Prediction of influenza antigenic variants using a novel sparse multitask learnin

描述（由申请人提供）：在美国，流感和流感相关并发症每年导致20多万人住院，约3.6万人死亡，疫苗接种是减少流感影响的主要选择。每年必须在全球范围内作出大量努力，以确定抗原变异并决定是否需要新的疫苗株。目前基于实验室的抗原表征过程是劳动密集型和耗时的，并且它已经成为产生有效流感疫苗接种计划的瓶颈。需要一种不需要这种实验室特征的可靠方法来快速鉴定流感抗原变异。本项目拟开发一种新的基于蛋白质序列输入预测流感抗原变异的稀疏多任务学习方法，并进一步将该方法应用于a /H3N2流感病毒抗原漂移途径的定位和流感爆发的抗原漂移模式研究。这种方法是基于流感抗原性将由血凝素（HA）蛋白序列和三级结构的某些特征决定的假设。这一假设得到了很好的证明，具有保守HAs的病毒在血清学反应中产生了交叉反应，在实验室实验和现场实践中也提供了交叉保护。该方法结合了多任务学习和稀疏学习，是一种新颖的方法。因此，该项目不仅将开发抗原变异筛选的重要技术，而且还将开发新的机器学习方法。该项目将促进疫苗株的选择，因为所提出的方法可以潜在地减少甚至消除血清学分析，这是流感监测中最劳动密集型的程序之一。此外，在本研究中确定的抗原性特异性特征和引起流感爆发的漂移模式将增强我们对抗原-抗体相互作用的理解，从而增强我们在流感免疫学和血清学方面的知识。此外，所提出的方法可能适用于表征具有显著抗原变异的其他病原体的抗原特性，例如轮状病毒。具体目标如下：(1)开发一种新的稀疏多任务学习方法，利用血凝素抑制（HI）数据生成抗原距离矩阵；(2)建立了预测硅中抗原变异的定量方法；(3)应用该方法研究季节性流感抗原漂移途径和导致流感暴发的抗原漂移模式。这项研究的本质是提出一种新的预测方法来测量流感病毒之间的抗原差异，这在流感疫苗株选择中是至关重要的。因此，我们将这个项目提交给广泛的挑战领域（06），使能技术和适合特定挑战06- gm -103：开发分子结构、识别和配体相互作用的预测方法。

DESCRIPTION (provided by applicant): Influenza and influenza related complication lead to more than 200,000 hospitalizations and approximately 36,000 deaths in the United States each year, and vaccination is the primary option for reducing influenza effect. A large amount of global efforts has to be made each year to identify antigenic variants and decide whether new vaccine strains are needed. Current laboratory based antigenic characterization processes are labor intensive and time consuming, and it has been the bottleneck for generating an effective influenza vaccination program. A robust method without such a laboratory characterization is demanding for rapid identification of influenza antigenic variants. This project proposes to develop a novel sparse multitask learning method in predicting influenza antigenic variants solely based on the input of protein sequences, and further to apply this method in mapping antigenic drift pathway of A/H3N2 influenza viruses and studying antigenic drift patterns leading to influenza outbreaks. This method is based on the assumption that influenza antigenicity would be determined by certain features in hemagglutinin (HA) protein sequence and tertiary structure. This assumption was well evidenced that the viruses with conserved HAs generated cross-reactions in serological reactions and also provided cross- protection in both laboratory experiments and field practices. The proposed method is novel since it combines multitask learning and sparse learning. Therefore not only this project will develop significant technology for antigenic variant screen, but also new machine learning methods. This project will facilitate vaccine strain selection since the proposed method can potentially reduce and even eliminate serological assay, one of the most labor intensive procedures, in influenza surveillance. In addition, the antigenicity specific features and the drift patterns causing influenza outbreaks to be identified in this study will enhance our understanding about antigen-antibody interaction thus enhance our knowledge in influenza immunology and serology. Furthermore, the proposed method is potentially applicable in characterizing antigenic properties of other pathogens with significant antigenic variations, for example, rotavirus. The specific aims are the following: (1) Development of a novel sparse multitask learning method in generating antigenic distance matrix using hemagglutinin inhibition (HI) data; (2) Development of a quantitative method for predicting antigenic variants in silicon; (3) Application of this method in studying seasonal influenza antigenic drift pathway and antigenic drift patterns leading to influenza outbreaks. This nature of this study is to address a novel predictive method for measuring antigenic divergence between influenza viruses, which is critical in influenza vaccine strain selection. Thus, we are submitting this project to the broad challenge area (06) Enabling Technologies and fit for the Specific Challenge 06-GM-103: development of predictive method for molecular structure, recognition, and ligand interaction. PUBLIC HEALTH RELEVANCE: This study is to develop a novel computational method for influenza antigenic variant prediction, which is very useful in influenza vaccine strain selection. This method will also be applied in studying antigenic drift patterns leading to influenza outbreak and epidemics.