the currently known SBPs.Meanwhile,8 incredibly suitable protein scaffolds for ProteinMPNN have been identified,from which the designed synthetic proteins calculate displayed good performance on binding ability to their corresponding protein targets.Therefore,the DL-based framework shown great potential in targetdirected de novo generation of synthetic protein library with high quality,which could assist experimental biologists to rational protein engineering to discover novel functional protein binders.
基金
supported by the National Natural Science Foundation of China(Grant No.21505009)
the Entrepreneurship and Innovation Support Plan for Chinese Overseas Students of Chongqing(cx2020127).
