CDS&E: Simulation- and Data-driven Peptide Antibody Design Targeting RBD and non-RBD Epitopes of SARS-CoV-2 Spike Protein

药物与蛋白质相互作用，破坏细菌和病毒感染。有效的药物通常是发现的，而不是设计出来的。抗体是免疫系统产生的蛋白质复合体，用于与病毒结合并使其失活。多肽是一串短的氨基酸，被设计成模拟抗体的蛋白质结合活性。蛋白质-蛋白质和蛋白质-多肽相互作用的许多方面还不清楚。这个项目将应用人工智能方法来理解这些互动。SARS-CoV-2刺突蛋白将作为研究的模型系统。最终的治疗性多肽设计模型将在各种软件平台上提供给研究社区。该项目还将支持向K-12学生宣传有关SARS-CoV-2病毒和病毒感染的知识。总体目标是开发一个混合机器学习-模拟(MLSim)平台，使我们能够更好地了解多肽药物和病毒蛋白之间的分子相互作用。模型病毒蛋白系统将是位于受体结合域(RBD)和非RBD的SARS-CoV-2刺突蛋白。将实施现有蛋白质-多肽相互作用数据模型的迁移学习技术。在线学习技术将允许用新获得的数据及时更新预测模型。多尺度模拟组件通过提供高保真输入数据和交叉验证预测来帮助机器学习部分。这些努力应该会导致对病毒蛋白-抗体相互作用的分子水平的洞察。预计该项目将产生两个关键成果。首先，一个支持模拟和数据的平台，集成了高通量、可定制的机器学习管道，用于快速筛选和筛选候选多肽，具有高保真全原子显式溶剂分子动力学模拟和自由能计算。第二项是对病毒蛋白-肽相互作用的基本见解，以及这些相互作用如何影响中和肽的设计。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Drugs interact with proteins to disrupt bacterial and viral infections. Effective drugs are usually discovered rather than designed. Antibodies are protein complexes generated by the immune system to bind to and inactivate viruses. Peptides are short strings of amino acids that are being designed to mimic the protein binding activity of antibodies. Many aspects of protein-protein and protein-peptide interactions are not clearly understood. This project will apply an artificial intelligence approach to understand those interactions. The SARS-CoV-2 spike protein will be the model system for study. The resulting model for therapeutic peptide design will be provided to the research community on a variety of software platforms. The project will also support outreach to K-12 students regarding the SARS-CoV-2 virus and viral infections. The overall objective is to develop a hybrid machine learning-simulation (MLSim) platform that allows us to better understand the molecular interaction between peptide drugs and viral proteins. The model viral protein system will be the SARS-CoV-2 spike proteins at both the receptor-binding domain (RBD) and the non-RBD. Transfer learning techniques for existing data models for protein-peptide interactions will be implemented. Online learning techniques will allow for the timely update of the predictive models with newly available data. The multiscale simulation component aids the machine learning part by supplying high-fidelity input data and cross-validating the predictions These efforts should result in molecular-level insight into viral protein-antibody interactions. There are two key outcomes anticipated from this project. First, a simulation- and data-enabled platform that integrates a high-throughput, customizable machine learning pipeline for fast screening and filtering peptide candidates, with high-fidelity all-atom explicit-solvent molecular dynamics simulation and free energy calculations. The second is fundamental insight into viral protein-peptide interactions and how those influence the design of neutralizing peptides.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.