Structure prediction and in silico screening of protein-peptide interactions

蛋白质-多肽相互作用广泛存在于许多细胞过程中，如信号转导、转录 调节和免疫反应。近年来，基于多肽的疗法引起了人们的极大关注， 越来越多的基于多肽的药物被设计和批准用于各种药物 疾病的威胁。因此，研究蛋白质与多肽的相互作用对于机理研究具有重要意义。 用于许多生物过程和多肽治疗开发。然而，由于困难和 通过X射线结晶学和核磁共振光谱确定这种结构的成本，目前只有一种 蛋白质数据库中蛋白质-肽复合体结构的数量有限。因此，预测蛋白质的能力- 多肽复合体结构将对理解重要的生物过程和 关于设计治疗性干预措施。然而，蛋白质-多肽复合体的结构预测是 具有挑战性，特别是由于多肽的灵活性。在这个项目中，我们将通过以下方式解决这一具有挑战性的问题 生物信息学和物理建模方法的创新整合。具体地说，我们建议实现 四个目标： 目标1：我们将为蛋白质-多肽结构预测开发新的深度学习模型。尽管取得了成功 深度学习在蛋白质结构预测和蛋白质-配体相互作用中的应用尚未见报道 已被应用于蛋白质-多肽结构预测，由于其灵活性和由此产生的很大程度 多肽中的自由。 目标2：我们将开发第一个用于搜索基于多肽的抑制剂的电子筛选方法，并将 构建用于筛选的新型多肽文库。我们的In Silo方法将是对有价值的 噬菌体展示和酵母双杂交系统等快速筛选多肽的实验技术 成本要低得多。 目标3：我们将把我们的计算算法转换成一个模块化的、可扩展的开源软件 可以免费分发给计算建模社区的包。 目标4：作为我们的计算机筛选方法的概念验证应用，我们将筛选新的多肽 与我的实验合作者合作，通过靶向β-内酰胺酶来对抗抗生素耐药性 他的专长是分子生物学、生物化学和微生物学。

Protein-peptide interactions are prevalent in many cellular processes, such as signal transduction, transcription regulation, and immune response. Peptide-based therapeutics have attracted much attention in recent years, and a significantly growing number of peptide-based medicines have been designed and approved for a variety of diseases. Therefore, studying protein-peptide interactions is of great significance for mechanistic investigation of many biological processes and for peptide therapeutic development. However, because of the difficulties and cost for determining such structures by X-ray crystallography and NMR spectroscopy, currently there are only a limited number of protein-peptide complex structures in the Protein Data Bank. Thus, the ability to predict protein- peptide complex structures will have a far-reaching impact on understanding important biological processes and on designing therapeutic interventions. However, structure prediction for protein-peptide complexes is challenging, particularly due to peptide flexibility. In this project, we will address this challenging issue by innovative integration of bioinformatics and physical modeling approaches. Specifically, we propose to achieve four goals: Goal #1: We will develop novel deep-learning models for protein-peptide structure prediction. Despite successful application of deep learning to protein structure prediction and protein-ligand interaction, deep learning has not been applied to protein-peptide structure prediction yet, due to the flexibility and the resulting large degrees of freedom in peptides. Goal #2: We will develop the first in silico screening method for the search of peptide-based inhibitors, and will construct novel peptide libraries for screening. Our in silico method will be an attractive complement to valuable experimental technologies such as phage display and yeast two-hybrid system for rapid peptide screening at much lower cost. Goal #3: We will convert our computational algorithms into a modular, extensible, open-source software package that can be disseminated to the computational modeling community at no cost. Goal #4. As a proof-of-concept application of our in silico screening method, we will screen for novel peptide leads by targeting β-lactamase to combat antibiotic resistance, in collaboration with my experimental collaborator whose expertise is in molecular biology, biochemistry and microbiology.