Genome Based Influenza Vaccine Strain Selection using Machine Learning

 描述（由申请人提供）： 甲型流感病毒引起大流行和季节性爆发，导致短时间内数千至数百万人丧生。接种疫苗是预防和尽量减少流感爆发影响的最佳选择。快速选择匹配良好的流感疫苗株是制定有效疫苗接种计划的关键。然而，这是一个不平凡的任务，因为在流感疫苗株选择的三个主要挑战：劳动和时间密集型病毒分离和血清学为基础的抗原性表征，在生产过程中鸡胚蛋中的选择株生长不良，和有偏见的采样流感监测。每年，全世界的许多科学家，包括来自美国的数千名科学家，都在共同努力选择最佳的疫苗株。然而，在过去的几十年里，仍然经常选择不正确的疫苗株。 基因组测序的最新进展使我们能够快速和经济地从分离株和临床样本中测序流感基因组。流感基因组测序已成为流感监测的常规和重要组成部分。该项目的目标是开发一种基于序列的流感抗原变异识别策略，并利用基因组数据优化疫苗株选择。为了实现这些目标，我们将开发基于机器学习的计算方法，通过直接使用流感病毒的基因组序列来估计它们之间的抗原距离。然后，我们将确定流感基因组中影响流感抗原漂移事件的关键残基和突变。这些信息将使我们能够选择最有希望的病毒株作为疫苗生产的候选株。由于经济的病毒生产需要选择的病毒株在鸡胚蛋中容易生长，我们还提出了一种基于机器学习的方法，可以根据其序列信息预测病毒株的生长能力。这种基于基因组的流感疫苗株筛选系统将被开发用于检测甲型流感病毒的抗原变异。 该项目将帮助我们提供利用基因组签名确定流感抗原性和鸡胚蛋中的生长能力的基础技术，这是高效和有效流感疫苗株开发的两个关键问题。所得的基于基因组的疫苗株选择策略将显著减少血清学表征所需的人力，减少选择将在蛋中良好生长的有效株所需的时间，并增加正确的流感疫苗候选物选择的可能性。因此，该项目将导致流感预防和控制方面的重大技术进步。

 DESCRIPTION (provided by applicant): Influenza A virus causes both pandemic and seasonal outbreaks, leading to loss of from thousands to millions of human lives within a short time period. Vaccination is the best option to prevent and minimize the effects of influenza outbreaks. Rapid selection of a well-matched influenza vaccine strain is the key to developing an effective vaccination program. However, this is a non-trivial task due to three major challenges in influenza vaccine strain selection: labor an time intensive virus isolation and serology-based antigenic characterization, poor growth of selected strains in chicken embryonic eggs during production, and biased sampling in influenza surveillance. Each year, many scientists worldwide, including thousands from the United States, are working altogether to select an optimal vaccine strain. However, incorrect vaccine strains have still been frequently chosen in the past decades. Recent advances in genomic sequencing allow us to rapidly and economically sequence influenza genomes from the isolates and from the clinical samples. Sequencing influenza genomes has become a routine and important component in influenza surveillance. The objectives of this project are to develop a sequence-based strategy for influenza antigenic variant identification and to optimize vaccine strain selection using genomic data. To achieve these aims, we will develop machine learning based computational methods to estimate antigenic distances among influenza viruses by directly using their genome sequences. We will then identify the key residues and mutations in influenza genomes affecting influenza antigenic drift events. Such information will allow us to select most promising virus strains as candidates for vaccine production. Since economical virus production requires the selected virus strains to grow easily in chicken embryonic eggs, we also propose the development of a machine learning based method that can predict the growth ability of a virus strain based on its sequence information. This integrated genome based influenza vaccine strain selection system will be developed for detecting antigenic variants for influenza A viruses. This project will help us provide fundamental technology that employs genomic signatures determining influenza antigenicity and growth ability in chicken embryonic eggs, which are the two key issues for efficient and effective influenza vaccine strain development. The resulting genome based vaccine strain selection strategy will significantly reduce the human labor needed for serological characterization, decrease the time required to select an effective strain that will grow well in eggs, and increase the likelihood of correct influenza vaccine candidate selection. Thus, this project will lead to significant technological advances in influenza prevention and control.